C# 5.0 in a Nutshell 5th

C# 5.0 IN A NUTSHELL Fifth Edition Joseph Albahari and Ben Albahari Beijing • Cambridge • Farnham • Köln • Sebastopol • Tokyo C# 5.0 in a Nutshell, Fifth Edition by Joseph Albahari and Ben Albahari Copyright © 2012 Joseph Albahari and Ben Albahari. All rights reserved. Printed in the United States of America. Published by O’Reilly Media, Inc., 1005 Gravenstein Highway North, Sebastopol, CA 95472. O’Reilly books may be purchased for educational, business, or sales promotional use. Online editions are also available for most titles (http://my.safaribooksonline.com). For more infor- mation, contact our corporate/institutional sales department: 800-998-9938 or corporate@oreilly.com. Editor: Rachel Roumeliotis Production Editor: Melanie Yarbrough Copyeditor: Nancy Reinhardt Proofreader: Jennifer Knight Indexer: Jay Marchand Cover Designer: Karen Montgomery Interior Designer: David Futato Illustrator: Robert Romano June 2012: Fifth Edition. Revision History for the Fifth Edition: 2012-06-08 First release See http://oreilly.com/catalog/errata.csp?isbn=9781449320102 for release details. Nutshell Handbook, the Nutshell Handbook logo, and the O’Reilly logo are registered trade- marks of O’Reilly Media, Inc. C# 5.0 in a Nutshell, the cover image of a numidian crane, and related trade dress are trademarks of O’Reilly Media, Inc. Many of the designations used by manufacturers and sellers to distinguish their products are claimed as trademarks. Where those designations appear in this book, and O’Reilly Media, Inc., was aware of a trademark claim, the designations have been printed in caps or initial caps. While every precaution has been taken in the preparation of this book, the publisher and authors assume no responsibility for errors or omissions, or for damages resulting from the use of the information contained herein. ISBN: 978-1-449-32010-2 [M] 1340210346 Table of Contents Preface ............................................................. xi 1. Introducing C# and the .NET Framework ............................ 1 Object Orientation 1 Type Safety 2 Memory Management 2 Platform Support 3 C#’s Relationship with the CLR 3 The CLR and .NET Framework 3 C# and Windows Runtime 5 What’s New in C# 5.0 6 What’s New in C# 4.0 6 What’s New in C# 3.0 7 2. C# Language Basics .............................................. 9 A First C# Program 9 Syntax 12 Type Basics 15 Numeric Types 23 Boolean Type and Operators 30 Strings and Characters 32 Arrays 34 Variables and Parameters 38 Expressions and Operators 47 Statements 51 Namespaces 59 iii 3. Creating Types in C# ............................................ 67 Classes 67 Inheritance 80 The object Type 89 Structs 93 Access Modifiers 94 Interfaces 96 Enums 102 Nested Types 105 Generics 106 4. Advanced C# ................................................. 119 Delegates 119 Events 128 Lambda Expressions 135 Anonymous Methods 139 try Statements and Exceptions 140 Enumeration and Iterators 148 Nullable Types 153 Operator Overloading 158 Extension Methods 162 Anonymous Types 164 Dynamic Binding 165 Attributes 173 Caller Info Attributes (C# 5) 175 Unsafe Code and Pointers 177 Preprocessor Directives 180 XML Documentation 182 5. Framework Overview .......................................... 187 The CLR and Core Framework 189 Applied Technologies 194 6. Framework Fundamentals ...................................... 201 String and Text Handling 201 Dates and Times 214 Dates and Time Zones 221 Formatting and Parsing 227 Standard Format Strings and Parsing Flags 233 Other Conversion Mechanisms 240 Globalization 244 Working with Numbers 245 Enums 249 Tuples 252 The Guid Struct 253 iv | Table of Contents Equality Comparison 254 Order Comparison 264 Utility Classes 267 7. Collections ................................................... 271 Enumeration 271 The ICollection and IList Interfaces 279 The Array Class 282 Lists, Queues, Stacks, and Sets 291 Dictionaries 299 Customizable Collections and Proxies 306 Plugging in Equality and Order 312 8. LINQ Queries ................................................. 319 Getting Started 319 Fluent Syntax 321 Query Expressions 328 Deferred Execution 332 Subqueries 338 Composition Strategies 342 Projection Strategies 345 Interpreted Queries 347 LINQ to SQL and Entity Framework 354 Building Query Expressions 368 9. LINQ Operators ............................................... 375 Overview 377 Filtering 379 Projecting 383 Joining 395 Ordering 403 Grouping 406 Set Operators 409 Conversion Methods 410 Element Operators 413 Aggregation Methods 415 Quantifiers 419 Generation Methods 420 10. LINQ to XML .................................................. 423 Architectural Overview 423 X-DOM Overview 424 Instantiating an X-DOM 427 Navigating and Querying 430 Updating an X-DOM 435 Table of Contents | v Working with Values 438 Documents and Declarations 441 Names and Namespaces 444 Annotations 450 Projecting into an X-DOM 450 11. Other XML Technologies ........................................ 457 XmlReader 458 XmlWriter 467 Patterns for Using XmlReader/XmlWriter 469 XmlDocument 473 XPath 477 XSD and Schema Validation 481 XSLT 484 12. Disposal and Garbage Collection ................................. 485 IDisposable, Dispose, and Close 485 Automatic Garbage Collection 490 Finalizers 493 How the Garbage Collector Works 497 Managed Memory Leaks 501 Weak References 505 13. Diagnostics and Code Contracts .................................. 509 Conditional Compilation 509 Debug and Trace Classes 512 Code Contracts Overview 516 Preconditions 520 Postconditions 524 Assertions and Object Invariants 527 Contracts on Interfaces and Abstract Methods 528 Dealing with Contract Failure 529 Selectively Enforcing Contracts 531 Static Contract Checking 533 Debugger Integration 535 Processes and Process Threads 536 StackTrace and StackFrame 537 Windows Event Logs 538 Performance Counters 541 The Stopwatch Class 545 14. Concurrency & Asynchrony ...................................... 547 Introduction 547 Threading 548 Tasks 565 vi | Table of Contents Principles of Asynchrony 573 Asynchronous Functions in C# 5.0 578 Asynchronous Patterns 594 Obsolete Patterns 601 15. Streams and I/O .............................................. 605 Stream Architecture 605 Using Streams 607 Stream Adapters 621 Compression Streams 629 Working with Zip Files 631 File and Directory Operations 632 File I/O in Windows Runtime 642 Memory-Mapped Files 644 Isolated Storage 647 16. Networking .................................................. 653 Network Architecture 653 Addresses and Ports 655 URIs 656 Client-Side Classes 658 Working with HTTP 671 Writing an HTTP Server 677 Using FTP 680 Using DNS 682 Sending Mail with SmtpClient 683 Using TCP 683 Receiving POP3 Mail with TCP 687 TCP in Windows Runtime 689 17. Serialization ................................................. 691 Serialization Concepts 691 The Data Contract Serializer 695 Data Contracts and Collections 705 Extending Data Contracts 707 The Binary Serializer 710 Binary Serialization Attributes 712 Binary Serialization with ISerializable 715 XML Serialization 719 18. Assemblies ................................................... 729 What’s in an Assembly 729 Strong Names and Assembly Signing 734 Assembly Names 737 Authenticode Signing 739 Table of Contents | vii The Global Assembly Cache 743 Resources and Satellite Assemblies 745 Resolving and Loading Assemblies 754 Deploying Assemblies Outside the Base Folder 759 Packing a Single-File Executable 760 Working with Unreferenced Assemblies 762 19. Reflection and Metadata ....................................... 765 Reflecting and Activating Types 766 Reflecting and Invoking Members 773 Reflecting Assemblies 785 Working with Attributes 786 Dynamic Code Generation 792 Emitting Assemblies and Types 799 Emitting Type Members 803 Emitting Generic Methods and Types 808 Awkward Emission Targets 810 Parsing IL 814 20. Dynamic Programming ........................................ 821 The Dynamic Language Runtime 821 Numeric Type Unification 823 Dynamic Member Overload Resolution 824 Implementing Dynamic Objects 830 Interoperating with Dynamic Languages 833 21. Security ..................................................... 837 Permissions 837 Code Access Security (CAS) 842 Allowing Partially Trusted Callers 845 The Transparency Model 847 Sandboxing Another Assembly 855 Operating System Security 858 Identity and Role Security 861 Cryptography Overview 862 Windows Data Protection 863 Hashing 864 Symmetric Encryption 865 Public Key Encryption and Signing 870 22. Advanced Threading ........................................... 875 Synchronization Overview 876 Exclusive Locking 876 Locking and Thread Safety 884 Non-Exclusive Locking 890 viii | Table of Contents Signaling with Event Wait Handles 895 The Barrier Class 903 Lazy Initialization 904 Thread-Local Storage 907 Interrupt and Abort 909 Suspend and Resume 910 Timers 911 23. Parallel Programming ......................................... 915 Why PFX? 915 PLINQ 918 The Parallel Class 931 Task Parallelism 938 Working with AggregateException 947 Concurrent Collections 949 BlockingCollection 952 24. Application Domains .......................................... 957 Application Domain Architecture 957 Creating and Destroying Application Domains 958 Using Multiple Application Domains 960 Using DoCallBack 962 Monitoring Application Domains 963 Domains and Threads 963 Sharing Data Between Domains 965 25. Native and COM Interoperability ................................. 971 Calling into Native DLLs 971 Type Marshaling 972 Callbacks from Unmanaged Code 975 Simulating a C Union 975 Shared Memory 976 Mapping a Struct to Unmanaged Memory 979 COM Interoperability 983 Calling a COM Component from C# 985 Embedding Interop Types 988 Primary Interop Assemblies 989 Exposing C# Objects to COM 990 26. Regular Expressions ........................................... 991 Regular Expression Basics 992 Quantifiers 996 Zero-Width Assertions 997 Groups 1000 Replacing and Splitting Text 1001 Table of Contents | ix Cookbook Regular Expressions 1003 Regular Expressions Language Reference 1006 Index ............................................................ 1011 x | Table of Contents Preface C# 5.0 represents the fourth major update to Microsoft’s flagship programming language, positioning C# as a language with unusual flexibility and breadth. At one end, it offers high-level abstractions such as query expressions and asynchronous continuations, while at the other end, it provides low-level power through constructs such as custom value types and the optional use of pointers. The price of this growth is that there’s more than ever to learn. Although tools such as Microsoft’s IntelliSense—and online references—are excellent in helping you on the job, they presume an existing map of conceptual knowledge. This book provides exactly that map of knowledge in a concise and unified style—free of clutter and long introductions. Like the past two editions, C# 5.0 in a Nutshell is organized entirely around concepts and use cases, making it friendly both to sequential reading and to random browsing. It also plumbs significant depths while assuming only basic background knowledge —making it accessible to intermediate as well as advanced readers. This book covers C#, the CLR, and the core Framework assemblies. We’ve chosen this focus to allow space for difficult topics such as concurrency, security, and application domains—without compromising depth or readability. Features new to C# 5.0 and the associated Framework are flagged so that you can also use this book as a C# 4.0 reference. Intended Audience This book targets intermediate to advanced audiences. No prior knowledge of C# is required, but some general programming experience is necessary. For the begin- ner, this book complements, rather than replaces, a tutorial-style introduction to programming. If you’re already familiar with C# 4.0, you’ll find a reorganized section on concur- rency, including thorough coverage of C# 5.0’s asynchronous functions and its xi associated types. We also describe the principles of asynchronous programming and how it helps with efficiency and thread-safety. This book is an ideal companion to any of the vast array of books that focus on an applied technology such as WPF, ASP.NET, or WCF. The areas of the language and .NET Framework that such books omit, C# 5.0 in a Nutshell covers in detail— and vice versa. If you’re looking for a book that skims every .NET Framework technology, this is not for you. This book is also unsuitable if you want to learn about APIs specific to tablet or Windows Phone development. How This Book Is Organized The first three chapters after the introduction concentrate purely on C#, starting with the basics of syntax, types, and variables, and finishing with advanced topics such as unsafe code and preprocessor directives. If you’re new to the language, you should read these chapters sequentially. The remaining chapters cover the core .NET Framework, including such topics as LINQ, XML, collections, code contracts, concurrency, I/O and networking, mem- ory management, reflection, dynamic programming, attributes, security, application domains, and native interoperability. You can read most of these chapters randomly, except for Chapters 6 and 7, which lay a foundation for subsequent topics. The three chapters on LINQ are also best read in sequence, and some chapters assume some knowledge of concurrency, which we cover in Chapter 14. What You Need to Use This Book The examples in this book require a C# 5.0 compiler and Microsoft .NET Frame- work 4.5. You will also find Microsoft’s .NET documentation useful to look up individual types and members (which is available online). While it’s possible to write source code in Notepad and invoke the compiler from the command line, you’ll be much more productive with a code scratchpad for instantly testing code snippets, plus an Integrated Development Environment (IDE) for producing executables and libraries. For a code scratchpad, download LINQPad 4.40 or later from www.linqpad.net (free). LINQPad fully supports C# 5.0 and is maintained by one of the authors. For an IDE, download Microsoft Visual Studio 2012: any edition is suitable for what’s taught in this book, except the free express edition. xii | Preface All code listings for Chapter 2 through Chapter 10, plus the chapters on concurrency, parallel programming, and dynamic programming are available as interactive (editable) LINQPad samples. You can download the whole lot in a single click: go to LINQPad’s Samples tab at the bottom left, click “Download more samples,” and choose “C# 5.0 in a Nutshell.” Conventions Used in This Book The book uses basic UML notation to illustrate relationships between types, as shown in Figure P-1. A slanted rectangle means an abstract class; a circle means an interface. A line with a hollow triangle denotes inheritance, with the triangle pointing to the base type. A line with an arrow denotes a one-way association; a line without an arrow denotes a two-way association. The following typographical conventions are used in this book: Italic Indicates new terms, URIs, filenames, and directories Constant width Indicates C# code, keywords and identifiers, and program output Constant width bold Shows a highlighted section of code Figure P-1. Sample diagram Preface | xiii Constant width italic Shows text that should be replaced with user-supplied values This icon signifies a tip, suggestion, or general note. This icon indicates a warning or caution. Using Code Examples This book is here to help you get your job done. In general, you may use the code in this book in your programs and documentation. You do not need to contact us for permission unless you’re reproducing a significant portion of the code. For example, writing a program that uses several chunks of code from this book does not require permission. Selling or distributing a CD-ROM of examples from O’Reilly books does require permission. Answering a question by citing this book and quoting example code does not require permission. Incorporating a significant amount of example code from this book into your product’s documentation does require per- mission. We appreciate, but do not require, attribution. For example: “C# 5.0 in a Nut- shell by Joseph Albahari and Ben Albahari. Copyright 2012 Joseph Albahari and Ben Albahari, 978-1-449-32010-2.” If you feel your use of code examples falls outside fair use or the permission given here, feel free to contact us at permissions@oreilly.com. Safari® Books Online Safari Books Online (www.safaribooksonline.com) is an on-demand digital library that delivers expert content in both book and video form from the world’s leading authors in technology and business. Technology professionals, software developers, web designers, and business and creative professionals use Safari Books Online as their primary resource for research, problem solving, learning, and certification training. Safari Books Online offers a range of product mixes and pricing programs for or- ganizations, government agencies, and individuals. Subscribers have access to thousands of books, training videos, and prepublication manuscripts in one fully searchable database from publishers like O’Reilly Media, Prentice Hall Professional, Addison-Wesley Professional, Microsoft Press, Sams, Que, Peachpit Press, Focal Press, Cisco Press, John Wiley & Sons, Syngress, Morgan Kaufmann, IBM Red- books, Packt, Adobe Press, FT Press, Apress, Manning, New Riders, McGraw-Hill, xiv | Preface Jones & Bartlett, Course Technology, and dozens more. For more information about Safari Books Online, please visit us online. How to Contact Us Please address comments and questions concerning this book to the publisher: O’Reilly Media, Inc. 1005 Gravenstein Highway North Sebastopol, CA 95472 800-998-9938 (in the United States or Canada) 707-829-0515 (international or local) 707-829-0104 (fax) We have a web page for this book, where we list errata, examples, and any additional information. You can access this page at: http://oreil.ly/csharp5_IAN To comment or ask technical questions about this book, send email to: bookquestions@oreilly.com For more information about our books, courses, conferences, and news, see our website at http://www.oreilly.com. Find us on Facebook: http://facebook.com/oreilly Follow us on Twitter: http://twitter.com/oreillymedia Watch us on YouTube: http://www.youtube.com/oreillymedia Acknowledgments Joseph Albahari First, I want to thank my brother, Ben Albahari, for persuading me to take on C# 3.0 in a Nutshell, whose success has spawned two subsequent editions. Ben shares my willingness to question conventional wisdom, and the tenacity to pull things apart until it becomes clear how they really work. It’s been an honor to have superb technical reviewers on the team. This edition owes much to two legendary individuals at Microsoft: Eric Lippert (C# compiler team) and Stephen Toub (Parallel Programming team). I can’t thank you enough for your extensive and useful feedback—and for answering all my questions. I’m also immensely grateful to C# MVP Nicholas Paldino, whose keen eye and ability to pick up things that others miss, shaped this book and two previous editions. This book was built on C# 4.0 in a Nutshell, whose technical reviewers I owe a similar honor. Chris Burrows (C# compiler team) significantly polished the chapters on concurrency, dynamic programming, and the C# language. From the CLR team, I received invaluable input on security and memory management from Shawn Preface | xv Farkas, Brian Grunkemeyer, Maoni Stephens, and David DeWinter. And on Code Contracts, the feedback from Brian Grunkemeyer, Mike Barnett, and Melitta An- dersen raised the chapter to the next quality bar. I have the highest praise for Jon Skeet (author of C# in Depth and Stack Overflow extraordinaire), whose perceptive suggestions shaped the previous edition, C# MVPs Mitch Wheat and Brian Peek, and reviewers of the 3.0 edition, including Krzysztof Cwalina, Matt Warren, Joel Pobar, Glyn Griffiths, Ion Vasilian, Brad Abrams, Sam Gentile, and Adam Nathan. Finally, I want to thank the O’Reilly team, including my editor, Rachel Roumeliotis (a joy to work with), my excellent copy editor, Nancy Reinhardt, and members of my family, Miri and Sonia. Ben Albahari Because my brother wrote his acknowledgments first, you can infer most of what I want to say. :) We’ve actually both been programming since we were kids (we shared an Apple IIe; he was writing his own operating system while I was writing Hangman), so it’s cool that we’re now writing books together. I hope the enriching experience we had writing the book will translate into an enriching experience for you reading the book. I’d also like to thank my former colleagues at Microsoft. Many smart people work there, not just in terms of intellect but also in a broader emotional sense, and I miss working with them. In particular, I learned a lot from Brian Beckman, to whom I am indebted. xvi | Preface 1Introducing C# and the .NET Framework C# is a general-purpose, type-safe, object-oriented programming language. The goal of the language is programmer productivity. To this end, the language balances simplicity, expressiveness, and performance. The chief architect of the language since its first version is Anders Hejlsberg (creator of Turbo Pascal and architect of Delphi). The C# language is platform-neutral, but it was written to work well with the Microsoft .NET Framework. Object Orientation C# is a rich implementation of the object-orientation paradigm, which includes encapsulation, inheritance, and polymorphism. Encapsulation means creating a boundary around an object, to separate its external (public) behavior from its internal (private) implementation details. The distinctive features of C# from an object-oriented perspective are: Unified type system The fundamental building block in C# is an encapsulated unit of data and functions called a type. C# has a unified type system, where all types ultimately share a common base type. This means that all types, whether they represent business objects or are primitive types such as numbers, share the same basic set of functionality. For example, an instance of any type can be converted to a string by calling its ToString method. Classes and interfaces In a traditional object-oriented paradigm, the only kind of type is a class. In C#, there are several other kinds of types, one of which is an interface. An interface is like a class, except that it only describes members. The implementation for those members comes from types that implement the interface. Interfaces are particularly useful in scenarios where multiple inheritance is required (unlike 1 languages such as C++ and Eiffel, C# does not support multiple inheritance of classes). Properties, methods, and events In the pure object-oriented paradigm, all functions are methods (this is the case in Smalltalk). In C#, methods are only one kind of function member, which also includes properties and events (there are others, too). Properties are function members that encapsulate a piece of an object’s state, such as a button’s color or a label’s text. Events are function members that simplify acting on object state changes. Type Safety C# is primarily a type-safe language, meaning that instances of types can interact only through protocols they define, thereby ensuring each type’s internal consis- tency. For instance, C# prevents you from interacting with a string type as though it were an integer type. More specifically, C# supports static typing, meaning that the language enforces type safety at compile time. This is in addition to type safety being enforced at runtime. Static typing eliminates a large class of errors before a program is even run. It shifts the burden away from runtime unit tests onto the compiler to verify that all the types in a program fit together correctly. This makes large programs much easier to man- age, more predictable, and more robust. Furthermore, static typing allows tools such as IntelliSense in Visual Studio to help you write a program, since it knows for a given variable what type it is, and hence what methods you can call on that variable. C# also allows parts of your code to be dynamically typed via the dynamic keyword (introduced in C# 4). However, C# re- mains a predominantly statically typed language. C# is also called a strongly typed language because its type rules (whether enforced statically or at runtime) are very strict. For instance, you cannot call a function that’s designed to accept an integer with a floating-point number, unless you first explic- itly convert the floating-point number to an integer. This helps prevent mistakes. Strong typing also plays a role in enabling C# code to run in a sandbox—an envi- ronment where every aspect of security is controlled by the host. In a sandbox, it is important that you cannot arbitrarily corrupt the state of an object by bypassing its type rules. Memory Management C# relies on the runtime to perform automatic memory management. The Common Language Runtime has a garbage collector that executes as part of your program, reclaiming memory for objects that are no longer referenced. This frees programmers 2 | Chapter 1: Introducing C# and the .NET Framework from explicitly deallocating the memory for an object, eliminating the problem of incorrect pointers encountered in languages such as C++. C# does not eliminate pointers: it merely makes them unnecessary for most pro- gramming tasks. For performance-critical hotspots and interoperability, pointers may be used, but they are permitted only in blocks that are explicitly marked unsafe. Platform Support C# is typically used for writing code that runs on Windows platforms. Although Microsoft standardized the C# language through ECMA, the total amount of re- sources (both inside and outside of Microsoft) dedicated to supporting C# on non- Windows platforms is relatively small. This means that languages such as Java are sensible choices when multiplatform support is of primary concern. Having said this, C# can be used to write cross-platform code in the following scenarios: • C# code may run on the server and dish up HTML that can run on any platform. This is precisely the case for ASP.NET. • C# code may run on a runtime other than the Microsoft Common Language Runtime. The most notable example is the Mono project, which has its own C# compiler and runtime, running on Linux, Solaris, Mac OS X, and Windows. • C# code may run on a host that supports Microsoft Silverlight (supported for Windows and Mac OS X). This technology is analogous to Adobe’s Flash Player. C#’s Relationship with the CLR C# depends on a runtime equipped with a host of features such as automatic mem- ory management and exception handling. The design of C# closely maps to the design of Microsoft’s Common Language Runtime (CLR), which provides these run- time features (although C# is technically independent of the CLR). Furthermore, the C# type system maps closely to the CLR type system (e.g., both share the same definitions for predefined types). The CLR and .NET Framework The .NET Framework consists of the CLR plus a vast set of libraries. The libraries consist of core libraries (which this book is concerned with) and applied libraries, which depend on the core libraries. Figure 1-1 is a visual overview of those libraries (and also serves as a navigational aid to the book). The CLR is the runtime for executing managed code. C# is one of several managed languages that get compiled into managed code. Managed code is packaged into an assembly, in the form of either an executable file (an .exe) or a library (a .dll), along with type information, or metadata. Managed code is represented in Intermediate Language or IL. When the CLR loads an assembly, it converts the IL into the native code of the machine, such as x86. This The CLR and .NET Framework | 3 Introduction conversion is done by the CLR’s JIT (Just-In-Time) compiler. An assembly retains almost all of the original source language constructs, which makes it easy to inspect and even generate code dynamically. Red Gate’s .NET Reflector application is an invaluable tool for examining the contents of an assembly. You can also use it as a decompiler. The CLR performs as a host for numerous runtime services. Examples of these serv- ices include memory management, the loading of libraries, and security services. The CLR is language-neutral, allowing developers to build applications in multiple languages (e.g., C#, Visual Basic .NET, Managed C++, Delphi.NET, Chrome .NET, and J#). The .NET Framework contains libraries for writing just about any Windows- or web-based application. Chapter 5 gives an overview of the .NET Framework libraries. Figure 1-1. Topics covered in this book and the chapters in which they are found. Topics not covered are shown outside the large circle. 4 | Chapter 1: Introducing C# and the .NET Framework C# and Windows Runtime C# 5.0 also interoperates with Windows Runtime (WinRT) libraries. WinRT is an execution interface and runtime environment for accessing libraries in a language- neutral and object-oriented fashion. It ships with Windows 8 and is (in part) an enhanced version of Microsoft’s Component Object Model or COM (see Chapter 25). Windows 8 ships with a set of unmanaged WinRT libraries which serve as a frame- work for touch-enabled Metro-style applications delivered through Microsoft’s application store. (The term WinRT also refers to these libraries.) Being WinRT, the libraries can easily be consumed not only from C# and VB, but C++ and JavaScript. Some WinRT libraries can also be consumed in normal non- tablet applications. However, taking a dependency on WinRT gives your application a minimum OS requirement of Windows 8. (And into the future, taking a dependency on the next version of WinRT would give your program a minimum OS require- ment of Windows 9.) The WinRT libraries support the new Metro user interface (for writing immersive touch-first applications), mobile device-specific features (sensors, text messaging and so on), and a range of core functionality that overlaps with parts of the .NET Framework. Because of this overlap, Visual Studio includes a reference profile (a set of .NET reference assemblies) for Metro projects that hides the portions of the .NET Framework that overlap with WinRT. This profile also hides large portions of the .NET Framework considered unnecessary for tablet apps (such as accessing a database). Microsoft’s application store, which controls the distribution of software to consumer devices, rejects any program that attempts to access a hidden type. A reference assembly exists purely to compile against and may have a restricted set of types and members. This allows devel- opers to install the full .NET Framework on their machines while coding certain projects as though they had only a subset. The actual functionality comes at runtime from assemblies in the Global Assembly Cache (see Chapter 18) which may superset the reference assemblies. Hiding most of the .NET Framework eases the learning curve for developers new to the Microsoft platform, although there are two more important goals: • It sandboxes applications (restricts functionality to reduce the impact of mal- ware). For instance, arbitrary file access is forbidden, and there the ability to start or communicate with other programs on the computer is extremely restricted. • It allows low-powered Metro-only tablets to ship with a reduced .NET Frame- work (Metro profile), lowering the OS footprint. C# and Windows Runtime | 5 Introduction What distinguishes WinRT from ordinary COM is that WinRT projects its libraries into a multitude of languages, namely C#, VB, C++ and JavaScript, so that each language sees WinRT types (almost) as though they were written especially for it. For example, WinRT will adapt capitalization rules to suit the standards of the target language, and will even remap some functions and interfaces. WinRT assemblies also ship with rich metadata in .winmd files which have the same format as .NET assembly files, allowing transparent consumption without special ritual. In fact, you might even be unaware that you’re using WinRT rather than .NET types, aside of namespace differences. (Another clue is that WinRT types are subject to COM-style restrictions; for instance, they offer limited support for inheritance and generics.) WinRT/Metro does not supersede the full .NET Framework. The latter is still recommended (and necessary) for standard desktop and server-side development, and has the following advantages: • Programs are not restricted to running in a sandbox. • Programs can use the entire .NET Framework and any third-party library. • Application distribution does not rely on the Windows Store. • Applications can target the latest Framework version without requiring users to have the latest OS version. What’s New in C# 5.0 C# 5.0’s big new feature is support for asynchronous functions via two new keywords, async and await. Asynchronous functions enable asynchronous continu- ations, which make it easier to write responsive and thread-safe rich-client applica- tions. They also make it easy to write highly concurrent and efficient I/O-bound applications that don’t tie up a thread resource per operation. We cover asynchronous functions in detail in Chapter 14. What’s New in C# 4.0 The features new to C# 4.0 were: • Dynamic binding • Optional parameters and named arguments • Type variance with generic interfaces and delegates • COM interoperability improvements Dynamic binding (Chapters 4 and 20) defers binding—the process of resolving types and members—from compile time to runtime and is useful in scenarios that would otherwise require complicated reflection code. Dynamic binding is also useful when interoperating with dynamic languages and COM components. 6 | Chapter 1: Introducing C# and the .NET Framework Optional parameters (Chapter 2) allow functions to specify default parameter values so that callers can omit arguments and named arguments allow a function caller to identify an argument by name rather than position. Type variance rules were relaxed in C# 4.0 (Chapters 3 and 4), such that type pa- rameters in generic interfaces and generic delegates can be marked as covariant or contravariant, allowing more natural type conversions. COM interoperability (Chapter 25) was enhanced in C# 4.0 in three ways. First, arguments can be passed by reference without the ref keyword (particularly useful in conjunction with optional parameters). Second, assemblies that contain COM interop types can be linked rather than referenced. Linked interop types support type equivalence, avoiding the need for Primary Interop Assemblies and putting an end to versioning and deployment headaches. Third, functions that return COM-Variant types from linked interop types are mapped to dynamic rather than object, elimi- nating the need for casting. What’s New in C# 3.0 The features added to C# 3.0 were mostly centered on Language Integrated Query capabilities or LINQ for short. LINQ enables queries to be written directly within a C# program and checked statically for correctness, and query both local collections (such as lists or XML documents) or remote data sources (such as a database). The C# 3.0 features added to support LINQ comprised implicitly typed local variables, anonymous types, object initializers, lambda expressions, extension methods, query expressions and expression trees. Implicitly typed local variables (var keyword, Chapter 2) let you omit the variable type in a declaration statement, allowing the compiler to infer it. This reduces clutter as well as allowing anonymous types (Chapter 4), which are simple classes created on the fly that are commonly used in the final output of LINQ queries. Arrays can also be implicitly typed (Chapter 2). Object initializers (Chapter 3) simplify object construction by allowing properties to be set inline after the constructor call. Object initializers work with both named and anonymous types. Lambda expressions (Chapter 4) are miniature functions created by the compiler on the fly, and are particularly useful in “fluent” LINQ queries (Chapter 8). Extension methods (Chapter 4) extend an existing type with new methods (without altering the type’s definition), making static methods feel like instance methods. LINQ’s query operators are implemented as extension methods. Query expressions (Chapter 8) provide a higher-level syntax for writing LINQ queries that can be substantially simpler when working with multiple sequences or range variables. Expression trees (Chapter 8) are miniature code DOMs (Document Object Models) that describe lambda expressions assigned to the special type Expression. Expression trees make it possible for LINQ queries to execute remotely (e.g., What’s New in C# 3.0 | 7 Introduction on a database server) because they can be introspected and translated at runtime (e.g., into a SQL statement). C# 3.0 also added automatic properties and partial methods. Automatic properties (Chapter 3) cut the work in writing properties that simply get/ set a private backing field by having the compiler do that work automatically. Partial methods (Chapter 3) let an auto-generated partial class provide customizable hooks for manual authoring which “melt away” if unused. 8 | Chapter 1: Introducing C# and the .NET Framework 2C# Language Basics In this chapter, we introduce the basics of the C# language. All programs and code snippets in this and the following two chapters are available as interactive samples in LINQPad. Working through these samples in conjunction with the book accelerates learning in that you can edit the samples and in- stantly see the results without needing to set up projects and solutions in Visual Studio. To download the samples, click the Samples tab in LINQPad and then click “Download more samples.” LINQPad is free— go to http://www.linqpad.net. A First C# Program Here is a program that multiplies 12 by 30 and prints the result, 360, to the screen. The double forward slash indicates that the remainder of a line is a comment. using System; // Importing namespace class Test // Class declaration { static void Main() // Method declaration { int x = 12 * 30; // Statement 1 Console.WriteLine (x); // Statement 2 } // End of method } // End of class At the heart of this program lie two statements: int x = 12 * 30; Console.WriteLine (x); 9 Statements in C# execute sequentially and are terminated by a semicolon (or a code block, as we’ll see later). The first statement computes the expression 12 * 30 and stores the result in a local variable, named x, which is an integer type. The second statement calls the Console class’s WriteLine method, to print the variable x to a text window on the screen. A method performs an action in a series of statements, called a statement block—a pair of braces containing zero or more statements. We defined a single method named Main: static void Main() { ... } Writing higher-level functions that call upon lower-level functions simplifies a pro- gram. We can refactor our program with a reusable method that multiplies an integer by 12 as follows: using System; class Test { static void Main() { Console.WriteLine (FeetToInches (30)); // 360 Console.WriteLine (FeetToInches (100)); // 1200 } static int FeetToInches (int feet) { int inches = feet * 12; return inches; } } A method can receive input data from the caller by specifying parameters and out- put data back to the caller by specifying a return type. We defined a method called FeetToInches that has a parameter for inputting feet, and a return type for outputting inches: static int FeetToInches (int feet ) {...} The literals 30 and 100 are the arguments passed to the FeetToInches method. The Main method in our example has empty parentheses because it has no parameters, and is void because it doesn’t return any value to its caller: static void Main() C# recognizes a method called Main as signaling the default entry point of execution. The Main method may optionally return an integer (rather than void) in order to return a value to the execution environment (where a non-zero value typically indi- cates an error). The Main method can also optionally accept an array of strings as a parameter (that will be populated with any arguments passed to the executable). 10 | Chapter 2: C# Language Basics For example: static int Main (string[] args) {...} An array (such as string[]) represents a fixed number of ele- ments of a particular type. Arrays are specified by placing square brackets after the element type and are described in “Ar- rays” on page 34. Methods are one of several kinds of functions in C#. Another kind of function we used was the * operator, used to perform multiplication. There are also construc- tors, properties, events, indexers, and finalizers. In our example, the two methods are grouped into a class. A class groups function members and data members to form an object-oriented building block. The Con sole class groups members that handle command-line input/output functionality, such as the WriteLine method. Our Test class groups two methods—the Main method and the FeetToInches method. A class is a kind of type, which we will examine in “Type Basics” on page 15. At the outermost level of a program, types are organized into namespaces. The using directive was used to make the System namespace available to our application, to use the Console class. We could define all our classes within the TestPrograms namespace, as follows: using System; namespace TestPrograms { class Test {...} class Test2 {...} } The .NET Framework is organized into nested namespaces. For example, this is the namespace that contains types for handling text: using System.Text; The using directive is there for convenience; you can also refer to a type by its fully qualified name, which is the type name prefixed with its namespace, such as System.Text.StringBuilder. Compilation The C# compiler compiles source code, specified as a set of files with the .cs exten- sion, into an assembly. An assembly is the unit of packaging and deployment in .NET. An assembly can be either an application or a library. A normal console or Windows application has a Main method and is an .exe file. A library is a .dll and is equivalent to an .exe without an entry point. Its purpose is to be called upon (ref- erenced) by an application or by other libraries. The .NET Framework is a set of libraries. A First C# Program | 11 C# Basics The name of the C# compiler is csc.exe. You can either use an IDE such as Visual Studio to compile, or call csc manually from the command line. To compile manually, first save a program to a file such as MyFirstProgram.cs, and then go to the command line and invoke csc (located under %SystemRoot%\Microsoft.NET \Framework\ where %SystemRoot% is your Windows direc- tory) as follows: csc MyFirstProgram.cs This produces an application named MyFirstProgram.exe. To produce a library (.dll), do the following: csc /target:library MyFirstProgram.cs We explain assemblies in detail in Chapter 18. Syntax C# syntax is inspired by C and C++ syntax. In this section, we will describe C#’s elements of syntax, using the following program: using System; class Test { static void Main() { int x = 12 * 30; Console.WriteLine (x); } } Identifiers and Keywords Identifiers are names that programmers choose for their classes, methods, variables, and so on. These are the identifiers in our example program, in the order they appear: System Test Main x Console WriteLine An identifier must be a whole word, essentially made up of Unicode characters starting with a letter or underscore. C# identifiers are case-sensitive. By convention, parameters, local variables, and private fields should be in camel case (e.g., myVari able), and all other identifiers should be in Pascal case (e.g., MyMethod). Keywords are names reserved by the compiler that you can’t use as identifiers. These are the keywords in our example program: using class static void int 12 | Chapter 2: C# Language Basics Here is the full list of C# keywords: abstract as base bool break byte case catch char checked class const continue decimal default delegate do double else enum event explicit extern false finally fixed float for foreach goto if implicit in int interface internal is lock long namespace new null object operator out override params private protected public readonly ref return sbyte sealed short sizeof stackalloc static string struct switch this throw true try typeof uint ulong unchecked unsafe ushort using virtual void volatile while Avoiding conflicts If you really want to use an identifier that clashes with a keyword, you can do so by qualifying it with the @ prefix. For instance: class class {...} // Illegal class @class {...} // Legal The @ symbol doesn’t form part of the identifier itself. So @myVariable is the same as myVariable. The @ prefix can be useful when consuming libraries written in other .NET languages that have different keywords. Contextual keywords Some keywords are contextual, meaning they can also be used as identifiers— without an @ symbol. These are: add ascending async dynamic equals from in into join partial remove select where yield Syntax | 13 C# Basics await by descending get global group let on orderby set value var With contextual keywords, ambiguity cannot arise within the context in which they are used. Literals, Punctuators, and Operators Literals are primitive pieces of data lexically embedded into the program. The literals we used in our example program are 12 and 30. Punctuators help demarcate the structure of the program. These are the punctuators we used in our example program: { } ; The braces group multiple statements into a statement block. The semicolon terminates a statement. (Statement blocks, however, do not require a semicolon.) Statements can wrap multiple lines: Console.WriteLine (1 + 2 + 3 + 4 + 5 + 6 + 7 + 8 + 9 + 10); An operator transforms and combines expressions. Most operators in C# are de- noted with a symbol, such as the multiplication operator, *. We will discuss oper- ators in more detail later in the chapter. These are the operators we used in our example program: . () * = A period denotes a member of something (or a decimal point with numeric literals). Parentheses are used when declaring or calling a method; empty parentheses are used when the method accepts no arguments. An equals sign performs assignment. (The double equals sign, ==, performs equality comparison, as we’ll see later.) Comments C# offers two different styles of source-code documentation: single-line comments and multiline comments. A single-line comment begins with a double forward slash and continues until the end of the line. For example: int x = 3; // Comment about assigning 3 to x A multiline comment begins with /* and ends with */. For example: int x = 3; /* This is a comment that spans two lines */ Comments may embed XML documentation tags, explained in “XML Documen- tation” on page 182 in Chapter 4. 14 | Chapter 2: C# Language Basics Type Basics A type defines the blueprint for a value. In our example, we used two literals of type int with values 12 and 30. We also declared a variable of type int whose name was x: static void Main() { int x = 12 * 30; Console.WriteLine (x); } A variable denotes a storage location that can contain different values over time. In contrast, a constant always represents the same value (more on this later): const int y = 360; All values in C# are instances of a type. The meaning of a value, and the set of possible values a variable can have, is determined by its type. Predefined Type Examples Predefined types are types that are specially supported by the compiler. The int type is a predefined type for representing the set of integers that fit into 32 bits of memory, from −231 to 231−1. We can perform functions such as arithmetic with instances of the int type as follows: int x = 12 * 30; Another predefined C# type is string. The string type represents a sequence of characters, such as “.NET” or “http://oreilly.com”. We can work with strings by calling functions on them as follows: string message = "Hello world"; string upperMessage = message.ToUpper(); Console.WriteLine (upperMessage); // HELLO WORLD int x = 2012; message = message + x.ToString(); Console.WriteLine (message); // Hello world2012 The predefined bool type has exactly two possible values: true and false. The bool type is commonly used to conditionally branch execution flow based with an if statement. For example: bool simpleVar = false; if (simpleVar) Console.WriteLine ("This will not print"); int x = 5000; bool lessThanAMile = x < 5280; if (lessThanAMile) Console.WriteLine ("This will print"); Type Basics | 15 C# Basics In C#, predefined types (also referred to as built-in types) are recognized with a C# keyword. The System namespace in the .NET Framework contains many important types that are not predefined by C# (e.g., DateTime). Custom Type Examples Just as we can build complex functions from simple functions, we can build complex types from primitive types. In this example, we will define a custom type named UnitConverter—a class that serves as a blueprint for unit conversions: using System; public class UnitConverter { int ratio; // Field public UnitConverter (int unitRatio) {ratio = unitRatio; } // Constructor public int Convert (int unit) {return unit * ratio; } // Method } class Test { static void Main() { UnitConverter feetToInchesConverter = new UnitConverter (12); UnitConverter milesToFeetConverter = new UnitConverter (5280); Console.WriteLine (feetToInchesConverter.Convert(30)); // 360 Console.WriteLine (feetToInchesConverter.Convert(100)); // 1200 Console.WriteLine (feetToInchesConverter.Convert( milesToFeetConverter.Convert(1))); // 63360 } } Members of a type A type contains data members and function members. The data member of UnitConverter is the field called ratio. The function members of UnitConverter are the Convert method and the UnitConverter’s constructor. Symmetry of predefined types and custom types A beautiful aspect of C# is that predefined types and custom types have few differ- ences. The predefined int type serves as a blueprint for integers. It holds data—32 bits—and provides function members that use that data, such as ToString. Similarly, our custom UnitConverter type acts as a blueprint for unit conversions. It holds data —the ratio—and provides function members to use that data. Constructors and instantiation Data is created by instantiating a type. Predefined types can be instantiated simply by using a literal such as 12 or "Hello, world". The new operator creates instances of 16 | Chapter 2: C# Language Basics a custom type. We created and declared an instance of the UnitConverter type with this statement: UnitConverter feetToInchesConverter = new UnitConverter (12); Immediately after the new operator instantiates an object, the object’s constructor is called to perform initialization. A constructor is defined like a method, except that the method name and return type are reduced to the name of the enclosing type: public class UnitConverter { ... public UnitConverter (int unitRatio) { ratio = unitRatio; } ... } Instance versus static members The data members and function members that operate on the instance of the type are called instance members. The UnitConverter’s Convert method and the int’s ToString method are examples of instance members. By default, members are in- stance members. Data members and function members that don’t operate on the instance of the type, but rather on the type itself, must be marked as static. The Test.Main and Con sole.WriteLine methods are static methods. The Console class is actually a static class, which means all its members are static. You never actually create instances of a Console—one console is shared across the whole application. To contrast instance from static members, in the following code the instance field Name pertains to an instance of a particular Panda, whereas Population pertains to the set of all Panda instances: public class Panda { public string Name; // Instance field public static int Population; // Static field public Panda (string n) // Constructor { Name = n; // Assign the instance field Population = Population + 1; // Increment the static Population field } } The following code creates two instances of the Panda, prints their names, and then prints the total population: using System; class Test { static void Main() { Panda p1 = new Panda ("Pan Dee"); Panda p2 = new Panda ("Pan Dah"); Type Basics | 17 C# Basics Console.WriteLine (p1.Name); // Pan Dee Console.WriteLine (p2.Name); // Pan Dah Console.WriteLine (Panda.Population); // 2 } } The public keyword The public keyword exposes members to other classes. In this example, if the Name field in Panda was not public, the Test class could not access it. Marking a member public is how a type communicates: “Here is what I want other types to see— everything else is my own private implementation details.” In object-oriented terms, we say that the public members encapsulate the private members of the class. Conversions C# can convert between instances of compatible types. A conversion always creates a new value from an existing one. Conversions can be either implicit or explicit: implicit conversions happen automatically, and explicit conversions require a cast. In the following example, we implicitly convert an int to a long type (which has twice the bitwise capacity of an int) and explicitly cast an int to a short type (which has half the capacity of an int): int x = 12345; // int is a 32-bit integer long y = x; // Implicit conversion to 64-bit integer short z = (short)x; // Explicit conversion to 16-bit integer Implicit conversions are allowed when both of the following are true: • The compiler can guarantee they will always succeed. • No information is lost in conversion.1 Conversely, explicit conversions are required when one of the following is true: • The compiler cannot guarantee they will always succeed. • Information may be lost during conversion. (If the compiler can determine that a conversion will always fail, both kinds of con- version are prohibited. Conversions that involve generics can also fail in certain conditions—see “Type Parameters and Conversions” on page 113 in Chapter 3.) 1. A minor caveat is that very large long values lose some precision when converted to double. 18 | Chapter 2: C# Language Basics The numeric conversions that we just saw are built into the lan- guage. C# also supports reference conversions and boxing conversions (see Chapter 3) as well as custom conversions (see “Operator Overloading” on page 158 in Chapter 4). The com- piler doesn’t enforce the aforementioned rules with custom conversions, so it’s possible for badly designed types to behave otherwise. Value Types Versus Reference Types All C# types fall into the following categories: • Value types • Reference types • Generic type parameters • Pointer types In this section, we’ll describe value types and reference types. We’ll cover generic type parameters in “Gener- ics” on page 106 in Chapter 3, and pointer types in “Unsafe Code and Pointers” on page 177 in Chapter 4. Value types comprise most built-in types (specifically, all numeric types, the char type, and the bool type) as well as custom struct and enum types. Reference types comprise all class, array, delegate, and interface types. (This includes the predefined string type.) The fundamental difference between value types and reference types is how they are handled in memory. Value types The content of a value type variable or constant is simply a value. For example, the content of the built-in value type, int, is 32 bits of data. You can define a custom value type with the struct keyword (see Figure 2-1): public struct Point { public int X, Y; } Figure 2-1. A value-type instance in memory The assignment of a value-type instance always copies the instance. Type Basics | 19 C# Basics For example: static void Main() { Point p1 = new Point(); p1.X = 7; Point p2 = p1; // Assignment causes copy Console.WriteLine (p1.X); // 7 Console.WriteLine (p2.X); // 7 p1.X = 9; // Change p1.X Console.WriteLine (p1.X); // 9 Console.WriteLine (p2.X); // 7 } Figure 2-2 shows that p1 and p2 have independent storage. Figure 2-2. Assignment copies a value-type instance Reference types A reference type is more complex than a value type, having two parts: an object and the reference to that object. The content of a reference-type variable or constant is a reference to an object that contains the value. Here is the Point type from our previous example rewritten as a class, rather than a struct (shown in Figure 2-3): public class Point { public int X, Y; } Figure 2-3. A reference-type instance in memory Assigning a reference-type variable copies the reference, not the object instance. This allows multiple variables to refer to the same object—something not ordinarily 20 | Chapter 2: C# Language Basics possible with value types. If we repeat the previous example, but with Point now a class, an operation to p1 affects p2: static void Main() { Point p1 = new Point(); p1.X = 7; Point p2 = p1; // Copies p1 reference Console.WriteLine (p1.X); // 7 Console.WriteLine (p2.X); // 7 p1.X = 9; // Change p1.X Console.WriteLine (p1.X); // 9 Console.WriteLine (p2.X); // 9 } Figure 2-4 shows that p1 and p2 are two references that point to the same object. Figure 2-4. Assignment copies a reference Null A reference can be assigned the literal null, indicating that the reference points to no object: class Point {...} ... Point p = null; Console.WriteLine (p == null); // True // The following line generates a runtime error // (a NullReferenceException is thrown): Console.WriteLine (p.X); In contrast, a value type cannot ordinarily have a null value: struct Point {...} ... Point p = null; // Compile-time error int x = null; // Compile-time error Type Basics | 21 C# Basics C# also has a construct called nullable types for representing value-type nulls (see “Nullable Types” on page 153 in Chap- ter 4). Storage overhead Value-type instances occupy precisely the memory required to store their fields. In this example, Point takes eight bytes of memory: struct Point { int x; // 4 bytes int y; // 4 bytes } Technically, the CLR positions fields within the type at an ad- dress that’s a multiple of the fields’ size (up to a maximum of eight bytes). Thus, the following actually consumes 16 bytes of memory (with the seven bytes following the first field “wasted”): struct A { byte b; long l; } Reference types require separate allocations of memory for the reference and object. The object consumes as many bytes as its fields, plus additional administrative overhead. The precise overhead is intrinsically private to the implementation of the .NET runtime, but at minimum the overhead is eight bytes, used to store a key to the object’s type, as well as temporary information such as its lock state for mul- tithreading and a flag to indicate whether it has been fixed from movement by the garbage collector. Each reference to an object requires an extra four or eight bytes, depending on whether the .NET runtime is running on a 32- or 64-bit platform. Predefined Type Taxonomy The predefined types in C# are: Value types • Numeric • Signed integer (sbyte, short, int, long) • Unsigned integer (byte, ushort, uint, ulong) • Real number (float, double, decimal) • Logical (bool) • Character (char) Reference types • String (string) • Object (object) 22 | Chapter 2: C# Language Basics Predefined types in C# alias Framework types in the System namespace. There is only a syntactic difference between these two statements: int i = 5; System.Int32 i = 5; The set of predefined value types excluding decimal are known as primitive types in the CLR. Primitive types are so called because they are supported directly via in- structions in compiled code, and this usually translates to direct support on the underlying processor. For example: // Underlying hexadecimal representation int i = 7; // 0x7 bool b = true; // 0x1 char c = 'A'; // 0x41 float f = 0.5f; // uses IEEE floating-point encoding The System.IntPtr and System.UIntPtr types are also primitive (see Chapter 25). Numeric Types C# has the predefined numeric types shown in Table 2-1. Table 2-1. Predefined numeric types in C# C# type System type Suffix Size Range Integral—signed sbyte SByte 8 bits −27 to 27−1 short Int16 16 bits −215 to 215−1 int Int32 32 bits −231 to 231−1 long Int64 L 64 bits −263 to 263−1 Integral—unsigned byte Byte 8 bits 0 to 28−1 ushort UInt16 16 bits 0 to 216−1 uint UInt32 U 32 bits 0 to 232−1 ulong UInt64 UL 64 bits 0 to 264−1 Real float Single F 32 bits ± (~10−45 to 1038) double Double D 64 bits ± (~10−324 to 10308) decimal Decimal M 128 bits ± (~10−28 to 1028) Of the integral types, int and long are first-class citizens and are favored by both C# and the runtime. The other integral types are typically used for interoperability or when space efficiency is paramount. Numeric Types | 23 C# Basics Of the real number types, float and double are called floating-point types2 and are typically used for scientific calculations. The decimal type is typically used for financial calculations, where base-10–accurate arithmetic and high precision are required. Numeric Literals Integral literals can use decimal or hexadecimal notation; hexadecimal is denoted with the 0x prefix. For example: int x = 127; long y = 0x7F; Real literals can use decimal and/or exponential notation. For example: double d = 1.5; double million = 1E06; Numeric literal type inference By default, the compiler infers a numeric literal to be either double or an integral type: • If the literal contains a decimal point or the exponential symbol (E), it is a double. • Otherwise, the literal’s type is the first type in this list that can fit the literal’s value: int, uint, long, and ulong. For example: Console.WriteLine ( 1.0.GetType()); // Double (double) Console.WriteLine ( 1E06.GetType()); // Double (double) Console.WriteLine ( 1.GetType()); // Int32 (int) Console.WriteLine ( 0xF0000000.GetType()); // UInt32 (uint) Numeric suffixes Numeric suffixes explicitly define the type of a literal. Suffixes can be either lower- or uppercase, and are as follows: Category C# type Example F float float f = 1.0F; D double double d = 1D; M decimal decimal d = 1.0M; U uint uint i = 1U; L long long i = 1L; UL ulong ulong i = 1UL; 2. Technically, decimal is a floating-point type too, although it’s not referred to as such in the C# language specification. 24 | Chapter 2: C# Language Basics The suffixes U and L are rarely necessary, because the uint, long, and ulong types can nearly always be either inferred or implicitly converted from int: long i = 5; // Implicit lossless conversion from int literal to long The D suffix is technically redundant, in that all literals with a decimal point are inferred to be double. And you can always add a decimal point to a numeric literal: double x = 4.0; The F and M suffixes are the most useful and should always be applied when speci- fying float or decimal literals. Without the F suffix, the following line would not compile, because 4.5 would be inferred to be of type double, which has no implicit conversion to float: float f = 4.5F; The same principle is true for a decimal literal: decimal d = −1.23M; // Will not compile without the M suffix. We describe the semantics of numeric conversions in detail in the following section. Numeric Conversions Integral to integral conversions Integral conversions are implicit when the destination type can represent every pos- sible value of the source type. Otherwise, an explicit conversion is required. For example: int x = 12345; // int is a 32-bit integral long y = x; // Implicit conversion to 64-bit integral short z = (short)x; // Explicit conversion to 16-bit integral Floating-point to floating-point conversions A float can be implicitly converted to a double, since a double can represent every possible value of a float. The reverse conversion must be explicit. Floating-point to integral conversions All integral types may be implicitly converted to all floating-point types: int i = 1; float f = i; The reverse conversion must be explicit: int i2 = (int)f; When you cast from a floating-point number to an integral, any fractional portion is truncated; no rounding is performed. The static class System.Convert provides methods that round while converting between various numeric types (see Chapter 6). Numeric Types | 25 C# Basics Implicitly converting a large integral type to a floating-point type preserves magni- tude but may occasionally lose precision. This is because floating-point types always have more magnitude than integral types, but may have less precision. Rewriting our example with a larger number demonstrates this: int i1 = 100000001; float f = i1; // Magnitude preserved, precision lost int i2 = (int)f; // 100000000 Decimal conversions All integral types can be implicitly converted to the decimal type, since a decimal can represent every possible C# integral value. All other numeric conversions to and from a decimal type must be explicit. Arithmetic Operators The arithmetic operators (+, -, *, /, %) are defined for all numeric types except the 8- and 16-bit integral types: + Addition - Subtraction * Multiplication / Division % Remainder after division Increment and Decrement Operators The increment and decrement operators (++, −−) increment and decrement numeric types by 1. The operator can either follow or precede the variable, depending on whether you want its value before or after the increment/decrement. For example: int x = 0, y = 0; Console.WriteLine (x++); // Outputs 0; x is now 1 Console.WriteLine (++y); // Outputs 1; y is now 1 Specialized Integral Operations Integral division Division operations on integral types always truncate remainders (round towards zero). Dividing by a variable whose value is zero generates a runtime error (a DivideByZeroException): int a = 2 / 3; // 0 int b = 0; int c = 5 / b; // throws DivideByZeroException Dividing by the literal or constant 0 generates a compile-time error. 26 | Chapter 2: C# Language Basics Integral overflow At runtime, arithmetic operations on integral types can overflow. By default, this happens silently—no exception is thrown and the result exhibits “wraparound” behavior, as though the computation was done on a larger integer type and the extra significant bits discarded. For example, decrementing the minimum possible int value results in the maximum possible int value: int a = int.MinValue; a--; Console.WriteLine (a == int.MaxValue); // True Integral arithmetic overflow check operators The checked operator tells the runtime to generate an OverflowException rather than overflowing silently when an integral expression or statement exceeds the arithmetic limits of that type. The checked operator affects expressions with the ++, −−, +, − (binary and unary), *, /, and explicit conversion operators between integral types. The checked operator has no effect on the double and float types (which overflow to special “infinite” values, as we’ll see soon) and no effect on the decimal type (which is always checked). checked can be used around either an expression or a statement block. For example: int a = 1000000; int b = 1000000; int c = checked (a * b); // Checks just the expression. checked // Checks all expressions { // in statement block. ... c = a * b; ... } You can make arithmetic overflow checking the default for all expressions in a program by compiling with the /checked+ command-line switch (in Visual Studio, go to Advanced Build Settings). If you then need to disable overflow checking just for specific expressions or statements, you can do so with the unchecked operator. For example, the following code will not throw exceptions—even if compiled with /checked+: int x = int.MaxValue; int y = unchecked (x + 1); unchecked { int z = x + 1; } Numeric Types | 27 C# Basics Overflow checking for constant expressions Regardless of the /checked compiler switch, expressions evaluated at compile time are always overflow-checked—unless you apply the unchecked operator: int x = int.MaxValue + 1; // Compile-time error int y = unchecked (int.MaxValue + 1); // No errors Bitwise operators C# supports the following bitwise operators: Operator Meaning Sample expression Result ~ Complement ~0xfU 0xfffffff0U & And 0xf0 & 0x33 0x30 | Or 0xf0 | 0x33 0xf3 ^ Exclusive Or 0xff00 ^ 0x0ff0 0xf0f0 << Shift left 0x20 << 2 0x80 >> Shift right 0x20 >> 1 0x10 8- and 16-Bit Integrals The 8- and 16-bit integral types are byte, sbyte, short, and ushort. These types lack their own arithmetic operators, so C# implicitly converts them to larger types as required. This can cause a compile-time error when trying to assign the result back to a small integral type: short x = 1, y = 1; short z = x + y; // Compile-time error In this case, x and y are implicitly converted to int so that the addition can be per- formed. This means the result is also an int, which cannot be implicitly cast back to a short (because it could cause loss of data). To make this compile, we must add an explicit cast: short z = (short) (x + y); // OK Special Float and Double Values Unlike integral types, floating-point types have values that certain operations treat specially. These special values are NaN (Not a Number), +∞, −∞, and −0. The float and double classes have constants for NaN, +∞, and −∞, as well as other values (MaxValue, MinValue, and Epsilon). For example: Console.WriteLine (double.NegativeInfinity); // -Infinity 28 | Chapter 2: C# Language Basics The constants that represent special values for double and float are as follows: Special value Double constant Float constant NaN double.NaN float.NaN +∞ double.PositiveInfinity float.PositiveInfinity −∞ double.NegativeInfinity float.NegativeInfinity −0 −0.0 −0.0f Dividing a nonzero number by zero results in an infinite value. For example: Console.WriteLine ( 1.0 / 0.0); // Infinity Console.WriteLine (−1.0 / 0.0); // -Infinity Console.WriteLine ( 1.0 / −0.0); // -Infinity Console.WriteLine (−1.0 / −0.0); // Infinity Dividing zero by zero, or subtracting infinity from infinity, results in a NaN. For example: Console.WriteLine ( 0.0 / 0.0); // NaN Console.WriteLine ((1.0 / 0.0) − (1.0 / 0.0)); // NaN When using ==, a NaN value is never equal to another value, even another NaN value: Console.WriteLine (0.0 / 0.0 == double.NaN); // False To test whether a value is NaN, you must use the float.IsNaN or double.IsNaN method: Console.WriteLine (double.IsNaN (0.0 / 0.0)); // True When using object.Equals, however, two NaN values are equal: Console.WriteLine (object.Equals (0.0 / 0.0, double.NaN)); // True NaNs are sometimes useful in representing special values. In WPF, double.NaN represents a measurement whose value is “Automatic”. Another way to represent such a value is with a nullable type (Chapter 4); another is with a custom struct that wraps a numeric type and adds an additional field (Chapter 3). float and double follow the specification of the IEEE 754 format types, supported natively by almost all processors. You can find detailed information on the behavior of these types at http://www.ieee.org. double Versus decimal double is useful for scientific computations (such as computing spatial coordinates). decimal is useful for financial computations and values that are “man-made” rather than the result of real-world measurements. Here’s a summary of the differences. Numeric Types | 29 C# Basics Category Double decimal Internal representation Base 2 Base 10 Decimal precision 15–16 significant figures 28–29 significant figures Range ±(~10−324 to ~10308) ±(~10−28 to ~1028) Special values +0, −0, +∞, −∞, and NaN None Speed Native to processor Non-native to processor (about 10 times slower than double) Real Number Rounding Errors float and double internally represent numbers in base 2. For this reason, only num- bers expressible in base 2 are represented precisely. Practically, this means most literals with a fractional component (which are in base 10) will not be represented precisely. For example: float tenth = 0.1f; // Not quite 0.1 float one = 1f; Console.WriteLine (one - tenth * 10f); // −1.490116E-08 This is why float and double are bad for financial calculations. In contrast, deci mal works in base 10 and so can precisely represent numbers expressible in base 10 (as well as its factors, base 2 and base 5). Since real literals are in base 10, decimal can precisely represent numbers such as 0.1. However, neither double nor decimal can precisely represent a fractional number whose base 10 representation is recurring: decimal m = 1M / 6M; // 0.1666666666666666666666666667M double d = 1.0 / 6.0; // 0.16666666666666666 This leads to accumulated rounding errors: decimal notQuiteWholeM = m+m+m+m+m+m; // 1.0000000000000000000000000002M double notQuiteWholeD = d+d+d+d+d+d; // 0.99999999999999989 which breaks equality and comparison operations: Console.WriteLine (notQuiteWholeM == 1M); // False Console.WriteLine (notQuiteWholeD < 1.0); // True Boolean Type and Operators C#’s bool type (aliasing the System.Boolean type) is a logical value that can be as- signed the literal true or false. Although a Boolean value requires only one bit of storage, the runtime will use one byte of memory, since this is the minimum chunk that the runtime and processor can efficiently work with. To avoid space inefficiency in the case of arrays, the Framework provides a BitArray class in the System.Collections namespace that is designed to use just one bit per Boolean value. 30 | Chapter 2: C# Language Basics Bool Conversions No conversions can be made from the bool type to numeric types or vice versa. Equality and Comparison Operators == and != test for equality and inequality of any type, but always return a bool value.3 Value types typically have a very simple notion of equality: int x = 1; int y = 2; int z = 1; Console.WriteLine (x == y); // False Console.WriteLine (x == z); // True For reference types, equality, by default, is based on reference, as opposed to the actual value of the underlying object (more on this in Chapter 6): public class Dude { public string Name; public Dude (string n) { Name = n; } } ... Dude d1 = new Dude ("John"); Dude d2 = new Dude ("John"); Console.WriteLine (d1 == d2); // False Dude d3 = d1; Console.WriteLine (d1 == d3); // True The equality and comparison operators, ==, !=, <, >, >=, and <=, work for all numeric types, but should be used with caution with real numbers (as we saw in “Real Num- ber Rounding Errors” on page 30). The comparison operators also work on enum type members, by comparing their underlying integral values. We describe this in “Enums” on page 102 in Chapter 3. We explain the equality and comparison operators in greater detail in “Operator Overloading” on page 158 in Chapter 4, and in “Equality Compari- son” on page 254 and “Order Comparison” on page 264 in Chapter 6. Conditional Operators The && and || operators test for and and or conditions. They are frequently used in conjunction with the ! operator, which expresses not. In this example, the UseUm brella method returns true if it’s rainy or sunny (to protect us from the rain or the sun), as long as it’s not also windy (since umbrellas are useless in the wind): static bool UseUmbrella (bool rainy, bool sunny, bool windy) { return !windy && (rainy || sunny); } 3. It’s possible to overload these operators (Chapter 4) such that they return a non-bool type, but this is almost never done in practice. Boolean Type and Operators | 31 C# Basics The && and || operators short-circuit evaluation when possible. In the preceding example, if it is windy, the expression (rainy || sunny) is not even evaluated. Short- circuiting is essential in allowing expressions such as the following to run without throwing a NullReferenceException: if (sb != null && sb.Length > 0) ... The & and | operators also test for and and or conditions: return !windy & (rainy | sunny); The difference is that they do not short-circuit. For this reason, they are rarely used in place of conditional operators. Unlike in C and C++, the & and | operators perform (non-short- circuiting) Boolean comparisons when applied to bool expres- sions. The & and | operators perform bitwise operations only when applied to numbers. The conditional operator (more commonly called the ternary operator, as it’s the only operator that takes three operands) has the form q ? a : b, where if condition q is true, a is evaluated, else b is evaluated. For example: static int Max (int a, int b) { return (a > b) ? a : b; } The conditional operator is particularly useful in LINQ queries (Chapter 8). Strings and Characters C#’s char type (aliasing the System.Char type) represents a Unicode character and occupies 2 bytes. A char literal is specified inside single quotes: char c = 'A'; // Simple character Escape sequences express characters that cannot be expressed or interpreted literally. An escape sequence is a backslash followed by a character with a special meaning. For example: char newLine = '\n'; char backSlash = '\\'; The escape sequence characters are shown in Table 2-2. Table 2-2. Escape sequence characters Char Meaning Value \' Single quote 0x0027 \" Double quote 0x0022 \\ Backslash 0x005C 32 | Chapter 2: C# Language Basics Char Meaning Value \0 Null 0x0000 \a Alert 0x0007 \b Backspace 0x0008 \f Form feed 0x000C \n New line 0x000A \r Carriage return 0x000D \t Horizontal tab 0x0009 \v Vertical tab 0x000B The \u (or \x) escape sequence lets you specify any Unicode character via its four- digit hexadecimal code: char copyrightSymbol = '\u00A9'; char omegaSymbol = '\u03A9'; char newLine = '\u000A'; Char Conversions An implicit conversion from a char to a numeric type works for the numeric types that can accommodate an unsigned short. For other numeric types, an explicit con- version is required. String Type C#’s string type (aliasing the System.String type, covered in depth in Chapter 6) represents an immutable sequence of Unicode characters. A string literal is specified inside double quotes: string a = "Heat"; string is a reference type, rather than a value type. Its equality operators, however, follow value-type semantics: string a = "test"; string b = "test"; Console.Write (a == b); // True The escape sequences that are valid for char literals also work inside strings: string a = "Here's a tab:\t"; The cost of this is that whenever you need a literal backslash, you must write it twice: string a1 = "\\\\server\\fileshare\\helloworld.cs"; To avoid this problem, C# allows verbatim string literals. A verbatim string literal is prefixed with @ and does not support escape sequences. Strings and Characters | 33 C# Basics The following verbatim string is identical to the preceding one: string a2 = @ "\\server\fileshare\helloworld.cs"; A verbatim string literal can also span multiple lines: string escaped = "First Line\r\nSecond Line"; string verbatim = @"First Line Second Line"; // Assuming your IDE uses CR-LF line separators: Console.WriteLine (escaped == verbatim); // True You can include the double-quote character in a verbatim literal by writing it twice: string xml = @""; String concatenation The + operator concatenates two strings: string s = "a" + "b"; One of the operands may be a nonstring value, in which case ToString is called on that value. For example: string s = "a" + 5; // a5 Using the + operator repeatedly to build up a string is inefficient; a better solution is to use the System.Text.StringBuilder type (described in Chapter 6). String comparisons string does not support < and > operators for comparisons. You must use the string’s CompareTo method, described in Chapter 6. Arrays An array represents a fixed number of variables (called elements) of a particular type. The elements in an array are always stored in a contiguous block of memory, pro- viding highly efficient access. An array is denoted with square brackets after the element type. For example: char[] vowels = new char[5]; // Declare an array of 5 characters Square brackets also index the array, accessing a particular element by position: vowels[0] = 'a'; vowels[1] = 'e'; vowels[2] = 'i'; vowels[3] = 'o'; vowels[4] = 'u'; Console.WriteLine (vowels[1]); // e This prints “e” because array indexes start at 0. We can use a for loop statement to iterate through each element in the array. 34 | Chapter 2: C# Language Basics The for loop in this example cycles the integer i from 0 to 4: for (int i = 0; i < vowels.Length; i++) Console.Write (vowels[i]); // aeiou The Length property of an array returns the number of elements in the array. Once an array has been created, its length cannot be changed. The System.Collection namespace and subnamespaces provide higher-level data structures, such as dy- namically sized arrays and dictionaries. An array initialization expression lets you declare and populate an array in a single step: char[] vowels = new char[] {'a','e','i','o','u'}; or simply: char[] vowels = {'a','e','i','o','u'}; All arrays inherit from the System.Array class, providing common services for all arrays. These members include methods to get and set elements regardless of the array type, and are described in “The Array Class” on page 282 in Chapter 7. Default Element Initialization Creating an array always preinitializes the elements with default values. The default value for a type is the result of a bitwise zeroing of memory. For example, consider creating an array of integers. Since int is a value type, this allocates 1,000 integers in one contiguous block of memory. The default value for each element will be 0: int[] a = new int[1000]; Console.Write (a[123]); // 0 Value types versus reference types Whether an array element type is a value type or a reference type has important performance implications. When the element type is a value type, each element value is allocated as part of the array. For example: public struct Point { public int X, Y; } ... Point[] a = new Point[1000]; int x = a[500].X; // 0 Had Point been a class, creating the array would have merely allocated 1,000 null references: public class Point { public int X, Y; } ... Point[] a = new Point[1000]; int x = a[500].X; // Runtime error, NullReferenceException Arrays | 35 C# Basics To avoid this error, we must explicitly instantiate 1,000 Points after instantiating the array: Point[] a = new Point[1000]; for (int i = 0; i < a.Length; i++) // Iterate i from 0 to 999 a[i] = new Point(); // Set array element i with new point An array itself is always a reference type object, regardless of the element type. For instance, the following is legal: int[] a = null; Multidimensional Arrays Multidimensional arrays come in two varieties: rectangular and jagged. Rectangular arrays represent an n-dimensional block of memory, and jagged arrays are arrays of arrays. Rectangular arrays Rectangular arrays are declared using commas to separate each dimension. The fol- lowing declares a rectangular two-dimensional array, where the dimensions are 3 by 3: int[,] matrix = new int[3,3]; The GetLength method of an array returns the length for a given dimension (starting at 0): for (int i = 0; i < matrix.GetLength(0); i++) for (int j = 0; j < matrix.GetLength(1); j++) matrix[i,j] = i * 3 + j; A rectangular array can be initialized as follows (to create an array identical to the previous example): int[,] matrix = new int[,] { {0,1,2}, {3,4,5}, {6,7,8} }; Jagged arrays Jagged arrays are declared using successive square brackets to represent each di- mension. Here is an example of declaring a jagged two-dimensional array, where the outermost dimension is 3: int[][] matrix = new int[3][]; Interestingly, this is new int[3][] and not new int[][3]. Eric Lippert has written an excellent article on why this is so: see http://albahari.com/jagged. 36 | Chapter 2: C# Language Basics The inner dimensions aren’t specified in the declaration because, unlike a rectan- gular array, each inner array can be an arbitrary length. Each inner array is implicitly initialized to null rather than an empty array. Each inner array must be created manually: for (int i = 0; i < matrix.Length; i++) { matrix[i] = new int[3]; // Create inner array for (int j = 0; j < matrix[i].Length; j++) matrix[i][j] = i * 3 + j; } A jagged array can be initialized as follows (to create an array identical to the previous example with an additional element at the end): int[][] matrix = new int[][] { new int[] {0,1,2}, new int[] {3,4,5}, new int[] {6,7,8,9} }; Simplified Array Initialization Expressions There are two ways to shorten array initialization expressions. The first is to omit the new operator and type qualifications: char[] vowels = {'a','e','i','o','u'}; int[,] rectangularMatrix = { {0,1,2}, {3,4,5}, {6,7,8} }; int[][] jaggedMatrix = { new int[] {0,1,2}, new int[] {3,4,5}, new int[] {6,7,8} }; The second approach is to use the var keyword, which tells the compiler to implicitly type a local variable: var i = 3; // i is implicitly of type int var s = "sausage"; // s is implicitly of type string // Therefore: var rectMatrix = new int[,] // rectMatrix is implicitly of type int[,] { {0,1,2}, {3,4,5}, {6,7,8} Arrays | 37 C# Basics }; var jaggedMat = new int[][] // jaggedMat is implicitly of type int[][] { new int[] {0,1,2}, new int[] {3,4,5}, new int[] {6,7,8} }; Implicit typing can be taken one stage further with arrays: you can omit the type qualifier after the new keyword and have the compiler infer the array type: var vowels = new[] {'a','e','i','o','u'}; // Compiler infers char[] For this to work, the elements must all be implicitly convertible to a single type (and at least one of the elements must be of that type, and there must be exactly one best type). For example: var x = new[] {1,10000000000}; // all convertible to long Bounds Checking All array indexing is bounds-checked by the runtime. An IndexOutOfRangeExcep tion is thrown if you use an invalid index: int[] arr = new int[3]; arr[3] = 1; // IndexOutOfRangeException thrown As with Java, array bounds checking is necessary for type safety and simplifies debugging. Generally, the performance hit from bounds checking is minor, and the JIT (Just-in-Time) compiler can perform optimizations, such as determining in advance whether all indexes will be safe before entering a loop, thus avoiding a check on each iteration. In addition, C# provides “unsafe” code that can explicitly by- pass bounds checking (see “Unsafe Code and Point- ers” on page 177 in Chapter 4). Variables and Parameters A variable represents a storage location that has a modifiable value. A variable can be a local variable, parameter (value, ref, or out), field (instance or static), or array element. The Stack and the Heap The stack and the heap are the places where variables and constants reside. Each has very different lifetime semantics. 38 | Chapter 2: C# Language Basics Stack The stack is a block of memory for storing local variables and parameters. The stack logically grows and shrinks as a function is entered and exited. Consider the fol- lowing method (to avoid distraction, input argument checking is ignored): static int Factorial (int x) { if (x == 0) return 1; return x * Factorial (x-1); } This method is recursive, meaning that it calls itself. Each time the method is entered, a new int is allocated on the stack, and each time the method exits, the int is deallocated. Heap The heap is a block of memory in which objects (i.e., reference-type instances) reside. Whenever a new object is created, it is allocated on the heap, and a reference to that object is returned. During a program’s execution, the heap starts filling up as new objects are created. The runtime has a garbage collector that periodically deallocates objects from the heap, so your computer does not run out of memory. An object is eligible for deallocation as soon as it’s not referenced by anything that’s itself “alive.” In the following example, we start by creating a StringBuilder object referenced by the variable ref1, and then write out its content. That StringBuilder object is then immediately eligible for garbage collection, because nothing subsequently uses it. Then, we create another StringBuilder referenced by variable ref2, and copy that reference to ref3. Even though ref2 is not used after that point, ref3 keeps the same StringBuilder object alive—ensuring that it doesn’t become eligible for collection until we’ve finished using ref3. using System; using System.Text; class Test { static void Main() { StringBuilder ref1 = new StringBuilder ("object1"); Console.WriteLine (ref1); // The StringBuilder referenced by ref1 is now eligible for GC. StringBuilder ref2 = new StringBuilder ("object2"); StringBuilder ref3 = ref2; // The StringBuilder referenced by ref2 is NOT yet eligible for GC. Console.WriteLine (ref3); // object2 } } Variables and Parameters | 39 C# Basics Value-type instances (and object references) live wherever the variable was declared. If the instance was declared as a field within an object, or as an array element, that instance lives on the heap. You can’t explicitly delete objects in C#, as you can in C++. An unreferenced object is eventually collected by the garbage collector. The heap also stores static fields and constants. Unlike objects allocated on the heap (which can get garbage-collected), these live until the application domain is torn down. Definite Assignment C# enforces a definite assignment policy. In practice, this means that outside of an unsafe context, it’s impossible to access uninitialized memory. Definite assignment has three implications: • Local variables must be assigned a value before they can be read. • Function arguments must be supplied when a method is called (unless marked as optional—see “Optional parameters” on page 45). • All other variables (such as fields and array elements) are automatically initial- ized by the runtime. For example, the following code results in a compile-time error: static void Main() { int x; Console.WriteLine (x); // Compile-time error } Fields and array elements are automatically initialized with the default values for their type. The following code outputs 0, because array elements are implicitly assigned to their default values: static void Main() { int[] ints = new int[2]; Console.WriteLine (ints[0]); // 0 } The following code outputs 0, because fields are implicitly assigned a default value: class Test { static int x; static void Main() { Console.WriteLine (x); } // 0 } 40 | Chapter 2: C# Language Basics Default Values All type instances have a default value. The default value for the predefined types is the result of a bitwise zeroing of memory: Type Default value All reference types null All numeric and enum types 0 char type '\0' bool type false You can obtain the default value for any type using the default keyword (in practice, this is useful with generics which we’ll cover in Chapter 3): decimal d = default (decimal); The default value in a custom value type (i.e., struct) is the same as the default value for each field defined by the custom type. Parameters A method has a sequence of parameters. Parameters define the set of arguments that must be provided for that method. In this example, the method Foo has a single parameter named p, of type int: static void Foo (int p) { p = p + 1; // Increment p by 1 Console.WriteLine(p); // Write p to screen } static void Main() { Foo (8); } You can control how parameters are passed with the ref and out modifiers: Parameter modifier Passed by Variable must be definitely assigned (None) Value Going in ref Reference Going in out Reference Going out Passing arguments by value By default, arguments in C# are passed by value, which is by far the most common case. This means a copy of the value is created when passed to the method: class Test { static void Foo (int p) { p = p + 1; // Increment p by 1 Console.WriteLine (p); // Write p to screen Variables and Parameters | 41 C# Basics } static void Main() { int x = 8; Foo (x); // Make a copy of x Console.WriteLine (x); // x will still be 8 } } Assigning p a new value does not change the contents of x, since p and x reside in different memory locations. Passing a reference-type argument by value copies the reference, but not the object. In the following example, Foo sees the same StringBuilder object that Main instan- tiated, but has an independent reference to it. In other words, sb and fooSB are sep- arate variables that reference the same StringBuilder object: class Test { static void Foo (StringBuilder fooSB) { fooSB.Append ("test"); fooSB = null; } static void Main() { StringBuilder sb = new StringBuilder(); Foo (sb); Console.WriteLine (sb.ToString()); // test } } Because fooSB is a copy of a reference, setting it to null doesn’t make sb null. (If, however, fooSB was declared and called with the ref modifier, sb would become null.) The ref modifier To pass by reference, C# provides the ref parameter modifier. In the following example, p and x refer to the same memory locations: class Test { static void Foo (ref int p) { p = p + 1; // Increment p by 1 Console.WriteLine (p); // Write p to screen } static void Main() { int x = 8; Foo (ref x); // Ask Foo to deal directly with x Console.WriteLine (x); // x is now 9 42 | Chapter 2: C# Language Basics } } Now assigning p a new value changes the contents of x. Notice how the ref modifier is required both when writing and when calling the method.4 This makes it very clear what’s going on. The ref modifier is essential in implementing a swap method (later, in “Gener- ics” on page 106 in Chapter 3, we will show how to write a swap method that works with any type): class Test { static void Swap (ref string a, ref string b) { string temp = a; a = b; b = temp; } static void Main() { string x = "Penn"; string y = "Teller"; Swap (ref x, ref y); Console.WriteLine (x); // Teller Console.WriteLine (y); // Penn } } A parameter can be passed by reference or by value, regardless of whether the parameter type is a reference type or a value type. The out modifier An out argument is like a ref argument, except it: • Need not be assigned before going into the function • Must be assigned before it comes out of the function The out modifier is most commonly used to get multiple return values back from a method. For example: class Test { static void Split (string name, out string firstNames, out string lastName) { int i = name.LastIndexOf (' '); firstNames = name.Substring (0, i); 4. An exception to this rule is when calling COM methods. We discuss this in Chapter 25. Variables and Parameters | 43 C# Basics lastName = name.Substring (i + 1); } static void Main() { string a, b; Split ("Stevie Ray Vaughan", out a, out b); Console.WriteLine (a); // Stevie Ray Console.WriteLine (b); // Vaughan } } Like a ref parameter, an out parameter is passed by reference. Implications of passing by reference When you pass an argument by reference, you alias the storage location of an existing variable rather than create a new storage location. In the following example, the variables x and y represent the same instance: class Test { static int x; static void Main() { Foo (out x); } static void Foo (out int y) { Console.WriteLine (x); // x is 0 y = 1; // Mutate y Console.WriteLine (x); // x is 1 } } The params modifier The params parameter modifier may be specified on the last parameter of a method so that the method accepts any number of parameters of a particular type. The parameter type must be declared as an array. For example: class Test { static int Sum (params int[] ints) { int sum = 0; for (int i = 0; i < ints.Length; i++) sum += ints[i]; // Increase sum by ints[i] return sum; } static void Main() { int total = Sum (1, 2, 3, 4); Console.WriteLine (total); // 10 } } 44 | Chapter 2: C# Language Basics You can also supply a params argument as an ordinary array. The first line in Main is semantically equivalent to this: int total = Sum (new int[] { 1, 2, 3, 4 } ); Optional parameters From C# 4.0, methods, constructors, and indexers (Chapter 3) can declare optional parameters. A parameter is optional if it specifies a default value in its declaration: void Foo (int x = 23) { Console.WriteLine (x); } Optional parameters may be omitted when calling the method: Foo(); // 23 The default argument of 23 is actually passed to the optional parameter x—the com- piler bakes the value 23 into the compiled code at the calling side. The preceding call to Foo is semantically identical to: Foo (23); because the compiler simply substitutes the default value of an optional parameter wherever it is used. Adding an optional parameter to a public method that’s called from another assembly requires recompilation of both assemblies—just as though the parameter were mandatory. The default value of an optional parameter must be specified by a constant expres- sion, or a parameterless constructor of a value type. Optional parameters cannot be marked with ref or out. Mandatory parameters must occur before optional parameters in both the method declaration and the method call (the exception is with params arguments, which still always come last). In the following example, the explicit value of 1 is passed to x, and the default value of 0 is passed to y: void Foo (int x = 0, int y = 0) { Console.WriteLine (x + ", " + y); } void Test() { Foo(1); // 1, 0 } To do the converse (pass a default value to x and an explicit value to y) you must combine optional parameters with named arguments. Named arguments Rather than identifying an argument by position, you can identify an argument by name. For example: void Foo (int x, int y) { Console.WriteLine (x + ", " + y); } void Test() Variables and Parameters | 45 C# Basics { Foo (x:1, y:2); // 1, 2 } Named arguments can occur in any order. The following calls to Foo are semantically identical: Foo (x:1, y:2); Foo (y:2, x:1); A subtle difference is that argument expressions are evaluated in the order in which they appear at the calling site. In general, this makes a difference only with interdependent side-effecting expressions such as the following, which writes 0, 1: int a = 0; Foo (y: ++a, x: --a); // ++a is evaluated first Of course, you would almost certainly avoid writing such code in practice! You can mix named and positional parameters: Foo (1, y:2); However, there is a restriction: positional parameters must come before named arguments. So we couldn’t call Foo like this: Foo (x:1, 2); // Compile-time error Named arguments are particularly useful in conjunction with optional parameters. For instance, consider the following method: void Bar (int a = 0, int b = 0, int c = 0, int d = 0) { ... } We can call this supplying only a value for d as follows: Bar (d:3); This is particularly useful when calling COM APIs, and is discussed in detail in Chapter 25. var—Implicitly Typed Local Variables It is often the case that you declare and initialize a variable in one step. If the compiler is able to infer the type from the initialization expression, you can use the keyword var (introduced in C# 3.0) in place of the type declaration. For example: var x = "hello"; var y = new System.Text.StringBuilder(); var z = (float)Math.PI; This is precisely equivalent to: string x = "hello"; System.Text.StringBuilder y = new System.Text.StringBuilder(); float z = (float)Math.PI; 46 | Chapter 2: C# Language Basics Because of this direct equivalence, implicitly typed variables are statically typed. For example, the following generates a compile-time error: var x = 5; x = "hello"; // Compile-time error; x is of type int var can decrease code readability in the case when you can’t deduce the type purely by looking at the variable declaration. For example: Random r = new Random(); var x = r.Next(); What type is x? In “Anonymous Types” on page 164 in Chapter 4, we will describe a scenario where the use of var is mandatory. Expressions and Operators An expression essentially denotes a value. The simplest kinds of expressions are constants and variables. Expressions can be transformed and combined using oper- ators. An operator takes one or more input operands to output a new expression. Here is an example of a constant expression: 12 We can use the * operator to combine two operands (the literal expressions 12 and 30), as follows: 12 * 30 Complex expressions can be built because an operand may itself be an expression, such as the operand (12 * 30) in the following example: 1 + (12 * 30) Operators in C# can be classed as unary, binary, or ternary—depending on the number of operands they work on (one, two, or three). The binary operators always use infix notation, where the operator is placed between the two operands. Primary Expressions Primary expressions include expressions composed of operators that are intrinsic to the basic plumbing of the language. Here is an example: Math.Log (1) This expression is composed of two primary expressions. The first expression per- forms a member-lookup (with the . operator), and the second expression performs a method call (with the () operator). Expressions and Operators | 47 C# Basics Void Expressions A void expression is an expression that has no value. For example: Console.WriteLine (1) A void expression, since it has no value, cannot be used as an operand to build more complex expressions: 1 + Console.WriteLine (1) // Compile-time error Assignment Expressions An assignment expression uses the = operator to assign the result of another expres- sion to a variable. For example: x = x * 5 An assignment expression is not a void expression—it has a value of whatever was assigned, and so can be incorporated into another expression. In the following ex- ample, the expression assigns 2 to x and 10 to y: y = 5 * (x = 2) This style of expression can be used to initialize multiple values: a = b = c = d = 0 The compound assignment operators are syntactic shortcuts that combine assign- ment with another operator. For example: x *= 2 // equivalent to x = x * 2 x <<= 1 // equivalent to x = x << 1 (A subtle exception to this rule is with events, which we describe in Chapter 4: the += and −= operators here are treated specially and map to the event’s add and remove accessors.) Operator Precedence and Associativity When an expression contains multiple operators, precedence and associativity determine the order of their evaluation. Operators with higher precedence execute before operators of lower precedence. If the operators have the same precedence, the operator’s associativity determines the order of evaluation. Precedence The following expression: 1 + 2 * 3 is evaluated as follows because * has a higher precedence than +: 1 + (2 * 3) 48 | Chapter 2: C# Language Basics Left-associative operators Binary operators (except for assignment, lambda, and null coalescing operators) are left-associative; in other words, they are evaluated from left to right. For example, the following expression: 8 / 4 / 2 is evaluated as follows due to left associativity: ( 8 / 4 ) / 2 // 1 You can insert parentheses to change the actual order of evaluation: 8 / ( 4 / 2 ) // 4 Right-associative operators The assignment operators, lambda, null coalescing, and conditional operator are right-associative; in other words, they are evaluated from right to left. Right asso- ciativity allows multiple assignments such as the following to compile: x = y = 3; This first assigns 3 to y, and then assigns the result of that expression (3) to x. Operator Table Table 2-3 lists C#’s operators in order of precedence. Operators in the same category have the same precedence. We explain user-overloadable operators in “Operator Overloading” on page 158 in Chapter 4. Table 2-3. C# operators (categories in order of precedence) Category Operator symbol Operator name Example User-overloadable Primary . Member access x.y No -> (unsafe) Pointer to struct x->y No () Function call x() No [] Array/index a[x] Via indexer ++ Post-increment x++ Yes −− Post-decrement x−− Yes new Create instance new Foo() No stackalloc Unsafe stack allocation stackal loc(10) No typeof Get type from identifier typeof(int) No checked Integral overflow check on checked(x) No unchecked Integral overflow check off unchecked(x) No default Default value default(char ) No await Await await myTask No Expressions and Operators | 49 C# Basics Category Operator symbol Operator name Example User-overloadable Unary sizeof Get size of struct sizeof(int)w No + Positive value of +x Yes − Negative value of −x Yes ! Not !x Yes - Bitwise complement -x Yes ++ Pre-increment ++x Yes −− Pre-decrement −−x Yes () Cast (int)x No * (unsafe) Value at address *x No & (unsafe) Address of value &x No Multiplicative * Multiply x * y Yes / Divide x / y Yes % Remainder x % y Yes Additive + Add x + y Yes − Subtract x − y Yes Shift << Shift left x >> 1 Yes >> Shift right x << 1 Yes Relational < Less than x < y Yes > Greater than x > y Yes <= Less than or equal to x <= y Yes >= Greater than or equal to x >= y Yes is Type is or is subclass of x is y No as Type conversion x as y No Equality == Equals x == y Yes != Not equals x != y Yes Logical And & And x & y Yes Logical Xor ^ Exclusive Or x ^ y Yes Logical Or | Or x | y Yes Conditional And && Conditional And x && y Via & Conditional Or || Conditional Or x || y Via | Null coalescing ?? Null coalescing x ?? y No Conditional ?: Conditional isTrue ? then ThisValue : elseThis Value No Assignment & Lambda = Assign x = y No *= Multiply self by x *= 2 Via * 50 | Chapter 2: C# Language Basics Category Operator symbol Operator name Example User-overloadable /= Divide self by x /= 2 Via / += Add to self x += 2 Via + −= Subtract from self x −= 2 Via − <<= Shift self left by x <<= 2 Via << >>= Shift self right by x >>= 2 Via >> &= And self by x &= 2 Via & ^= Exclusive-Or self by x ^= 2 Via ^ |= Or self by x |= 2 Via | => Lambda x => x + 1 No Statements Functions comprise statements that execute sequentially in the textual order in which they appear. A statement block is a series of statements appearing between braces (the {} tokens). Declaration Statements A declaration statement declares a new variable, optionally initializing the variable with an expression. A declaration statement ends in a semicolon. You may declare multiple variables of the same type in a comma-separated list. For example: string someWord = "rosebud"; int someNumber = 42; bool rich = true, famous = false; A constant declaration is like a variable declaration, except that it cannot be changed after it has been declared, and the initialization must occur with the declaration (see “Constants” on page 76 in Chapter 3): const double c = 2.99792458E08; c += 10; // Compile-time Error Local variables The scope of a local variable or local constant extends throughout the current block. You cannot declare another local variable with the same name in the current block or in any nested blocks. For example: static void Main() { int x; { int y; int x; // Error - x already defined } { int y; // OK - y not in scope Statements | 51 C# Basics } Console.Write (y); // Error - y is out of scope } A variable’s scope extends in both directions throughout its code block. This means that if we moved the initial declaration of x in this example to the bottom of the method, we’d get the same error. This is in contrast to C++ and is somewhat peculiar, given that it’s not legal to refer to a variable or constant before it’s declared. Expression Statements Expression statements are expressions that are also valid statements. An expression statement must either change state or call something that might change state. Changing state essentially means changing a variable. The possible expression state- ments are: • Assignment expressions (including increment and decrement expressions) • Method call expressions (both void and nonvoid) • Object instantiation expressions Here are some examples: // Declare variables with declaration statements: string s; int x, y; System.Text.StringBuilder sb; // Expression statements x = 1 + 2; // Assignment expression x++; // Increment expression y = Math.Max (x, 5); // Assignment expression Console.WriteLine (y); // Method call expression sb = new StringBuilder(); // Assignment expression new StringBuilder(); // Object instantiation expression When you call a constructor or a method that returns a value, you’re not obliged to use the result. However, unless the constructor or method changes state, the state- ment is completely useless: new StringBuilder(); // Legal, but useless new string ('c', 3); // Legal, but useless x.Equals (y); // Legal, but useless Selection Statements C# has the following mechanisms to conditionally control the flow of program execution: • Selection statements (if, switch) • Conditional operator (?:) 52 | Chapter 2: C# Language Basics • Loop statements (while, do..while, for, foreach) This section covers the simplest two constructs: the if-else statement and the switch statement. The if statement An if statement executes a statement if a bool expression is true. For example: if (5 < 2 * 3) Console.WriteLine ("true"); // true The statement can be a code block: if (5 < 2 * 3) { Console.WriteLine ("true"); Console.WriteLine ("Let's move on!"); } The else clause An if statement can optionally feature an else clause: if (2 + 2 == 5) Console.WriteLine ("Does not compute"); else Console.WriteLine ("False"); // False Within an else clause, you can nest another if statement: if (2 + 2 == 5) Console.WriteLine ("Does not compute"); else if (2 + 2 == 4) Console.WriteLine ("Computes"); // Computes Changing the flow of execution with braces An else clause always applies to the immediately preceding if statement in the statement block. For example: if (true) if (false) Console.WriteLine(); else Console.WriteLine ("executes"); This is semantically identical to: if (true) { if (false) Console.WriteLine(); else Console.WriteLine ("executes"); } Statements | 53 C# Basics We can change the execution flow by moving the braces: if (true) { if (false) Console.WriteLine(); } else Console.WriteLine ("does not execute"); With braces, you explicitly state your intention. This can improve the readability of nested if statements—even when not required by the compiler. A notable exception is with the following pattern: static void TellMeWhatICanDo (int age) { if (age >= 35) Console.WriteLine ("You can be president!"); else if (age >= 21) Console.WriteLine ("You can drink!"); else if (age >= 18) Console.WriteLine ("You can vote!"); else Console.WriteLine ("You can wait!"); } Here, we’ve arranged the if and else statements to mimic the “elsif” construct of other languages (and C#’s #elif preprocessor directive). Visual Studio’s auto- formatting recognizes this pattern and preserves the indentation. Semantically, though, each if statement following an else statement is functionally nested within the else clause. The switch statement switch statements let you branch program execution based on a selection of possible values that a variable may have. switch statements may result in cleaner code than multiple if statements, since switch statements require an expression to be evaluated only once. For instance: static void ShowCard(int cardNumber) { switch (cardNumber) { case 13: Console.WriteLine ("King"); break; case 12: Console.WriteLine ("Queen"); break; case 11: Console.WriteLine ("Jack"); break; case −1: // Joker is −1 goto case 12; // In this game joker counts as queen default: // Executes for any other cardNumber Console.WriteLine (cardNumber); 54 | Chapter 2: C# Language Basics break; } } You can only switch on an expression of a type that can be statically evaluated, which restricts it to the built-in integral types, bool, and enum types (and nullable versions of these—see Chapter 4), and string type. At the end of each case clause, you must say explicitly where execution is to go next, with some kind of jump statement. Here are the options: • break (jumps to the end of the switch statement) • goto case x (jumps to another case clause) • goto default (jumps to the default clause) • Any other jump statement—namely, return, throw, continue, or goto label When more than one value should execute the same code, you can list the common cases sequentially: switch (cardNumber) { case 13: case 12: case 11: Console.WriteLine ("Face card"); break; default: Console.WriteLine ("Plain card"); break; } This feature of a switch statement can be pivotal in terms of producing cleaner code than multiple if-else statements. Iteration Statements C# enables a sequence of statements to execute repeatedly with the while, do- while, for, and foreach statements. while and do-while loops while loops repeatedly execute a body of code while a bool expression is true. The expression is tested before the body of the loop is executed. For example: int i = 0; while (i < 3) { Console.WriteLine (i); i++; } OUTPUT: 0 1 2 Statements | 55 C# Basics do-while loops differ in functionality from while loops only in that they test the expression after the statement block has executed (ensuring that the block is always executed at least once). Here’s the preceding example rewritten with a do-while loop: int i = 0; do { Console.WriteLine (i); i++; } while (i < 3); for loops for loops are like while loops with special clauses for initialization and iteration of a loop variable. A for loop contains three clauses as follows: for (initialization-clause; condition-clause; iteration-clause) statement-or-statement-block Initialization clause Executed before the loop begins; used to initialize one or more iteration variables Condition clause The bool expression that, while true, will execute the body Iteration clause Executed after each iteration of the statement block; used typically to update the iteration variable For example, the following prints the numbers 0 through 2: for (int i = 0; i < 3; i++) Console.WriteLine (i); The following prints the first 10 Fibonacci numbers (where each number is the sum of the previous two): for (int i = 0, prevFib = 1, curFib = 1; i < 10; i++) { Console.WriteLine (prevFib); int newFib = prevFib + curFib; prevFib = curFib; curFib = newFib; } Any of the three parts of the for statement may be omitted. One can implement an infinite loop such as the following (though while(true) may be used instead): for (;;) Console.WriteLine ("interrupt me"); foreach loops The foreach statement iterates over each element in an enumerable object. Most of the types in C# and the .NET Framework that represent a set or list of elements are enumerable. For example, both an array and a string are enumerable. 56 | Chapter 2: C# Language Basics Here is an example of enumerating over the characters in a string, from the first character through to the last: foreach (char c in "beer") // c is the iteration variable Console.WriteLine (c); OUTPUT: b e e r We define enumerable objects in “Enumeration and Iterators” on page 148 in Chapter 4. Jump Statements The C# jump statements are break, continue, goto, return, and throw. Jump statements obey the reliability rules of try statements (see “try Statements and Exceptions” on page 140 in Chapter 4). This means that: • A jump out of a try block always executes the try’s finally block before reaching the target of the jump. • A jump cannot be made from the inside to the outside of a finally block (except via throw). The break statement The break statement ends the execution of the body of an iteration or switch statement: int x = 0; while (true) { if (x++ > 5) break ; // break from the loop } // execution continues here after break ... The continue statement The continue statement forgoes the remaining statements in a loop and makes an early start on the next iteration. The following loop skips even numbers: for (int i = 0; i < 10; i++) { if ((i % 2) == 0) // If i is even, continue; // continue with next iteration Console.Write (i + " "); } Statements | 57 C# Basics OUTPUT: 1 3 5 7 9 The goto statement The goto statement transfers execution to another label within a statement block. The form is as follows: goto statement-label; Or, when used within a switch statement: goto case case-constant; A label is a placeholder in a code block that precedes a statement, denoted with a colon suffix. The following iterates the numbers 1 through 5, mimicking a for loop: int i = 1; startLoop: if (i <= 5) { Console.Write (i + " "); i++; goto startLoop; } OUTPUT: 1 2 3 4 5 The goto case case-constant transfers execution to another case in a switch block (see “The switch statement” on page 54). The return statement The return statement exits the method and must return an expression of the method’s return type if the method is nonvoid: static decimal AsPercentage (decimal d) { decimal p = d * 100m; return p; // Return to the calling method with value } A return statement can appear anywhere in a method (except in a finally block). The throw statement The throw statement throws an exception to indicate an error has occurred (see “try Statements and Exceptions” on page 140 in Chapter 4): if (w == null) throw new ArgumentNullException (...); Miscellaneous Statements The using statement provides an elegant syntax for calling Dispose on objects that implement IDisposable, within a finally block (see “try Statements and Excep- 58 | Chapter 2: C# Language Basics tions” on page 140 in Chapter 4 and “IDisposable, Dispose, and Close” on page 485 in Chapter 12). C# overloads the using keyword to have independent meanings in different contexts. Specifically, the using directive is different from the using statement. The lock statement is a shortcut for calling the Enter and Exit methods of the Monitor class (see Chapters 14 and 23). Namespaces A namespace is a domain for type names. Types are typically organized into hier- archical namespaces, making them easier to find and avoiding conflicts. For exam- ple, the RSA type that handles public key encryption is defined within the following namespace: System.Security.Cryptography A namespace forms an integral part of a type’s name. The following code calls RSA’s Create method: System.Security.Cryptography.RSA rsa = System.Security.Cryptography.RSA.Create(); Namespaces are independent of assemblies, which are units of deployment such as an .exe or .dll (described in Chapter 18). Namespaces also have no impact on member visibility— public, internal, private, and so on. The namespace keyword defines a namespace for types within that block. For example: namespace Outer.Middle.Inner { class Class1 {} class Class2 {} } The dots in the namespace indicate a hierarchy of nested namespaces. The code that follows is semantically identical to the preceding example: namespace Outer { namespace Middle { namespace Inner { class Class1 {} class Class2 {} Namespaces | 59 C# Basics } } } You can refer to a type with its fully qualified name, which includes all namespaces from the outermost to the innermost. For example, we could refer to Class1 in the preceding example as Outer.Middle.Inner.Class1. Types not defined in any namespace are said to reside in the global namespace. The global namespace also includes top-level namespaces, such as Outer in our example. The using Directive The using directive imports a namespace, allowing you to refer to types without their fully qualified names. The following imports the previous example’s Outer.Mid dle.Inner namespace: using Outer.Middle.Inner; class Test { static void Main() { Class1 c; // Don't need fully qualified name } } It’s legal (and often desirable) to define the same type name in different namespaces. However, you’d typically do so only if it was unlikely for a consumer to want to import both namespaces at once. A good example, from the .NET Framework, is the TextBox class which is defined both in System.Windows.Con trols (WPF) and System.Web.UI.WebControls (ASP.NET). Rules Within a Namespace Name scoping Names declared in outer namespaces can be used unqualified within inner name- spaces. In this example, the names Middle and Class1 are implicitly imported into Inner: namespace Outer { namespace Middle { class Class1 {} namespace Inner { class Class2 : Class1 {} } 60 | Chapter 2: C# Language Basics } } If you want to refer to a type in a different branch of your namespace hierarchy, you can use a partially qualified name. In the following example, we base SalesReport on Common.ReportBase: namespace MyTradingCompany { namespace Common { class ReportBase {} } namespace ManagementReporting { class SalesReport : Common.ReportBase {} } } Name hiding If the same type name appears in both an inner and an outer namespace, the inner name wins. To refer to the type in the outer namespace, you must qualify its name. For example: namespace Outer { class Foo { } namespace Inner { class Foo { } class Test { Foo f1; // = Outer.Inner.Foo Outer.Foo f2; // = Outer.Foo } } } All type names are converted to fully qualified names at compile time. Intermediate Language (IL) code contains no unqualified or partially qualified names. Repeated namespaces You can repeat a namespace declaration, as long as the type names within the name- spaces don’t conflict: namespace Outer.Middle.Inner { class Class1 {} } Namespaces | 61 C# Basics namespace Outer.Middle.Inner { class Class2 {} } We can even break the example into two source files such that we could compile each class into a different assembly. Source file 1: namespace Outer.Middle.Inner { class Class1 {} } Source file 2: namespace Outer.Middle.Inner { class Class2 {} } Nested using directive You can nest a using directive within a namespace. This allows you to scope the using directive within a namespace declaration. In the following example, Class1 is visible in one scope, but not in another: namespace N1 { class Class1 {} } namespace N2 { using N1; class Class2 : Class1 {} } namespace N2 { class Class3 : Class1 {} // Compile-time error } Aliasing Types and Namespaces Importing a namespace can result in type-name collision. Rather than importing the whole namespace, you can import just the specific types you need, giving each type an alias. For example: using PropertyInfo2 = System.Reflection.PropertyInfo; class Program { PropertyInfo2 p; } 62 | Chapter 2: C# Language Basics An entire namespace can be aliased, as follows: using R = System.Reflection; class Program { R.PropertyInfo p; } Advanced Namespace Features Extern Extern aliases allow your program to reference two types with the same fully quali- fied name (i.e., the namespace and type name are identical). This is an unusual scenario and can occur only when the two types come from different assemblies. Consider the following example. Library 1: // csc target:library /out:Widgets1.dll widgetsv1.cs namespace Widgets { public class Widget {} } Library 2: // csc target:library /out:Widgets2.dll widgetsv2.cs namespace Widgets { public class Widget {} } Application: // csc /r:Widgets1.dll /r:Widgets2.dll application.cs using Widgets; class Test { static void Main() { Widget w = new Widget(); } } The application cannot compile, because Widget is ambiguous. Extern aliases can resolve the ambiguity in our application: // csc /r:W1=Widgets1.dll /r:W2=Widgets2.dll application.cs extern alias W1; extern alias W2; class Test { static void Main() Namespaces | 63 C# Basics { W1.Widgets.Widget w1 = new W1.Widgets.Widget(); W2.Widgets.Widget w2 = new W2.Widgets.Widget(); } } Namespace alias qualifiers As we mentioned earlier, names in inner namespaces hide names in outer name- spaces. However, sometimes even the use of a fully qualified type name does not resolve the conflict. Consider the following example: namespace N { class A { public class B {} // Nested type static void Main() { new A.B(); } // Instantiate class B } } namespace A { class B {} } The Main method could be instantiating either the nested class B, or the class B within the namespace A. The compiler always gives higher precedence to identifiers in the current namespace; in this case, the nested B class. To resolve such conflicts, a namespace name can be qualified, relative to one of the following: • The global namespace—the root of all namespaces (identified with the con- textual keyword global) • The set of extern aliases The :: token is used for namespace alias qualification. In this example, we qualify using the global namespace (this is most commonly seen in auto-generated code to avoid name conflicts). namespace N { class A { static void Main() { System.Console.WriteLine (new A.B()); System.Console.WriteLine (new global::A.B()); } public class B {} } } namespace A 64 | Chapter 2: C# Language Basics { class B {} } Here is an example of qualifying with an alias (adapted from the example in “Ex- tern” on page 63): extern alias W1; extern alias W2; class Test { static void Main() { W1::Widgets.Widget w1 = new W1::Widgets.Widget(); W2::Widgets.Widget w2 = new W2::Widgets.Widget(); } } Namespaces | 65 C# Basics 3Creating Types in C# In this chapter, we will delve into types and type members. Classes A class is the most common kind of reference type. The simplest possible class dec- laration is as follows: class YourClassName { } A more complex class optionally has the following: Preceding the keyword class Attributes and class modifiers. The non-nested class modifiers are public, inter nal, abstract, sealed, static, unsafe, and partial Following YourClassName Generic type parameters, a base class, and interfaces Within the braces Class members (these are methods, properties, indexers, events, fields, constructors, overloaded operators, nested types, and a finalizer) This chapter covers all of these constructs except attributes, operator functions, and the unsafe keyword, which are covered in Chapter 4. The following sections enu- merate each of the class members. Fields A field is a variable that is a member of a class or struct. For example: class Octopus { string name; public int Age = 10; } 67 Fields allow the following modifiers: Static modifier static Access modifiers public internal private protected Inheritance modifier new Unsafe code modifier unsafe Read-only modifier readonly Threading modifier volatile The readonly modifier The readonly modifier prevents a field from being modified after construction. A read-only field can be assigned only in its declaration or within the enclosing type’s constructor. Field initialization Field initialization is optional. An uninitialized field has a default value (0, \0, null, false). Field initializers run before constructors: public int Age = 10; Declaring multiple fields together For convenience, you may declare multiple fields of the same type in a comma- separated list. This is a convenient way for all the fields to share the same attributes and field modifiers. For example: static readonly int legs = 8, eyes = 2; Methods A method performs an action in a series of statements. A method can receive input data from the caller by specifying parameters and output data back to the caller by specifying a return type. A method can specify a void return type, indicating that it doesn’t return any value to its caller. A method can also output data back to the caller via ref/out parameters. A method’s signature must be unique within the type. A method’s signature com- prises its name and parameter types (but not the parameter names, nor the return type). Methods allow the following modifiers: Static modifier static Access modifiers public internal private protected Inheritance modifiers new virtual abstract override sealed Partial method modifier partial Unmanaged code modifiers unsafe extern 68 | Chapter 3: Creating Types in C# Overloading methods A type may overload methods (have multiple methods with the same name), as long as the signatures are different. For example, the following methods can all coexist in the same type: void Foo (int x) {...} void Foo (double x) {...} void Foo (int x, float y) {...} void Foo (float x, int y) {...} However, the following pairs of methods cannot coexist in the same type, since the return type and the params modifier are not part of a method’s signature: void Foo (int x) {...} float Foo (int x) {...} // Compile-time error void Goo (int[] x) {...} void Goo (params int[] x) {...} // Compile-time error Pass-by-value versus pass-by-reference Whether a parameter is pass-by-value or pass-by-reference is also part of the signa- ture. For example, Foo(int) can coexist with either Foo(ref int) or Foo(out int). However, Foo(ref int) and Foo(out int) cannot coexist: void Foo (int x) {...} void Foo (ref int x) {...} // OK so far void Foo (out int x) {...} // Compile-time error Instance Constructors Constructors run initialization code on a class or struct. A constructor is defined like a method, except that the method name and return type are reduced to the name of the enclosing type: public class Panda { string name; // Define field public Panda (string n) // Define constructor { name = n; // Initialization code (set up field) } } ... Panda p = new Panda ("Petey"); // Call constructor Instance constructors allow the following modifiers: Access modifiers public internal private protected Unmanaged code modifiers unsafe extern Classes | 69 Creating Types Overloading constructors A class or struct may overload constructors. To avoid code duplication, one con- structor may call another, using the this keyword: using System; public class Wine { public decimal Price; public int Year; public Wine (decimal price) { Price = price; } public Wine (decimal price, int year) : this (price) { Year = year; } } When one constructor calls another, the called constructor executes first. You can pass an expression into another constructor as follows: public Wine (decimal price, DateTime year) : this (price, year.Year) { } The expression itself cannot make use of the this reference, for example, to call an instance method. (This is enforced because the object has not been initialized by the constructor at this stage, so any methods that you call on it are likely to fail.) It can, however, call static methods. Implicit parameterless constructors For classes, the C# compiler automatically generates a parameterless public con- structor if and only if you do not define any constructors. However, as soon as you define at least one constructor, the parameterless constructor is no longer automat- ically generated. For structs, a parameterless constructor is intrinsic to the struct; therefore, you can- not define your own. The role of a struct’s implicit parameterless constructor is to initialize each field with default values. Constructor and field initialization order We saw previously that fields can be initialized with default values in their declaration: class Player { int shields = 50; // Initialized first int health = 100; // Initialized second } Field initializations occur before the constructor is executed, and in the declaration order of the fields. Nonpublic constructors Constructors do not need to be public. A common reason to have a nonpublic con- structor is to control instance creation via a static method call. The static method 70 | Chapter 3: Creating Types in C# could be used to return an object from a pool rather than necessarily creating a new object, or return various subclasses based on input arguments: public class Class1 { Class1() {} // Private constructor public static Class1 Create (...) { // Perform custom logic here to return an instance of Class1 ... } } Object Initializers To simplify object initialization, any accessible fields or properties of an object can be set via an object initializer directly after construction. For example, consider the following class: public class Bunny { public string Name; public bool LikesCarrots; public bool LikesHumans; public Bunny () {} public Bunny (string n) { Name = n; } } Using object initializers, you can instantiate Bunny objects as follows: // Note parameterless constructors can omit empty parentheses Bunny b1 = new Bunny { Name="Bo", LikesCarrots=true, LikesHumans=false }; Bunny b2 = new Bunny ("Bo") { LikesCarrots=true, LikesHumans=false }; The code to construct b1 and b2 is precisely equivalent to: Bunny temp1 = new Bunny(); // temp1 is a compiler-generated name temp1.Name = "Bo"; temp1.LikesCarrots = true; temp1.LikesHumans = false; Bunny b1 = temp1; Bunny temp2 = new Bunny ("Bo"); temp2.LikesCarrots = true; temp2.LikesHumans = false; Bunny b2 = temp2; The temporary variables are to ensure that if an exception is thrown during initial- ization, you can’t end up with a half-initialized object. Object initializers were introduced in C# 3.0. Classes | 71 Creating Types Object Initializers Versus Optional Parameters Instead of using object initializers, we could make Bunny’s constructor accept optional parameters: public Bunny (string name, bool likesCarrots = false, bool likesHumans = false) { Name = name; LikesCarrots = likesCarrots; LikesHumans = likesHumans; } This would allow us to construct a Bunny as follows: Bunny b1 = new Bunny (name: "Bo", likesCarrots: true); An advantage of this approach is that we could make Bunny’s fields (or proper- ties, as we’ll explain shortly) read-only if we choose. Making fields or properties read-only is good practice when there’s no valid reason for them to change throughout the life of the object. The disadvantage in this approach is that each optional parameter value is baked into the calling site. In other words, C# translates our constructor call into this: Bunny b1 = new Bunny ("Bo", true, false); This can be problematic if we instantiate the Bunny class from another assembly, and later modify Bunny by adding another optional parameter—such as likes Cats. Unless the referencing assembly is also recompiled, it will continue to call the (now nonexistent) constructor with three parameters and fail at runtime. (A subtler problem is that if we changed the value of one of the optional parameters, callers in other assemblies would continue to use the old optional value until they were recompiled.) Hence, optional parameters are best avoided in public functions if you want to offer binary compatibility between assembly versions. The this Reference The this reference refers to the instance itself. In the following example, the Marry method uses this to set the partner’s mate field: public class Panda { public Panda Mate; public void Marry (Panda partner) { Mate = partner; partner.Mate = this; } } 72 | Chapter 3: Creating Types in C# The this reference also disambiguates a local variable or parameter from a field. For example: public class Test { string name; public Test (string name) { this.name = name; } } The this reference is valid only within nonstatic members of a class or struct. Properties Properties look like fields from the outside, but internally they contain logic, like methods do. For example, you can’t tell by looking at the following code whether CurrentPrice is a field or a property: Stock msft = new Stock(); msft.CurrentPrice = 30; msft.CurrentPrice -= 3; Console.WriteLine (msft.CurrentPrice); A property is declared like a field, but with a get/set block added. Here’s how to implement CurrentPrice as a property: public class Stock { decimal currentPrice; // The private "backing" field public decimal CurrentPrice // The public property { get { return currentPrice; } set { currentPrice = value; } } } get and set denote property accessors. The get accessor runs when the property is read. It must return a value of the property’s type. The set accessor runs when the property is assigned. It has an implicit parameter named value of the property’s type that you typically assign to a private field (in this case, currentPrice). Although properties are accessed in the same way as fields, they differ in that they give the implementer complete control over getting and setting its value. This control enables the implementer to choose whatever internal representation is needed, without exposing the internal details to the user of the property. In this example, the set method could throw an exception if value was outside a valid range of values. Throughout this book, we use public fields extensively to keep the examples free of distraction. In a real application, you would typically favor public properties over public fields, in order to promote encapsulation. Classes | 73 Creating Types Properties allow the following modifiers: Static modifier static Access modifiers public internal private protected Inheritance modifiers new virtual abstract override sealed Unmanaged code modifiers unsafe extern Read-only and calculated properties A property is read-only if it specifies only a get accessor, and it is write-only if it specifies only a set accessor. Write-only properties are rarely used. A property typically has a dedicated backing field to store the underlying data. However, a property can also be computed from other data. For example: decimal currentPrice, sharesOwned; public decimal Worth { get { return currentPrice * sharesOwned; } } Automatic properties The most common implementation for a property is a getter and/or setter that simply reads and writes to a private field of the same type as the property. An automatic property declaration instructs the compiler to provide this implementation. We can redeclare the first example in this section as follows: public class Stock { ... public decimal CurrentPrice { get; set; } } The compiler automatically generates a private backing field of a compiler-generated name that cannot be referred to. The set accessor can be marked private if you want to expose the property as read-only to other types. Automatic properties were in- troduced in C# 3.0. get and set accessibility The get and set accessors can have different access levels. The typical use case for this is to have a public property with an internal or private access modifier on the setter: public class Foo { private decimal x; public decimal X { get { return x; } private set { x = Math.Round (value, 2); } 74 | Chapter 3: Creating Types in C# } } Notice that you declare the property itself with the more permissive access level (public, in this case), and add the modifier to the accessor you want to be less accessible. CLR property implementation C# property accessors internally compile to methods called get_XXX and set_XXX: public decimal get_CurrentPrice {...} public void set_CurrentPrice (decimal value) {...} Simple nonvirtual property accessors are inlined by the JIT (Just-In-Time) compiler, eliminating any performance difference between accessing a property and a field. Inlining is an optimization in which a method call is replaced with the body of that method. With WinRT properties, the compiler assumes the put_XXX naming convention rather than set_XXX. Indexers Indexers provide a natural syntax for accessing elements in a class or struct that encapsulate a list or dictionary of values. Indexers are similar to properties, but are accessed via an index argument rather than a property name. The string class has an indexer that lets you access each of its char values via an int index: string s = "hello"; Console.WriteLine (s[0]); // 'h' Console.WriteLine (s[3]); // 'l' The syntax for using indexers is like that for using arrays, except that the index argument(s) can be of any type(s). Indexers have the same modifiers as properties (see “Proper- ties” on page 73). Implementing an indexer To write an indexer, define a property called this, specifying the arguments in square brackets. For instance: class Sentence { string[] words = "The quick brown fox".Split(); public string this [int wordNum] // indexer { get { return words [wordNum]; } set { words [wordNum] = value; } Classes | 75 Creating Types } } Here’s how we could use this indexer: Sentence s = new Sentence(); Console.WriteLine (s[3]); // fox s[3] = "kangaroo"; Console.WriteLine (s[3]); // kangaroo A type may declare multiple indexers, each with parameters of different types. An indexer can also take more than one parameter: public string this [int arg1, string arg2] { get { ... } set { ... } } If you omit the set accessor, an indexer becomes read-only. CLR indexer implementation Indexers internally compile to methods called get_Item and set_Item, as follows: public string get_Item (int wordNum) {...} public void set_Item (int wordNum, string value) {...} Constants A constant is a static field whose value can never change. A constant is evaluated statically at compile time and the compiler literally substitutes its value whenever used (rather like a macro in C++). A constant can be any of the built-in numeric types, bool, char, string, or an enum type. A constant is declared with the const keyword and must be initialized with a value. For example: public class Test { public const string Message = "Hello World"; } A constant is much more restrictive than a static readonly field—both in the types you can use and in field initialization semantics. A constant also differs from a static readonly field in that the evaluation of the constant occurs at compile time. For example: public static double Circumference (double radius) { return 2 * System.Math.PI * radius; } is compiled to: public static double Circumference (double radius) { return 6.2831853071795862 * radius; } 76 | Chapter 3: Creating Types in C# It makes sense for PI to be a constant, since it can never change. In contrast, a static readonly field can have a different value per application. A static readonly field is also advantageous when exposing to other assemblies a value that might change in a later version. For instance, suppose assembly X exposes a constant as follows: public const decimal ProgramVersion = 2.3; If assembly Y references X and uses this constant, the value 2.3 will be baked into assembly Y when compiled. This means that if X is later recompiled with the constant set to 2.4, Y will still use the old value of 2.3 until Y is recompiled. A static readonly field avoids this problem. Another way of looking at this is that any value that might change in the future is not constant by definition, and so should not be represented as one. Constants can also be declared local to a method. For example: static void Main() { const double twoPI = 2 * System.Math.PI; ... } Non-local constants allow the following modifiers: Access modifiers public internal private protected Inheritance modifier new Static Constructors A static constructor executes once per type, rather than once per instance. A type can define only one static constructor, and it must be parameterless and have the same name as the type: class Test { static Test() { Console.WriteLine ("Type Initialized"); } } The runtime automatically invokes a static constructor just prior to the type being used. Two things trigger this: • Instantiating the type • Accessing a static member in the type The only modifiers allowed by static constructors are unsafe and extern. Classes | 77 Creating Types If a static constructor throws an unhandled exception (Chap- ter 4), that type becomes unusable for the life of the application. Static constructors and field initialization order Static field initializers run just before the static constructor is called. If a type has no static constructor, field initializers will execute just prior to the type being used— or anytime earlier at the whim of the runtime. (This means that the presence of a static constructor may cause field initializers to execute later in the program than they would otherwise.) Static field initializers run in the order in which the fields are declared. The following example illustrates this: X is initialized to 0 and Y is initialized to 3. class Foo { public static int X = Y; // 0 public static int Y = 3; // 3 } If we swap the two field initializers around, both fields are initialized to 3. The next example prints 0 followed by 3 because the field initializer that instantiates a Foo executes before X is initialized to 3: class Program { static void Main() { Console.WriteLine (Foo.X); } // 3 } class Foo { public static Foo Instance = new Foo(); public static int X = 3; Foo() { Console.WriteLine (X); } // 0 } If we swap the two lines in boldface, the example prints 3 followed by 3. Static Classes A class can be marked static, indicating that it must be composed solely of static members and cannot be subclassed. The System.Console and System.Math classes are good examples of static classes. Finalizers Finalizers are class-only methods that execute before the garbage collector reclaims the memory for an unreferenced object. The syntax for a finalizer is the name of the class prefixed with the ~ symbol: class Class1 { 78 | Chapter 3: Creating Types in C# ~Class1() { ... } } This is actually C# syntax for overriding Object’s Finalize method, and the compiler expands it into the following method declaration: protected override void Finalize() { ... base.Finalize(); } We discuss garbage collection and finalizers fully in Chapter 12. Finalizers allow the following modifier: Unmanaged code modifier unsafe Partial Types and Methods Partial types allow a type definition to be split—typically across multiple files. A common scenario is for a partial class to be auto-generated from some other source (such as a Visual Studio template or designer), and for that class to be augmented with additional hand-authored methods. For example: // PaymentFormGen.cs - auto-generated partial class PaymentForm { ... } // PaymentForm.cs - hand-authored partial class PaymentForm { ... } Each participant must have the partial declaration; the following is illegal: partial class PaymentForm {} class PaymentForm {} Participants cannot have conflicting members. A constructor with the same param- eters, for instance, cannot be repeated. Partial types are resolved entirely by the compiler, which means that each participant must be available at compile time and must reside in the same assembly. There are two ways to specify a base class with partial classes: • Specify the (same) base class on each participant. For example: partial class PaymentForm : ModalForm {} partial class PaymentForm : ModalForm {} • Specify the base class on just one participant. For example: partial class PaymentForm : ModalForm {} partial class PaymentForm {} Classes | 79 Creating Types In addition, each participant can independently specify interfaces to implement. We cover base classes and interfaces in “Inheritance” on page 80 and “Inter- faces” on page 96. Partial methods A partial type may contain partial methods. These let an auto-generated partial type provide customizable hooks for manual authoring. For example: partial class PaymentForm // In auto-generated file { ... partial void ValidatePayment (decimal amount); } partial class PaymentForm // In hand-authored file { ... partial void ValidatePayment (decimal amount) { if (amount > 100) ... } } A partial method consists of two parts: a definition and an implementation. The definition is typically written by a code generator, and the implementation is typi- cally manually authored. If an implementation is not provided, the definition of the partial method is compiled away (as is the code that calls it). This allows auto- generated code to be liberal in providing hooks, without having to worry about bloat. Partial methods must be void and are implicitly private. Partial methods were introduced in C# 3.0. Inheritance A class can inherit from another class to extend or customize the original class. Inheriting from a class lets you reuse the functionality in that class instead of building it from scratch. A class can inherit from only a single class, but can itself be inherited by many classes, thus forming a class hierarchy. In this example, we start by defining a class called Asset: public class Asset { public string Name; } Next, we define classes called Stock and House, which will inherit from Asset. Stock and House get everything an Asset has, plus any additional members that they define: public class Stock : Asset // inherits from Asset { public long SharesOwned; 80 | Chapter 3: Creating Types in C# } public class House : Asset // inherits from Asset { public decimal Mortgage; } Here’s how we can use these classes: Stock msft = new Stock { Name="MSFT", SharesOwned=1000 }; Console.WriteLine (msft.Name); // MSFT Console.WriteLine (msft.SharesOwned); // 1000 House mansion = new House { Name="Mansion", Mortgage=250000 }; Console.WriteLine (mansion.Name); // Mansion Console.WriteLine (mansion.Mortgage); // 250000 The derived classes, Stock and House, inherit the Name property from the base class, Asset. A derived class is also called a subclass. A base class is also called a superclass. Polymorphism References are polymorphic. This means a variable of type x can refer to an object that subclasses x. For instance, consider the following method: public static void Display (Asset asset) { System.Console.WriteLine (asset.Name); } This method can display both a Stock and a House, since they are both Assets: Stock msft = new Stock ... ; House mansion = new House ... ; Display (msft); Display (mansion); Polymorphism works on the basis that subclasses (Stock and House) have all the features of their base class (Asset). The converse, however, is not true. If Display was modified to accept a House, you could not pass in an Asset: static void Main() { Display (new Asset()); } // Compile-time error public static void Display (House house) // Will not accept Asset { Inheritance | 81 Creating Types System.Console.WriteLine (house.Mortgage); } Casting and Reference Conversions An object reference can be: • Implicitly upcast to a base class reference • Explicitly downcast to a subclass reference Upcasting and downcasting between compatible reference types performs reference conversions: a new reference is (logically) created that points to the same object. An upcast always succeeds; a downcast succeeds only if the object is suitably typed. Upcasting An upcast operation creates a base class reference from a subclass reference. For example: Stock msft = new Stock(); Asset a = msft; // Upcast After the upcast, variable a still references the same Stock object as variable msft. The object being referenced is not itself altered or converted: Console.WriteLine (a == msft); // True Although a and msft refer to the identical object, a has a more restrictive view on that object: Console.WriteLine (a.Name); // OK Console.WriteLine (a.SharesOwned); // Error: SharesOwned undefined The last line generates a compile-time error because the variable a is of type Asset, even though it refers to an object of type Stock. To get to its SharesOwned field, you must downcast the Asset to a Stock. Downcasting A downcast operation creates a subclass reference from a base class reference. For example: Stock msft = new Stock(); Asset a = msft; // Upcast Stock s = (Stock)a; // Downcast Console.WriteLine (s.SharesOwned); // Console.WriteLine (s == a); // True Console.WriteLine (s == msft); // True As with an upcast, only references are affected—not the underlying object. A down- cast requires an explicit cast because it can potentially fail at runtime: House h = new House(); Asset a = h; // Upcast always succeeds Stock s = (Stock)a; // Downcast fails: a is not a Stock 82 | Chapter 3: Creating Types in C# If a downcast fails, an InvalidCastException is thrown. This is an example of runtime type checking (we will elaborate on this concept in “Static and Runtime Type Check- ing” on page 91). The as operator The as operator performs a downcast that evaluates to null (rather than throwing an exception) if the downcast fails: Asset a = new Asset(); Stock s = a as Stock; // s is null; no exception thrown This is useful when you’re going to subsequently test whether the result is null: if (s != null) Console.WriteLine (s.SharesOwned); Without such a test, a cast is advantageous, because if it fails, a more helpful exception is thrown. We can illustrate by com- paring the following two lines of code: int shares = ((Stock)a).SharesOwned; // Approach #1 int shares = (a as Stock).SharesOwned; // Approach #2 If a is not a Stock, the first line throws an InvalidCastExcep tion, which is an accurate description of what went wrong. The second line throws a NullReferenceException, which is ambig- uous. Was a not a Stock or was a null? Another way of looking at it is that with the cast operator, you’re saying to the compiler: “I’m certain of a value’s type; if I’m wrong, there’s a bug in my code, so throw an exception!” Whereas with the as operator, you’re uncertain of its type and want to branch according to the outcome at runtime. The as operator cannot perform custom conversions (see “Operator Overload- ing” on page 158 in Chapter 4) and it cannot do numeric conversions: long x = 3 as long; // Compile-time error The as and cast operators will also perform upcasts, although this is not terribly useful because an implicit conversion will do the job. The is operator The is operator tests whether a reference conversion would succeed; in other words, whether an object derives from a specified class (or implements an interface). It is often used to test before downcasting. if (a is Stock) Console.WriteLine (((Stock)a).SharesOwned); Inheritance | 83 Creating Types The is operator does not consider custom or numeric conversions, but it does con- sider unboxing conversions (see “The object Type” on page 89). Virtual Function Members A function marked as virtual can be overridden by subclasses wanting to provide a specialized implementation. Methods, properties, indexers, and events can all be declared virtual: public class Asset { public string Name; public virtual decimal Liability { get { return 0; } } } A subclass overrides a virtual method by applying the override modifier: public class Stock : Asset { public long SharesOwned; } public class House : Asset { public decimal Mortgage; public override decimal Liability { get { return Mortgage; } } } By default, the Liability of an Asset is 0. A Stock does not need to specialize this behavior. However, the House specializes the Liability property to return the value of the Mortgage: House mansion = new House { Name="McMansion", Mortgage=250000 }; Asset a = mansion; Console.WriteLine (mansion.Liability); // 250000 Console.WriteLine (a.Liability); // 250000 The signatures, return types, and accessibility of the virtual and overridden methods must be identical. An overridden method can call its base class implementation via the base keyword (we will cover this in “The base Keyword” on page 86). Calling virtual methods from a constructor is potentially dan- gerous because authors of subclasses are unlikely to know, when overriding the method, that they are working with a par- tially initialized object. In other words, the overriding method may end up accessing methods or properties which rely on fields not yet initialized by the constructor. Abstract Classes and Abstract Members A class declared as abstract can never be instantiated. Instead, only its concrete subclasses can be instantiated. 84 | Chapter 3: Creating Types in C# Abstract classes are able to define abstract members. Abstract members are like vir- tual members, except they don’t provide a default implementation. That implementation must be provided by the subclass, unless that subclass is also declared abstract: public abstract class Asset { // Note empty implementation public abstract decimal NetValue { get; } } public class Stock : Asset { public long SharesOwned; public decimal CurrentPrice; // Override like a virtual method. public override decimal NetValue { get { return CurrentPrice * SharesOwned; } } } Hiding Inherited Members A base class and a subclass may define identical members. For example: public class A { public int Counter = 1; } public class B : A { public int Counter = 2; } The Counter field in class B is said to hide the Counter field in class A. Usually, this happens by accident, when a member is added to the base type after an identical member was added to the subtype. For this reason, the compiler generates a warning, and then resolves the ambiguity as follows: • References to A (at compile time) bind to A.Counter. • References to B (at compile time) bind to B.Counter. Occasionally, you want to hide a member deliberately, in which case you can apply the new modifier to the member in the subclass. The new modifier does nothing more than suppress the compiler warning that would otherwise result: public class A { public int Counter = 1; } public class B : A { public new int Counter = 2; } The new modifier communicates your intent to the compiler—and other programmers—that the duplicate member is not an accident. C# overloads the new keyword to have independent meanings in different contexts. Specifically, the new operator is different from the new member modifier. Inheritance | 85 Creating Types new versus override Consider the following class hierarchy: public class BaseClass { public virtual void Foo() { Console.WriteLine ("BaseClass.Foo"); } } public class Overrider : BaseClass { public override void Foo() { Console.WriteLine ("Overrider.Foo"); } } public class Hider : BaseClass { public new void Foo() { Console.WriteLine ("Hider.Foo"); } } The differences in behavior between Overrider and Hider are demonstrated in the following code: Overrider over = new Overrider(); BaseClass b1 = over; over.Foo(); // Overrider.Foo b1.Foo(); // Overrider.Foo Hider h = new Hider(); BaseClass b2 = h; h.Foo(); // Hider.Foo b2.Foo(); // BaseClass.Foo Sealing Functions and Classes An overridden function member may seal its implementation with the sealed key- word to prevent it from being overridden by further subclasses. In our earlier virtual function member example, we could have sealed House’s implementation of Liabil ity, preventing a class that derives from House from overriding Liability, as follows: public sealed override decimal Liability { get { return Mortgage; } } You can also seal the class itself, implicitly sealing all the virtual functions, by ap- plying the sealed modifier to the class itself. Sealing a class is more common than sealing a function member. Although you can seal against overriding, you can’t seal a member against being hidden. The base Keyword The base keyword is similar to the this keyword. It serves two essential purposes: • Accessing an overridden function member from the subclass • Calling a base-class constructor (see the next section) 86 | Chapter 3: Creating Types in C# In this example, House uses the base keyword to access Asset’s implementation of Liability: public class House : Asset { ... public override decimal Liability { get { return base.Liability + Mortgage; } } } With the base keyword, we access Asset’s Liability property nonvirtually. This means we will always access Asset’s version of this property—regardless of the in- stance’s actual runtime type. The same approach works if Liability is hidden rather than overridden. (You can also access hidden members by casting to the base class before invoking the function.) Constructors and Inheritance A subclass must declare its own constructors. The base class’s constructors are accessible to the derived class, but are never automatically inherited. For example, if we define Baseclass and Subclass as follows: public class Baseclass { public int X; public Baseclass () { } public Baseclass (int x) { this.X = x; } } public class Subclass : Baseclass { } the following is illegal: Subclass s = new Subclass (123); Subclass must hence “redefine” any constructors it wants to expose. In doing so, however, it can call any of the base class’s constructors with the base keyword: public class Subclass : Baseclass { public Subclass (int x) : base (x) { } } The base keyword works rather like the this keyword, except that it calls a con- structor in the base class. Base-class constructors always execute first; this ensures that base initialization occurs before specialized initialization. Inheritance | 87 Creating Types Implicit calling of the parameterless base-class constructor If a constructor in a subclass omits the base keyword, the base type’s parameter- less constructor is implicitly called: public class BaseClass { public int X; public BaseClass() { X = 1; } } public class Subclass : BaseClass { public Subclass() { Console.WriteLine (X); } // 1 } If the base class has no accessible parameterless constructor, subclasses are forced to use the base keyword in their constructors. Constructor and field initialization order When an object is instantiated, initialization takes place in the following order: 1. From subclass to base class: a. Fields are initialized. b. Arguments to base-class constructor calls are evaluated. 2. From base class to subclass: a. Constructor bodies execute. The following code demonstrates: public class B { int x = 1; // Executes 3rd public B (int x) { ... // Executes 4th } } public class D : B { int y = 1; // Executes 1st public D (int x) : base (x + 1) // Executes 2nd { ... // Executes 5th } } 88 | Chapter 3: Creating Types in C# Overloading and Resolution Inheritance has an interesting impact on method overloading. Consider the follow- ing two overloads: static void Foo (Asset a) { } static void Foo (House h) { } When an overload is called, the most specific type has precedence: House h = new House (...); Foo(h); // Calls Foo(House) The particular overload to call is determined statically (at compile time) rather than at runtime. The following code calls Foo(Asset), even though the runtime type of a is House: Asset a = new House (...); Foo(a); // Calls Foo(Asset) If you cast Asset to dynamic (Chapter 4), the decision as to which overload to call is deferred until runtime, and is then based on the object’s actual type: Asset a = new House (...); Foo ((dynamic)a); // Calls Foo(House) The object Type object (System.Object) is the ultimate base class for all types. Any type can be upcast to object. To illustrate how this is useful, consider a general-purpose stack. A stack is a data structure based on the principle of LIFO—“Last-In First-Out.” A stack has two operations: push an object on the stack, and pop an object off the stack. Here is a simple implementation that can hold up to 10 objects: public class Stack { int position; object[] data = new object[10]; public void Push (object obj) { data[position++] = obj; } public object Pop() { return data[--position]; } } Because Stack works with the object type, we can Push and Pop instances of any type to and from the Stack: Stack stack = new Stack(); stack.Push ("sausage"); string s = (string) stack.Pop(); // Downcast, so explicit cast is needed Console.WriteLine (s); // sausage The object Type | 89 Creating Types object is a reference type, by virtue of being a class. Despite this, value types, such as int, can also be cast to and from object, and so be added to our stack. This feature of C# is called type unification and is demonstrated here: stack.Push (3); int three = (int) stack.Pop(); When you cast between a value type and object, the CLR must perform some special work to bridge the difference in semantics between value and reference types. This process is called boxing and unboxing. In “Generics” on page 106, we’ll describe how to improve our Stack class to better handle stacks with same-typed elements. Boxing and Unboxing Boxing is the act of converting a value-type instance to a reference-type instance. The reference type may be either the object class or an interface (which we will visit later in the chapter).1 In this example, we box an int into an object: int x = 9; object obj = x; // Box the int Unboxing reverses the operation, by casting the object back to the original value type: int y = (int)obj; // Unbox the int Unboxing requires an explicit cast. The runtime checks that the stated value type matches the actual object type, and throws an InvalidCastException if the check fails. For instance, the following throws an exception, because long does not exactly match int: object obj = 9; // 9 is inferred to be of type int long x = (long) obj; // InvalidCastException The following succeeds, however: object obj = 9; long x = (int) obj; As does this: object obj = 3.5; // 3.5 is inferred to be of type double int x = (int) (double) obj; // x is now 3 In the last example, (double) performs an unboxing and then (int) performs a numeric conversion. 1. The reference type may also be System.ValueType or System.Enum (Chapter 6). 90 | Chapter 3: Creating Types in C# Boxing conversions are crucial in providing a unified type sys- tem. The system is not perfect, however: we’ll see in “Gener- ics” on page 106 that variance with arrays and generics sup- ports only reference conversions and not boxing conversions: object[] a1 = new string[3]; // Legal object[] a2 = new int[3]; // Error Copying semantics of boxing and unboxing Boxing copies the value-type instance into the new object, and unboxing copies the contents of the object back into a value-type instance. In the following example, changing the value of i doesn’t change its previously boxed copy: int i = 3; object boxed = i; i = 5; Console.WriteLine (boxed); // 3 Static and Runtime Type Checking C# programs are type-checked both statically (at compile time) and at runtime (by the CLR). Static type checking enables the compiler to verify the correctness of your program without running it. The following code will fail because the compiler enforces static typing: int x = "5"; Runtime type checking is performed by the CLR when you downcast via a reference conversion or unboxing. For example: object y = "5"; int z = (int) y; // Runtime error, downcast failed Runtime type checking is possible because each object on the heap internally stores a little type token. This token can be retrieved by calling the GetType method of object. The GetType Method and typeof Operator All types in C# are represented at runtime with an instance of System.Type. There are two basic ways to get a System.Type object: • Call GetType on the instance. • Use the typeof operator on a type name. GetType is evaluated at runtime; typeof is evaluated statically at compile time (when generic type parameters are involved, it’s resolved by the just-in-time compiler). System.Type has properties for such things as the type’s name, assembly, base type, and so on. The object Type | 91 Creating Types For example: using System; public class Point { public int X, Y; } class Test { static void Main() { Point p = new Point(); Console.WriteLine (p.GetType().Name); // Point Console.WriteLine (typeof (Point).Name); // Point Console.WriteLine (p.GetType() == typeof(Point)); // True Console.WriteLine (p.X.GetType().Name); // Int32 Console.WriteLine (p.Y.GetType().FullName); // System.Int32 } } System.Type also has methods that act as a gateway to the runtime’s reflection model, described in Chapter 19. The ToString Method The ToString method returns the default textual representation of a type instance. This method is overridden by all built-in types. Here is an example of using the int type’s ToString method: int x = 1; string s = x.ToString(); // s is "1" You can override the ToString method on custom types as follows: public class Panda { public string Name; public override string ToString() { return Name; } } ... Panda p = new Panda { Name = "Petey" }; Console.WriteLine (p); // Petey If you don’t override ToString, the method returns the type name. When you call an overridden object member such as ToString directly on a value type, boxing doesn’t occur. Boxing then oc- curs only if you cast: int x = 1; string s1 = x.ToString(); // Calling on nonboxed value object box = x; string s2 = box.ToString(); // Calling on boxed value 92 | Chapter 3: Creating Types in C# Object Member Listing Here are all the members of object: public class Object { public Object(); public extern Type GetType(); public virtual bool Equals (object obj); public static bool Equals (object objA, object objB); public static bool ReferenceEquals (object objA, object objB); public virtual int GetHashCode(); public virtual string ToString(); protected virtual void Finalize(); protected extern object MemberwiseClone(); } We describe the Equals, ReferenceEquals, and GetHashCode methods in “Equality Comparison” on page 254 in Chapter 6. Structs A struct is similar to a class, with the following key differences: • A struct is a value type, whereas a class is a reference type. • A struct does not support inheritance (other than implicitly deriving from object, or more precisely, System.ValueType). A struct can have all the members a class can, except the following: • A parameterless constructor • A finalizer • Virtual members A struct is used instead of a class when value-type semantics are desirable. Good examples of structs are numeric types, where it is more natural for assignment to copy a value rather than a reference. Because a struct is a value type, each instance does not require instantiation of an object on the heap; this incurs a useful savings when creating many instances of a type. For instance, creating an array of value type requires only a single heap allocation. Structs | 93 Creating Types Struct Construction Semantics The construction semantics of a struct are as follows: • A parameterless constructor that you can’t override implicitly exists. This per- forms a bitwise-zeroing of its fields. • When you define a struct constructor, you must explicitly assign every field. • You can’t have field initializers in a struct. Here is an example of declaring and calling struct constructors: public struct Point { int x, y; public Point (int x, int y) { this.x = x; this.y = y; } } ... Point p1 = new Point (); // p1.x and p1.y will be 0 Point p2 = new Point (1, 1); // p1.x and p1.y will be 1 The next example generates three compile-time errors: public struct Point { int x = 1; // Illegal: cannot initialize field int y; public Point() {} // Illegal: cannot have // parameterless constructor public Point (int x) {this.x = x;} // Illegal: must assign field y } Changing struct to class makes this example legal. Access Modifiers To promote encapsulation, a type or type member may limit its accessibility to other types and other assemblies by adding one of five access modifiers to the declaration: public Fully accessible. This is the implicit accessibility for members of an enum or interface. internal Accessible only within containing assembly or friend assemblies. This is the default accessibility for non-nested types. private Accessible only within containing type. This is the default accessibility for members of a class or struct. protected Accessible only within containing type or subclasses. 94 | Chapter 3: Creating Types in C# protected internal The union of protected and internal accessibility. Eric Lippert explains it as follows: Everything is as private as possible by default, and each modifier makes the thing more accessible. So something that is protected internal is made more accessible in two ways. The CLR has the concept of the intersection of protected and internal accessibility, but C# does not support this. Examples Class2 is accessible from outside its assembly; Class1 is not: class Class1 {} // Class1 is internal (default) public class Class2 {} ClassB exposes field x to other types in the same assembly; ClassA does not: class ClassA { int x; } // x is private (default) class ClassB { internal int x; } Functions within Subclass can call Bar but not Foo: class BaseClass { void Foo() {} // Foo is private (default) protected void Bar() {} } class Subclass : BaseClass { void Test1() { Foo(); } // Error - cannot access Foo void Test2() { Bar(); } // OK } Friend Assemblies In advanced scenarios, you can expose internal members to other friend assemblies by adding the System.Runtime.CompilerServices.InternalsVisibleTo assembly at- tribute, specifying the name of the friend assembly as follows: [assembly: InternalsVisibleTo ("Friend")] If the friend assembly has a strong name (see Chapter 18), you must specify its full 160-byte public key: [assembly: InternalsVisibleTo ("StrongFriend, PublicKey=0024f000048c...")] You can extract the full public key from a strongly named assembly with a LINQ query (we explain LINQ in detail in Chapter 8): string key = string.Join ("", Assembly.GetExecutingAssembly().GetName().GetPublicKey() Access Modifiers | 95 Creating Types .Select (b => b.ToString ("x2")) .ToArray()); The companion sample in LINQPad invites you to browse to an assembly and then copies the assembly’s full public key to the clipboard. Accessibility Capping A type caps the accessibility of its declared members. The most common example of capping is when you have an internal type with public members. For example: class C { public void Foo() {} } C’s (default) internal accessibility caps Foo’s accessibility, effectively making Foo internal. A common reason Foo would be marked public is to make for easier refactoring, should C later be changed to public. Restrictions on Access Modifiers When overriding a base class function, accessibility must be identical on the over- ridden function. For example: class BaseClass { protected virtual void Foo() {} } class Subclass1 : BaseClass { protected override void Foo() {} } // OK class Subclass2 : BaseClass { public override void Foo() {} } // Error (An exception is when overriding a protected internal method in another assembly, in which case the override must simply be protected.) The compiler prevents any inconsistent use of access modifiers. For example, a sub- class itself can be less accessible than a base class, but not more: internal class A {} public class B : A {} // Error Interfaces An interface is similar to a class, but it provides a specification rather than an im- plementation for its members. An interface is special in the following ways: • Interface members are all implicitly abstract. In contrast, a class can provide both abstract members and concrete members with implementations. • A class (or struct) can implement multiple interfaces. In contrast, a class can inherit from only a single class, and a struct cannot inherit at all (aside from deriving from System.ValueType). An interface declaration is like a class declaration, but it provides no implementation for its members, since all its members are implicitly abstract. These members will be implemented by the classes and structs that implement the interface. An interface 96 | Chapter 3: Creating Types in C# can contain only methods, properties, events, and indexers, which noncoinciden- tally are precisely the members of a class that can be abstract. Here is the definition of the IEnumerator interface, defined in System.Collections: public interface IEnumerator { bool MoveNext(); object Current { get; } void Reset(); } Interface members are always implicitly public and cannot declare an access modi- fier. Implementing an interface means providing a public implementation for all its members: internal class Countdown : IEnumerator { int count = 11; public bool MoveNext () { return count-- > 0 ; } public object Current { get { return count; } } public void Reset() { throw new NotSupportedException(); } } You can implicitly cast an object to any interface that it implements. For example: IEnumerator e = new Countdown(); while (e.MoveNext()) Console.Write (e.Current); // 109876543210 Even though Countdown is an internal class, its members that implement IEnumerator can be called publicly by casting an instance of Countdown to IEnumerator. For instance, if a public type in the same assembly defined a method as follows: public static class Util { public static object GetCountDown() { return new CountDown(); } } a caller from another assembly could do this: IEnumerator e = (IEnumerator) Util.GetCountDown(); e.MoveNext(); If IEnumerator was itself defined as internal, this wouldn’t be possible. Extending an Interface Interfaces may derive from other interfaces. For instance: public interface IUndoable { void Undo(); } public interface IRedoable : IUndoable { void Redo(); } Interfaces | 97 Creating Types IRedoable “inherits” all the members of IUndoable. In other words, types that im- plement IRedoable must also implement the members of IUndoable. Explicit Interface Implementation Implementing multiple interfaces can sometimes result in a collision between mem- ber signatures. You can resolve such collisions by explicitly implementing an interface member. Consider the following example: interface I1 { void Foo(); } interface I2 { int Foo(); } public class Widget : I1, I2 { public void Foo () { Console.WriteLine ("Widget's implementation of I1.Foo"); } int I2.Foo() { Console.WriteLine ("Widget's implementation of I2.Foo"); return 42; } } Because both I1 and I2 have conflicting Foo signatures, Widget explicitly implements I2’s Foo method. This lets the two methods coexist in one class. The only way to call an explicitly implemented member is to cast to its interface: Widget w = new Widget(); w.Foo(); // Widget's implementation of I1.Foo ((I1)w).Foo(); // Widget's implementation of I1.Foo ((I2)w).Foo(); // Widget's implementation of I2.Foo Another reason to explicitly implement interface members is to hide members that are highly specialized and distracting to a type’s normal use case. For example, a type that implements ISerializable would typically want to avoid flaunting its ISerializable members unless explicitly cast to that interface. Implementing Interface Members Virtually An implicitly implemented interface member is, by default, sealed. It must be marked virtual or abstract in the base class in order to be overridden. For example: public interface IUndoable { void Undo(); } public class TextBox : IUndoable { public virtual void Undo() { Console.WriteLine ("TextBox.Undo"); } 98 | Chapter 3: Creating Types in C# } public class RichTextBox : TextBox { public override void Undo() { Console.WriteLine ("RichTextBox.Undo"); } } Calling the interface member through either the base class or the interface calls the subclass’s implementation: RichTextBox r = new RichTextBox(); r.Undo(); // RichTextBox.Undo ((IUndoable)r).Undo(); // RichTextBox.Undo ((TextBox)r).Undo(); // RichTextBox.Undo An explicitly implemented interface member cannot be marked virtual, nor can it be overridden in the usual manner. It can, however, be reimplemented. Reimplementing an Interface in a Subclass A subclass can reimplement any interface member already implemented by a base class. Reimplementation hijacks a member implementation (when called through the interface) and works whether or not the member is virtual in the base class. It also works whether a member is implemented implicitly or explicitly—although it works best in the latter case, as we will demonstrate. In the following example, TextBox implements IUndoable.Undo explicitly, and so it cannot be marked as virtual. In order to “override” it, RichTextBox must reimple- ment IUndoable’s Undo method: public interface IUndoable { void Undo(); } public class TextBox : IUndoable { void IUndoable.Undo() { Console.WriteLine ("TextBox.Undo"); } } public class RichTextBox : TextBox, IUndoable { public new void Undo() { Console.WriteLine ("RichTextBox.Undo"); } } Calling the reimplemented member through the interface calls the subclass’s implementation: RichTextBox r = new RichTextBox(); r.Undo(); // RichTextBox.Undo Case 1 ((IUndoable)r).Undo(); // RichTextBox.Undo Case 2 Interfaces | 99 Creating Types Assuming the same RichTextBox definition, suppose that TextBox implemented Undo implicitly: public class TextBox : IUndoable { public void Undo() { Console.WriteLine ("TextBox.Undo"); } } This would give us another way to call Undo, which would “break” the system, as shown in Case 3: RichTextBox r = new RichTextBox(); r.Undo(); // RichTextBox.Undo Case 1 ((IUndoable)r).Undo(); // RichTextBox.Undo Case 2 ((TextBox)r).Undo(); // TextBox.Undo Case 3 Case 3 demonstrates that reimplementation hijacking is effective only when a member is called through the interface and not through the base class. This is usually undesirable as it can mean inconsistent semantics. This makes reimplementation most appropriate as a strategy for overriding explicitly implemented interface members. Alternatives to interface reimplementation Even with explicit member implementation, interface reimplementation is prob- lematic for a couple of reasons: • The subclass has no way to call the base class method. • The base class author may not anticipate that a method be reimplemented and may not allow for the potential consequences. Reimplementation can be a good last resort when subclassing hasn’t been anticipa- ted. A better option, however, is to design a base class such that reimplementation will never be required. There are two ways to achieve this: • When implicitly implementing a member, mark it virtual if appropriate. • When explicitly implementing a member, use the following pattern if you anticipate that subclasses might need to override any logic: public class TextBox : IUndoable { void IUndoable.Undo() { Undo(); } // Calls method below protected virtual void Undo() { Console.WriteLine ("TextBox.Undo"); } } public class RichTextBox : TextBox { protected override void Undo() { Console.WriteLine("RichTextBox.Undo"); } } If you don’t anticipate any subclassing, you can mark the class as sealed to preempt interface reimplementation. 100 | Chapter 3: Creating Types in C# Interfaces and Boxing Converting a struct to an interface causes boxing. Calling an implicitly implemented member on a struct does not cause boxing: interface I { void Foo(); } struct S : I { public void Foo() {} } ... S s = new S(); s.Foo(); // No boxing. I i = s; // Box occurs when casting to interface. i.Foo(); Writing a Class Versus an Interface As a guideline: • Use classes and subclasses for types that naturally share an implementation. • Use interfaces for types that have independent implementations. Consider the following classes: abstract class Animal {} abstract class Bird : Animal {} abstract class Insect : Animal {} abstract class FlyingCreature : Animal {} abstract class Carnivore : Animal {} // Concrete classes: class Ostrich : Bird {} class Eagle : Bird, FlyingCreature, Carnivore {} // Illegal class Bee : Insect, FlyingCreature {} // Illegal class Flea : Insect, Carnivore {} // Illegal The Eagle, Bee, and Flea classes do not compile because inheriting from multiple classes is prohibited. To resolve this, we must convert some of the types to inter- faces. The question then arises, which types? Following our general rule, we could say that insects share an implementation, and birds share an implementation, so they remain classes. In contrast, flying creatures have independent mechanisms for flying, and carnivores have independent strategies for eating animals, so we would convert FlyingCreature and Carnivore to interfaces: interface IFlyingCreature {} interface ICarnivore {} In a typical scenario, Bird and Insect might correspond to a Windows control and a web control; FlyingCreature and Carnivore might correspond to IPrintable and IUndoable. Interfaces | 101 Creating Types Enums An enum is a special value type that lets you specify a group of named numeric constants. For example: public enum BorderSide { Left, Right, Top, Bottom } We can use this enum type as follows: BorderSide topSide = BorderSide.Top; bool isTop = (topSide == BorderSide.Top); // true Each enum member has an underlying integral value. By default: • Underlying values are of type int. • The constants 0, 1, 2... are automatically assigned, in the declaration order of the enum members. You may specify an alternative integral type, as follows: public enum BorderSide : byte { Left, Right, Top, Bottom } You may also specify an explicit underlying value for each enum member: public enum BorderSide : byte { Left=1, Right=2, Top=10, Bottom=11 } The compiler also lets you explicitly assign some of the enum members. The unassigned enum members keep incrementing from the last explicit value. The preceding example is equivalent to the following: public enum BorderSide : byte { Left=1, Right, Top=10, Bottom } Enum Conversions You can convert an enum instance to and from its underlying integral value with an explicit cast: int i = (int) BorderSide.Left; BorderSide side = (BorderSide) i; bool leftOrRight = (int) side <= 2; You can also explicitly cast one enum type to another. Suppose HorizontalAlign ment is defined as follows: public enum HorizontalAlignment { Left = BorderSide.Left, Right = BorderSide.Right, Center } 102 | Chapter 3: Creating Types in C# A translation between the enum types uses the underlying integral values: HorizontalAlignment h = (HorizontalAlignment) BorderSide.Right; // same as: HorizontalAlignment h = (HorizontalAlignment) (int) BorderSide.Right; The numeric literal 0 is treated specially by the compiler in an enum expression and does not require an explicit cast: BorderSide b = 0; // No cast required if (b == 0) ... There are two reasons for the special treatment of 0: • The first member of an enum is often used as the “default” value. • For combined enum types, 0 means “no flags.” Flags Enums You can combine enum members. To prevent ambiguities, members of a combin- able enum require explicitly assigned values, typically in powers of two. For example: [Flags] public enum BorderSides { None=0, Left=1, Right=2, Top=4, Bottom=8 } To work with combined enum values, you use bitwise operators, such as | and &. These operate on the underlying integral values: BorderSides leftRight = BorderSides.Left | BorderSides.Right; if ((leftRight & BorderSides.Left) != 0) Console.WriteLine ("Includes Left"); // Includes Left string formatted = leftRight.ToString(); // "Left, Right" BorderSides s = BorderSides.Left; s |= BorderSides.Right; Console.WriteLine (s == leftRight); // True s ^= BorderSides.Right; // Toggles BorderSides.Right Console.WriteLine (s); // Left By convention, the Flags attribute should always be applied to an enum type when its members are combinable. If you declare such an enum without the Flags attribute, you can still combine members, but calling ToString on an enum instance will emit a number rather than a series of names. By convention, a combinable enum type is given a plural rather than singular name. For convenience, you can include combination members within an enum declara- tion itself: [Flags] public enum BorderSides { None=0, Enums | 103 Creating Types Left=1, Right=2, Top=4, Bottom=8, LeftRight = Left | Right, TopBottom = Top | Bottom, All = LeftRight | TopBottom } Enum Operators The operators that work with enums are: = == != < > <= >= + - ^ & | ˜ += -= ++ -- sizeof The bitwise, arithmetic, and comparison operators return the result of processing the underlying integral values. Addition is permitted between an enum and an integral type, but not between two enums. Type-Safety Issues Consider the following enum: public enum BorderSide { Left, Right, Top, Bottom } Since an enum can be cast to and from its underlying integral type, the actual value it may have may fall outside the bounds of a legal enum member. For example: BorderSide b = (BorderSide) 12345; Console.WriteLine (b); // 12345 The bitwise and arithmetic operators can produce similarly invalid values: BorderSide b = BorderSide.Bottom; b++; // No errors An invalid BorderSide would break the following code: void Draw (BorderSide side) { if (side == BorderSide.Left) {...} else if (side == BorderSide.Right) {...} else if (side == BorderSide.Top) {...} else {...} // Assume BorderSide.Bottom } One solution is to add another else clause: ... else if (side == BorderSide.Bottom) ... else throw new ArgumentException ("Invalid BorderSide: " + side, "side"); Another workaround is to explicitly check an enum value for validity. The static Enum.IsDefined method does this job: BorderSide side = (BorderSide) 12345; Console.WriteLine (Enum.IsDefined (typeof (BorderSide), side)); // False 104 | Chapter 3: Creating Types in C# Unfortunately, Enum.IsDefined does not work for flagged enums. However, the fol- lowing helper method (a trick dependent on the behavior of Enum.ToString()) returns true if a given flagged enum is valid: static bool IsFlagDefined (Enum e) { decimal d; return !decimal.TryParse(e.ToString(), out d); } [Flags] public enum BorderSides { Left=1, Right=2, Top=4, Bottom=8 } static void Main() { for (int i = 0; i <= 16; i++) { BorderSides side = (BorderSides)i; Console.WriteLine (IsFlagDefined (side) + " " + side); } } Nested Types A nested type is declared within the scope of another type. For example: public class TopLevel { public class Nested { } // Nested class public enum Color { Red, Blue, Tan } // Nested enum } A nested type has the following features: • It can access the enclosing type’s private members and everything else the enclosing type can access. • It can be declared with the full range of access modifiers, rather than just public and internal. • The default accessibility for a nested type is private rather than internal. • Accessing a nested type from outside the enclosing type requires qualification with the enclosing type’s name (like when accessing static members). For example, to access Color.Red from outside our TopLevel class, we’d have to do this: TopLevel.Color color = TopLevel.Color.Red; All types (classes, structs, interfaces, delegates and enums) can be nested inside either a class or a struct. Here is an example of accessing a private member of a type from a nested type: public class TopLevel { static int x; Nested Types | 105 Creating Types class Nested { static void Foo() { Console.WriteLine (TopLevel.x); } } } Here is an example of applying the protected access modifier to a nested type: public class TopLevel { protected class Nested { } } public class SubTopLevel : TopLevel { static void Foo() { new TopLevel.Nested(); } } Here is an example of referring to a nested type from outside the enclosing type: public class TopLevel { public class Nested { } } class Test { TopLevel.Nested n; } Nested types are used heavily by the compiler itself when it generates private classes that capture state for constructs such as iterators and anonymous methods. If the sole reason for using a nested type is to avoid cluttering a namespace with too many types, consider using a nested name- space instead. A nested type should be used because of its stron- ger access control restrictions, or when the nested class must access private members of the containing class. Generics C# has two separate mechanisms for writing code that is reusable across different types: inheritance and generics. Whereas inheritance expresses reusability with a base type, generics express reusability with a “template” that contains “placeholder” types. Generics, when compared to inheritance, can increase type safety and reduce casting and boxing. C# generics and C++ templates are similar concepts, but they work differently. We explain this difference in “C# Generics Versus C++ Templates” on page 118. 106 | Chapter 3: Creating Types in C# Generic Types A generic type declares type parameters—placeholder types to be filled in by the consumer of the generic type, which supplies the type arguments. Here is a generic type Stack, designed to stack instances of type T. Stack declares a single type parameter T: public class Stack { int position; T[] data = new T[100]; public void Push (T obj) { data[position++] = obj; } public T Pop() { return data[--position]; } } We can use Stack as follows: Stack stack = new Stack(); stack.Push(5); stack.Push(10); int x = stack.Pop(); // x is 10 int y = stack.Pop(); // y is 5 Stack fills in the type parameter T with the type argument int, implicitly cre- ating a type on the fly (the synthesis occurs at runtime). Stack effectively has the following definition (substitutions appear in bold, with the class name hashed out to avoid confusion): public class ### { int position; int[] data; public void Push (int obj) { data[position++] = obj; } public int Pop() { return data[--position]; } } Technically, we say that Stack is an open type, whereas Stack is a closed type. At runtime, all generic type instances are closed—with the placeholder types filled in. This means that the following statement is illegal: var stack = new Stack(); // Illegal: What is T? unless inside a class or method which itself defines T as a type parameter: public class Stack { ... public Stack Clone() { Stack clone = new Stack(); // Legal ... } } Generics | 107 Creating Types Why Generics Exist Generics exist to write code that is reusable across different types. Suppose we needed a stack of integers, but we didn’t have generic types. One solution would be to hardcode a separate version of the class for every required element type (e.g., IntStack, StringStack, etc.). Clearly, this would cause considerable code duplica- tion. Another solution would be to write a stack that is generalized by using object as the element type: public class ObjectStack { int position; object[] data = new object[10]; public void Push (object obj) { data[position++] = obj; } public object Pop() { return data[--position]; } } An ObjectStack, however, wouldn’t work as well as a hardcoded IntStack for specif- ically stacking integers. Specifically, an ObjectStack would require boxing and downcasting that could not be checked at compile time: // Suppose we just want to store integers here: ObjectStack stack = new ObjectStack(); stack.Push ("s"); // Wrong type, but no error! int i = (int)stack.Pop(); // Downcast - runtime error What we need is both a general implementation of a stack that works for all element types, and a way to easily specialize that stack to a specific element type for increased type safety and reduced casting and boxing. Generics give us precisely this, by al- lowing us to parameterize the element type. Stack has the benefits of both Object Stack and IntStack. Like ObjectStack, Stack is written once to work generally across all types. Like IntStack, Stack is specialized for a particular type—the beauty is that this type is T, which we substitute on the fly. ObjectStack is functionally equivalent to Stack. Generic Methods A generic method declares type parameters within the signature of a method. With generic methods, many fundamental algorithms can be implemented in a general-purpose way only. Here is a generic method that swaps the contents of two variables of any type T: static void Swap (ref T a, ref T b) { T temp = a; a = b; 108 | Chapter 3: Creating Types in C# b = temp; } Swap can be used as follows: int x = 5; int y = 10; Swap (ref x, ref y); Generally, there is no need to supply type arguments to a generic method, because the compiler can implicitly infer the type. If there is ambiguity, generic methods can be called with the type arguments as follows: Swap (ref x, ref y); Within a generic type, a method is not classed as generic unless it introduces type parameters (with the angle bracket syntax). The Pop method in our generic stack merely uses the type’s existing type parameter, T, and is not classed as a generic method. Methods and types are the only constructs that can introduce type parameters. Properties, indexers, events, fields, constructors, operators, and so on cannot declare type parameters, although they can partake in any type parameters already declared by their enclosing type. In our generic stack example, for instance, we could write an indexer that returns a generic item: public T this [int index] { get { return data [index]; } } Similarly, constructors can partake in existing type parameters, but not introduce them: public Stack() { } // Illegal Declaring Type Parameters Type parameters can be introduced in the declaration of classes, structs, interfaces, delegates (covered in Chapter 4), and methods. Other constructs, such as properties, cannot introduce a type parameter, but can use one. For example, the property Value uses T: public struct Nullable { public T Value { get; } } A generic type or method can have multiple parameters. For example: class Dictionary {...} To instantiate: Dictionary myDic = new Dictionary(); Or: var myDic = new Dictionary(); Generics | 109 Creating Types Generic type names and method names can be overloaded as long as the number of type parameters is different. For example, the following two type names do not conflict: class A {} class A {} By convention, generic types and methods with a single type parameter typically name their parameter T, as long as the intent of the parameter is clear. When using multiple type parameters, each parameter is prefixed with T, but has a more descriptive name. typeof and Unbound Generic Types Open generic types do not exist at runtime: open generic types are closed as part of compilation. However, it is possible for an unbound generic type to exist at runtime —purely as a Type object. The only way to specify an unbound generic type in C# is with the typeof operator: class A {} class A {} ... Type a1 = typeof (A<>); // Unbound type (notice no type arguments). Type a2 = typeof (A<,>); // Use commas to indicate multiple type args. Open generic types are used in conjunction with the Reflection API (Chapter 19). You can also use the typeof operator to specify a closed type: Type a3 = typeof (A); or an open type (which is closed at runtime): class B { void X() { Type t = typeof (T); } } The default Generic Value The default keyword can be used to get the default value given a generic type pa- rameter. The default value for a reference type is null, and the default value for a value type is the result of bitwise-zeroing the value type’s fields: static void Zap (T[] array) { for (int i = 0; i < array.Length; i++) array[i] = default(T); } Generic Constraints By default, a type parameter can be substituted with any type whatsoever. Con- straints can be applied to a type parameter to require more specific type arguments. 110 | Chapter 3: Creating Types in C# These are the possible constraints: where T : base-class // Base-class constraint where T : interface // Interface constraint where T : class // Reference-type constraint where T : struct // Value-type constraint (excludes Nullable types) where T : new() // Parameterless constructor constraint where U : T // Naked type constraint In the following example, GenericClass requires T to derive from (or be iden- tical to) SomeClass and implement Interface1, and requires U to provide a parame- terless constructor: class SomeClass {} interface Interface1 {} class GenericClass where T : SomeClass, Interface1 where U : new() {...} Constraints can be applied wherever type parameters are defined, in both methods and type definitions. A base-class constraint specifies that the type parameter must subclass (or match) a particular class; an interface constraint specifies that the type parameter must im- plement that interface. These constraints allow instances of the type parameter to be implicitly converted to that class or interface. For example, suppose we want to write a generic Max method, which returns the maximum of two values. We can take advantage of the generic interface defined in the framework called IComparable: public interface IComparable // Simplified version of interface { int CompareTo (T other); } CompareTo returns a positive number if this is greater than other. Using this interface as a constraint, we can write a Max method as follows (to avoid distraction, null checking is omitted): static T Max (T a, T b) where T : IComparable { return a.CompareTo (b) > 0 ? a : b; } The Max method can accept arguments of any type implementing IComparable (which includes most built-in types such as int and string): int z = Max (5, 10); // 10 string last = Max ("ant", "zoo"); // zoo The class constraint and struct constraint specify that T must be a reference type or (non-nullable) value type. A great example of the struct constraint is the System.Nullable struct (we will discuss this class in depth in “Nullable Types” on page 153 in Chapter 4): struct Nullable where T : struct {...} Generics | 111 Creating Types The parameterless constructor constraint requires T to have a public parameterless constructor. If this constraint is defined, you can call new() on T: static void Initialize (T[] array) where T : new() { for (int i = 0; i < array.Length; i++) array[i] = new T(); } The naked type constraint requires one type parameter to derive from (or match) another type parameter. In this example, the method FilteredStack returns another Stack, containing only the subset of elements where the type parameter U is of the type parameter T: class Stack { Stack FilteredStack() where U : T {...} } Subclassing Generic Types A generic class can be subclassed just like a nongeneric class. The subclass can leave the base class’s type parameters open, as in the following example: class Stack {...} class SpecialStack : Stack {...} Or the subclass can close the generic type parameters with a concrete type: class IntStack : Stack {...} A subtype can also introduce fresh type arguments: class List {...} class KeyedList : List {...} Technically, all type arguments on a subtype are fresh: you could say that a subtype closes and then reopens the base type arguments. This means that a subclass can give new (and potentially more meaningful) names to the type arguments it reopens: class List {...} class KeyedList : List {...} Self-Referencing Generic Declarations A type can name itself as the concrete type when closing a type argument: public interface IEquatable { bool Equals (T obj); } public class Balloon : IEquatable { public string Color { get; set; } public int CC { get; set; } 112 | Chapter 3: Creating Types in C# public bool Equals (Balloon b) { if (b == null) return false; return b.Color == Color && b.CC == CC; } } The following are also legal: class Foo where T : IComparable { ... } class Bar where T : Bar { ... } Static Data Static data is unique for each closed type: class Bob { public static int Count; } class Test { static void Main() { Console.WriteLine (++Bob.Count); // 1 Console.WriteLine (++Bob.Count); // 2 Console.WriteLine (++Bob.Count); // 1 Console.WriteLine (++Bob.Count); // 1 } } Type Parameters and Conversions C#’s cast operator can perform several kinds of conversion, including: • Numeric conversion • Reference conversion • Boxing/unboxing conversion • Custom conversion (via operator overloading; see Chapter 4) The decision as to which kind of conversion will take place happens at compile time, based on the known types of the operands. This creates an interesting scenario with generic type parameters, because the precise operand types are unknown at compile time. If this leads to ambiguity, the compiler generates an error. The most common scenario is when you want to perform a reference conversion: StringBuilder Foo (T arg) { if (arg is StringBuilder) return (StringBuilder) arg; // Will not compile ... } Without knowledge of T’s actual type, the compiler is concerned that you might have intended this to be a custom conversion. The simplest solution is to instead use Generics | 113 Creating Types the as operator, which is unambiguous because it cannot perform custom conversions: StringBuilder Foo (T arg) { StringBuilder sb = arg as StringBuilder; if (sb != null) return sb; ... } A more general solution is to first cast to object. This works because conversions to/from object are assumed not to be custom conversions, but reference or boxing/ unboxing conversions. In this case, StringBuilder is a reference type, so it has to be a reference conversion: return (StringBuilder) (object) arg; Unboxing conversions can also introduce ambiguities. The following could be an unboxing, numeric, or custom conversion: int Foo (T x) { return (int) x; } // Compile-time error The solution, again, is to first cast to object and then to int (which then unambig- uously signals an unboxing conversion in this case): int Foo (T x) { return (int) (object) x; } Covariance Assuming A is convertible to B, X is covariant if X is convertible to X. With C#’s notion of covariance (and contravariance), “conver- tible” means convertible via an implicit reference conversion— such as A subclassing B, or A implementing B. Numeric conversions, boxing conversions, and custom conversions are not included. For instance, type IFoo is covariant for T if the following is legal: IFoo s = ...; IFoo b = s; From C# 4.0, generic interfaces permit covariance for (as do generic delegates—see Chapter 4), but generic classes do not. Arrays also support covariance (A[] can be converted to B[] if A has an implicit reference conversion to B), and are discussed here for comparison. 114 | Chapter 3: Creating Types in C# Covariance and contravariance (or simply “variance”) are advanced concepts. The motivation behind introducing and enhancing variance in C# was to allow generic interface and generic types (in particular, those defined in the Framework, such as IEnumerable) to work more as you’d expect. You can benefit from this without understanding the details behind co- variance and contravariance. Classes Generic classes are not covariant, to ensure static type safety. Consider the following: class Animal {} class Bear : Animal {} class Camel : Animal {} public class Stack // A simple Stack implementation { int position; T[] data = new T[100]; public void Push (T obj) { data[position++] = obj; } public T Pop() { return data[--position]; } } The following fails to compile: Stack bears = new Stack(); Stack animals = bears; // Compile-time error That restriction prevents the possibility of runtime failure with the following code: animals.Push (new Camel()); // Trying to add Camel to bears Lack of covariance, however, can hinder reusability. Suppose, for instance, we wanted to write a method to Wash a stack of animals: public class ZooCleaner { public static void Wash (Stack animals) {...} } Calling Wash with a stack of bears would generate a compile-time error. One work- around is to redefine the Wash method with a constraint: class ZooCleaner { public static void Wash (Stack animals) where T : Animal { ... } } We can now call Wash as follows: Stack bears = new Stack(); ZooCleaner.Wash (bears); Another solution is to have Stack implement a covariant generic interface, as we’ll see shortly. Generics | 115 Creating Types Arrays For historical reasons, array types are covariant. This means that B[] can be cast to A[] if B subclasses A (and both are reference types). For example: Bear[] bears = new Bear[3]; Animal[] animals = bears; // OK The downside of this reusability is that element assignments can fail at runtime: animals[0] = new Camel(); // Runtime error Interfaces As of C# 4.0, generic interfaces support covariance for type parameters marked with the out modifier. This modifier ensures that, unlike with arrays, covariance with interfaces is fully type-safe. To illustrate, suppose that our Stack class implements the following interface: public interface IPoppable { T Pop(); } The out modifier on T indicates that T is used only in output positions (e.g., return types for methods). The out modifier flags the interface as covariant and allows us to do this: var bears = new Stack(); bears.Push (new Bear()); // Bears implements IPoppable. We can convert to IPoppable: IPoppable animals = bears; // Legal Animal a = animals.Pop(); The cast from bears to animals is permitted by the compiler—by virtue of the interface being covariant. This is type-safe because the case the compiler is trying to avoid—pushing a Camel onto the stack—can’t occur as there’s no way to feed a Camel into an interface where T can appear only in output positions. Covariance (and contravariance) in interfaces is something that you typically consume: it’s less common that you need to write variant interfaces. Curiously, method parameters marked as out are not eligible for covariance, due to a limitation in the CLR. We can leverage the ability to cast covariantly to solve the reusability problem described earlier: public class ZooCleaner { public static void Wash (IPoppable animals) { ... } } The IEnumerator and IEnumerable interfaces described in Chapter 7 are marked as covariant. This allows you to cast IEnumerable to IEnumerable, for instance. 116 | Chapter 3: Creating Types in C# The compiler will generate an error if you use a covariant type parameter in an input position (e.g., a parameter to a method or a writable property). With both generic types and arrays, covariance (and contravar- iance) is valid only for elements with reference conversions—not boxing conversions. So, if you wrote a method that accepted a parameter of type IPoppable, you could call it with IPoppable, but not IPoppable. Contravariance We previously saw that, assuming that A allows an implicit reference conversion to B, a type X is covariant if X allows a reference conversion to X. A type is con- travariant when you can convert in the reverse direction—from X to X. This is supported with generic interfaces—when the generic type parameter only appears in input positions, designated with the in modifier. Extending our previous example, if the Stack class implements the following interface: public interface IPushable { void Push (T obj); } we can legally do this: IPushable animals = new Stack(); IPushable bears = animals; // Legal bears.Push (new Bear()); No member in IPushable outputs a T, so we can’t get into trouble by casting animals to bears (there’s no way to Pop, for instance, through that interface). Our Stack class can implement both IPushable and IPop pable—despite T having opposing variance annotations in the two interfaces! This works because you can exercise variance only through an interface; therefore, you must commit to the lens of either IPoppable or IPushable before performing a variant conversion. This lens then restricts you to the operations that are legal under the appropriate variance rules. This also illustrates why it would usually make no sense for classes (such as Stack) to be variant: concrete implementa- tions typically require data to flow in both directions. To give another example, consider the following interface, defined as part of the .NET Framework: public interface IComparer { // Returns a value indicating the relative ordering of a and b int Compare (T a, T b); } Generics | 117 Creating Types Because the interface is contravariant, we can use an IComparer to compare two strings: var objectComparer = Comparer.Default; // objectComparer implements IComparer IComparer stringComparer = objectComparer; int result = stringComparer.Compare ("Brett", "Jemaine"); Mirroring covariance, the compiler will report an error if you try to use a contra- variant parameter in an output position (e.g., as a return value, or in a readable property). C# Generics Versus C++ Templates C# generics are similar in application to C++ templates, but they work very differ- ently. In both cases, a synthesis between the producer and consumer must take place, where the placeholder types of the producer are filled in by the consumer. However, with C# generics, producer types (i.e., open types such as List) can be compiled into a library (such as mscorlib.dll). This works because the synthesis between the producer and the consumer that produces closed types doesn’t actually happen until runtime. With C++ templates, this synthesis is performed at compile time. This means that in C++ you don’t deploy template libraries as .dlls—they exist only as source code. It also makes it difficult to dynamically inspect, let alone create, para- meterized types on the fly. To dig deeper into why this is the case, consider the Max method in C#, once more: static T Max (T a, T b) where T : IComparable { return a.CompareTo (b) > 0 ? a : b; } Why couldn’t we have implemented it like this? static T Max (T a, T b) { return a > b ? a : b; // Compile error } The reason is that Max needs to be compiled once and work for all possible values of T. Compilation cannot succeed, because there is no single meaning for > across all values of T—in fact, not every T even has a > operator. In contrast, the following code shows the same Max method written with C++ templates. This code will be compiled separately for each value of T, taking on whatever semantics > has for a particular T, failing to compile if a particular T does not support the > operator: template T Max (T a, T b) { return a > b ? a : b; } 118 | Chapter 3: Creating Types in C# 4Advanced C# In this chapter, we cover advanced C# topics that build on concepts explored in Chapters 2 and 3. You should read the first four sections sequentially; you can read the remaining sections in any order. Delegates A delegate is an object that knows how to call a method. A delegate type defines the kind of method that delegate instances can call. Specifi- cally, it defines the method’s return type and its parameter types. The following defines a delegate type called Transformer: delegate int Transformer (int x); Transformer is compatible with any method with an int return type and a single int parameter, such as this: static int Square (int x) { return x * x; } Assigning a method to a delegate variable creates a delegate instance: Transformer t = Square; which can be invoked in the same way as a method: int answer = t(3); // answer is 9 Here’s a complete example: delegate int Transformer (int x); class Test { static void Main() { Transformer t = Square; // Create delegate instance int result = t(3); // Invoke delegate Console.WriteLine (result); // 9 119 } static int Square (int x) { return x * x; } } A delegate instance literally acts as a delegate for the caller: the caller invokes the delegate, and then the delegate calls the target method. This indirection decouples the caller from the target method. The statement: Transformer t = Square; is shorthand for: Transformer t = new Transformer (Square); Technically, we are specifying a method group when we refer to Square without brackets or arguments. If the method is over- loaded, C# will pick the correct overload based on the signature of the delegate to which it’s being assigned. The expression: t(3) is shorthand for: t.Invoke(3) A delegate is similar to a callback, a general term that captures constructs such as C function pointers. Writing Plug-in Methods with Delegates A delegate variable is assigned a method at runtime. This is useful for writing plug- in methods. In this example, we have a utility method named Transform that applies a transform to each element in an integer array. The Transform method has a delegate parameter, for specifying a plug-in transform. public delegate int Transformer (int x); class Util { public static void Transform (int[] values, Transformer t) { for (int i = 0; i < values.Length; i++) values[i] = t (values[i]); } } class Test { 120 | Chapter 4: Advanced C# static void Main() { int[] values = { 1, 2, 3 }; Util.Transform (values, Square); // Hook in the Square method foreach (int i in values) Console.Write (i + " "); // 1 4 9 } static int Square (int x) { return x * x; } } Multicast Delegates All delegate instances have multicast capability. This means that a delegate instance can reference not just a single target method, but also a list of target methods. The + and += operators combine delegate instances. For example: SomeDelegate d = SomeMethod1; d += SomeMethod2; The last line is functionally the same as: d = d + SomeMethod2; Invoking d will now call both SomeMethod1 and SomeMethod2. Delegates are invoked in the order they are added. The - and -= operators remove the right delegate operand from the left delegate operand. For example: d -= SomeMethod1; Invoking d will now cause only SomeMethod2 to be invoked. Calling + or += on a delegate variable with a null value works, and it is equivalent to assigning the variable to a new value: SomeDelegate d = null; d += SomeMethod1; // Equivalent (when d is null) to d = SomeMethod1; Similarly, calling −= on a delegate variable with a single target is equivalent to assigning null to that variable. Delegates are immutable, so when you call += or −=, you’re in fact creating a new delegate instance and assigning it to the existing variable. If a multicast delegate has a nonvoid return type, the caller receives the return value from the last method to be invoked. The preceding methods are still called, but their return values are discarded. In most scenarios in which multicast delegates are used, they have void return types, so this subtlety does not arise. Delegates | 121 Advanced C# All delegate types implicitly derive from System.MulticastDele gate, which inherits from System.Delegate. C# compiles +, -, +=, and -= operations made on a delegate to the static Combine and Remove methods of the System.Delegate class. Multicast delegate example Suppose you wrote a routine that took a long time to execute. That routine could regularly report progress to its caller by invoking a delegate. In this example, the HardWork routine has a ProgressReporter delegate parameter, which it invokes to indicate progress: public delegate void ProgressReporter (int percentComplete); public class Util { public static void HardWork (ProgressReporter p) { for (int i = 0; i < 10; i++) { p (i * 10); // Invoke delegate System.Threading.Thread.Sleep (100); // Simulate hard work } } } To monitor progress, the Main method creates a multicast delegate instance p, such that progress is monitored by two independent methods: class Test { static void Main() { ProgressReporter p = WriteProgressToConsole; p += WriteProgressToFile; Util.HardWork (p); } static void WriteProgressToConsole (int percentComplete) { Console.WriteLine (percentComplete); } static void WriteProgressToFile (int percentComplete) { System.IO.File.WriteAllText ("progress.txt", percentComplete.ToString()); } } Instance Versus Static Method Targets When an instance method is assigned to delegate object, the latter must maintain a reference not only to the method, but also to the instance to which the method 122 | Chapter 4: Advanced C# belongs. The System.Delegate class’s Target property represents this instance (and will be null for a delegate referencing a static method). For example: public delegate void ProgressReporter (int percentComplete); class Test { static void Main() { X x = new X(); ProgressReporter p = x.InstanceProgress; p(99); // 99 Console.WriteLine (p.Target == x); // True Console.WriteLine (p.Method); // Void InstanceProgress(Int32) } } class X { public void InstanceProgress (int percentComplete) { Console.WriteLine (percentComplete); } } Generic Delegate Types A delegate type may contain generic type parameters. For example: public delegate T Transformer (T arg); With this definition, we can write a generalized Transform utility method that works on any type: public class Util { public static void Transform (T[] values, Transformer t) { for (int i = 0; i < values.Length; i++) values[i] = t (values[i]); } } class Test { static void Main() { int[] values = { 1, 2, 3 }; Util.Transform (values, Square); // Hook in Square foreach (int i in values) Console.Write (i + " "); // 1 4 9 } static int Square (int x) { return x * x; } } Delegates | 123 Advanced C# The Func and Action Delegates With generic delegates, it becomes possible to write a small set of delegate types that are so general they can work for methods of any return type and any (reasonable) number of arguments. These delegates are the Func and Action delegates, defined in the System namespace (the in and out annotations indicate variance, which we will cover shortly): delegate TResult Func (); delegate TResult Func (T arg); delegate TResult Func (T1 arg1, T2 arg2); ... and so on, up to T16 delegate void Action (); delegate void Action (T arg); delegate void Action (T1 arg1, T2 arg2); ... and so on, up to T16 These delegates are extremely general. The Transformer delegate in our previous example can be replaced with a Func delegate that takes a single argument of type T and returns a same-typed value: public static void Transform (T[] values, Func transformer) { for (int i = 0; i < values.Length; i++) values[i] = transformer (values[i]); } The only practical scenarios not covered by these delegates are ref/out and pointer parameters. Prior to Framework 2.0, the Func and Action delegates did not exist (because generics did not exist). It’s for this historical rea- son that much of the Framework uses custom delegate types rather than Func and Action. Delegates Versus Interfaces A problem that can be solved with a delegate can also be solved with an interface. For instance, the following explains how to solve our filter problem using an ITrans former interface: public interface ITransformer { int Transform (int x); } public class Util { public static void TransformAll (int[] values, ITransformer t) { for (int i = 0; i < values.Length; i++) values[i] = t.Transform (values[i]); 124 | Chapter 4: Advanced C# } } class Squarer : ITransformer { public int Transform (int x) { return x * x; } } ... static void Main() { int[] values = { 1, 2, 3 }; Util.TransformAll (values, new Squarer()); foreach (int i in values) Console.WriteLine (i); } A delegate design may be a better choice than an interface design if one or more of these conditions are true: • The interface defines only a single method. • Multicast capability is needed. • The subscriber needs to implement the interface multiple times. In the ITransformer example, we don’t need to multicast. However, the interface defines only a single method. Furthermore, our subscriber may need to implement ITransformer multiple times, to support different transforms, such as square or cube. With interfaces, we’re forced into writing a separate type per transform, since Test can implement ITransformer only once. This is quite cumbersome: class Squarer : ITransformer { public int Transform (int x) { return x * x; } } class Cuber : ITransformer { public int Transform (int x) {return x * x * x; } } ... static void Main() { int[] values = { 1, 2, 3 }; Util.TransformAll (values, new Cuber()); foreach (int i in values) Console.WriteLine (i); } Delegates | 125 Advanced C# Delegate Compatibility Type compatibility Delegate types are all incompatible with each other, even if their signatures are the same: delegate void D1(); delegate void D2(); ... D1 d1 = Method1; D2 d2 = d1; // Compile-time error The following, however, is permitted: D2 d2 = new D2 (d1); Delegate instances are considered equal if they have the same method targets: delegate void D(); ... D d1 = Method1; D d2 = Method1; Console.WriteLine (d1 == d2); // True Multicast delegates are considered equal if they reference the same methods in the same order. Parameter compatibility When you call a method, you can supply arguments that have more specific types than the parameters of that method. This is ordinary polymorphic behavior. For exactly the same reason, a delegate can have more specific parameter types than its method target. This is called contravariance. Here’s an example: delegate void StringAction (string s); class Test { static void Main() { StringAction sa = new StringAction (ActOnObject); sa ("hello"); } static void ActOnObject (object o) { Console.WriteLine (o); // hello 126 | Chapter 4: Advanced C# } } (As with type parameter variance, delegates are variant only for reference conversions.) A delegate merely calls a method on someone else’s behalf. In this case, the String Action is invoked with an argument of type string. When the argument is then relayed to the target method, the argument gets implicitly upcast to an object. The standard event pattern is designed to help you leverage contravariance through its use of the common EventArgs base class. For example, you can have a single method invoked by two different delegates, one passing a MouseEventArgs and the other passing a KeyEventArgs. Return type compatibility If you call a method, you may get back a type that is more specific than what you asked for. This is ordinary polymorphic behavior. For exactly the same reason, a delegate target method may return a more specific type than described by the dele- gate. This is called covariance. For example: delegate object ObjectRetriever(); class Test { static void Main() { ObjectRetriever o = new ObjectRetriever (RetriveString); object result = o(); Console.WriteLine (result); // hello } static string RetriveString() { return "hello"; } } The ObjectRetriever expects to get back an object, but an object subclass will also do; delegate return types are covariant. Generic delegate type parameter variance In Chapter 3 we saw how generic interfaces support covariant and contravariant type parameters. The same capability exists for delegates, too (from C# 4.0 onward). If you’re defining a generic delegate type, it’s good practice to: • Mark a type parameter used only on the return value as covariant (out). • Mark any type parameters used only on parameters as contravariant (in). Doing so allows conversions to work naturally by respecting inheritance relation- ships between types. Delegates | 127 Advanced C# The following delegate (defined in the System namespace) supports covariance: delegate TResult Func(); allowing: Func x = ...; Func y = x; The following delegate (defined in the System namespace) supports contravariance: delegate void Action (T arg); allowing: Action x = ...; Action y = x; Events When using delegates, two emergent roles commonly appear: broadcaster and subscriber. The broadcaster is a type that contains a delegate field. The broadcaster decides when to broadcast, by invoking the delegate. The subscribers are the method target recipients. A subscriber decides when to start and stop listening, by calling += and −= on the broadcaster’s delegate. A subscriber does not know about, or interfere with, other subscribers. Events are a language feature that formalizes this pattern. An event is a construct that exposes just the subset of delegate features required for the broadcaster/sub- scriber model. The main purpose of events is to prevent subscribers from interfering with each other. The easiest way to declare an event is to put the event keyword in front of a delegate member: // Delegate definition public delegate void PriceChangedHandler (decimal oldPrice, decimal newPrice); public class Broadcaster { // Event declaration public event PriceChangedHandler PriceChanged; } Code within the Broadcaster type has full access to PriceChanged and can treat it as a delegate. Code outside of Broadcaster can only perform += and −= operations on the PriceChanged event. 128 | Chapter 4: Advanced C# How Do Events Work on the Inside? Three things happen under the covers when you declare an event as follows: public class Broadcaster { public event PriceChangedHandler PriceChanged; } First, the compiler translates the event declaration into something close to the following: PriceChangedHandler _priceChanged; // private delegate public event PriceChangedHandler PriceChanged { add { _priceChanged += value; } remove { _priceChanged -= value; } } The add and remove keywords denote explicit event accessors—which act rather like property accessors. We’ll describe how to write these later. Second, the compiler looks within the Broadcaster class for references to Price Changed that perform operations other than += or −=, and redirects them to the underlying _priceChanged delegate field. Third, the compiler translates += and −= operations on the event to calls to the event’s add and remove accessors. Interestingly, this makes the behavior of += and −= unique when applied to events: unlike in other scenarios, it’s not simply a shortcut for + and − followed by an assignment. Consider the following example. The Stock class fires its PriceChanged event every time the Price of the Stock changes: public delegate void PriceChangedHandler (decimal oldPrice, decimal newPrice); public class Stock { string symbol; decimal price; public Stock (string symbol) { this.symbol = symbol; } public event PriceChangedHandler PriceChanged; public decimal Price { get { return price; } set { if (price == value) return; // Exit if nothing has changed decimal oldPrice = price; price = value; if (PriceChanged != null) // If invocation list not PriceChanged (oldPrice, price); // empty, fire event. Events | 129 Advanced C# } } } If we remove the event keyword from our example so that PriceChanged becomes an ordinary delegate field, our example would give the same results. However, Stock would be less robust, in that subscribers could do the following things to interfere with each other: • Replace other subscribers by reassigning PriceChanged (instead of using the += operator). • Clear all subscribers (by setting PriceChanged to null). • Broadcast to other subscribers by invoking the delegate. WinRT events have slightly different semantics in that attaching to an event returns a token which is required to detach from the event. The compiler transparently bridges this gap (by main- taining an internal dictionary of tokens) so that you can con- sume WinRT events as though they were ordinary CLR events. Standard Event Pattern The .NET Framework defines a standard pattern for writing events. Its purpose is to provide consistency across both Framework and user code. At the core of the standard event pattern is System.EventArgs: a predefined Framework class with no members (other than the static Empty property). EventArgs is a base class for con- veying information for an event. In our Stock example, we would subclass EventArgs to convey the old and new prices when a PriceChanged event is fired: public class PriceChangedEventArgs : System.EventArgs { public readonly decimal LastPrice; public readonly decimal NewPrice; public PriceChangedEventArgs (decimal lastPrice, decimal newPrice) { LastPrice = lastPrice; NewPrice = newPrice; } } For reusability, the EventArgs subclass is named according to the information it contains (rather than the event for which it will be used). It typically exposes data as properties or as read-only fields. With an EventArgs subclass in place, the next step is to choose or define a delegate for the event. 130 | Chapter 4: Advanced C# There are three rules: • It must have a void return type. • It must accept two arguments: the first of type object, and the second a subclass of EventArgs. The first argument indicates the event broadcaster, and the second argument contains the extra information to convey. • Its name must end with EventHandler. The Framework defines a generic delegate called System.EventHandler<> that satis- fies these rules: public delegate void EventHandler (object source, TEventArgs e) where TEventArgs : EventArgs; Before generics existed in the language (prior to C# 2.0), we would have had to instead write a custom delegate as follows: public delegate void PriceChangedHandler (object sender, PriceChangedEventArgs e); For historical reasons, most events within the Framework use delegates defined in this way. The next step is to define an event of the chosen delegate type. Here, we use the generic EventHandler delegate: public class Stock { ... public event EventHandler PriceChanged; } Finally, the pattern requires that you write a protected virtual method that fires the event. The name must match the name of the event, prefixed with the word On, and then accept a single EventArgs argument: public class Stock { ... public event EventHandler PriceChanged; protected virtual void OnPriceChanged (PriceChangedEventArgs e) { if (PriceChanged != null) PriceChanged (this, e); } } Events | 131 Advanced C# In multithreaded scenarios (Chapter 14), you need to assign the delegate to a temporary variable before testing and invoking it, to avoid an obvious thread-safety error: var temp = PriceChanged; if (temp != null) temp (this, e); This provides a central point from which subclasses can invoke or override the event (assuming the class is not sealed). Here’s the complete example: using System; public class PriceChangedEventArgs : EventArgs { public readonly decimal LastPrice; public readonly decimal NewPrice; public PriceChangedEventArgs (decimal lastPrice, decimal newPrice) { LastPrice = lastPrice; NewPrice = newPrice; } } public class Stock { string symbol; decimal price; public Stock (string symbol) {this.symbol = symbol;} public event EventHandler PriceChanged; protected virtual void OnPriceChanged (PriceChangedEventArgs e) { if (PriceChanged != null) PriceChanged (this, e); } public decimal Price { get { return price; } set { if (price == value) return; decimal oldPrice = price; price = value; OnPriceChanged (new PriceChangedEventArgs (oldPrice, price)); } } } class Test { static void Main() 132 | Chapter 4: Advanced C# { Stock stock = new Stock ("THPW"); stock.Price = 27.10M; // Register with the PriceChanged event stock.PriceChanged += stock_PriceChanged; stock.Price = 31.59M; } static void stock_PriceChanged (object sender, PriceChangedEventArgs e) { if ((e.NewPrice - e.LastPrice) / e.LastPrice > 0.1M) Console.WriteLine ("Alert, 10% stock price increase!"); } } The predefined nongeneric EventHandler delegate can be used when an event doesn’t carry extra information. In this example, we rewrite Stock such that the Price Changed event is fired after the price changes, and no information about the event is necessary, other than it happened. We also make use of the EventArgs.Empty prop- erty, in order to avoid unnecessarily instantiating an instance of EventArgs. public class Stock { string symbol; decimal price; public Stock (string symbol) { this.symbol = symbol; } public event EventHandler PriceChanged; protected virtual void OnPriceChanged (EventArgs e) { if (PriceChanged != null) PriceChanged (this, e); } public decimal Price { get { return price; } set { if (price == value) return; price = value; OnPriceChanged (EventArgs.Empty); } } } Event Accessors An event’s accessors are the implementations of its += and −= functions. By default, accessors are implemented implicitly by the compiler. Consider this event declaration: public event EventHandler PriceChanged; Events | 133 Advanced C# The compiler converts this to the following: • A private delegate field • A public pair of event accessor functions (add_PriceChanged and remove_Price Changed), whose implementations forward the += and −= operations to the pri- vate delegate field You can take over this process by defining explicit event accessors. Here’s a manual implementation of the PriceChanged event from our previous example: private EventHandler _priceChanged; // Declare a private delegate public event EventHandler PriceChanged { add { _priceChanged += value; } remove { _priceChanged -= value; } } This example is functionally identical to C#’s default accessor implementation (ex- cept that C# also ensures thread safety around updating the delegate via a lock-free compare-and-swap algorithm—see http://albahari.com/threading). By defining event accessors ourselves, we instruct C# not to generate default field and accessor logic. With explicit event accessors, you can apply more complex strategies to the storage and access of the underlying delegate. There are three scenarios where this is useful: • When the event accessors are merely relays for another class that is broadcasting the event. • When the class exposes a large number of events, where most of the time very few subscribers exist, such as a Windows control. In such cases, it is better to store the subscriber’s delegate instances in a dictionary, since a dictionary will contain less storage overhead than dozens of null delegate field references. • When explicitly implementing an interface that declares an event. Here is an example that illustrates the last point: public interface IFoo { event EventHandler Ev; } class Foo : IFoo { private EventHandler ev; event EventHandler IFoo.Ev { add { ev += value; } remove { ev -= value; } } } 134 | Chapter 4: Advanced C# The add and remove parts of an event are compiled to add_XXX and remove_XXX methods. Event Modifiers Like methods, events can be virtual, overridden, abstract, or sealed. Events can also be static: public class Foo { public static event EventHandler StaticEvent; public virtual event EventHandler VirtualEvent; } Lambda Expressions A lambda expression is an unnamed method written in place of a delegate instance. The compiler immediately converts the lambda expression to either: • A delegate instance. • An expression tree, of type Expression, representing the code inside the lambda expression in a traversable object model. This allows the lambda expression to be interpreted later at runtime (see “Building Query Expres- sions” on page 368 in Chapter 8). Given the following delegate type: delegate int Transformer (int i); we could assign and invoke the lambda expression x => x * x as follows: Transformer sqr = x => x * x; Console.WriteLine (sqr(3)); // 9 Internally, the compiler resolves lambda expressions of this type by writing a private method, and moving the expression’s code into that method. A lambda expression has the following form: (parameters) => expression-or-statement-block For convenience, you can omit the parentheses if and only if there is exactly one parameter of an inferable type. In our example, there is a single parameter, x, and the expression is x * x: x => x * x; Lambda Expressions | 135 Advanced C# Each parameter of the lambda expression corresponds to a delegate parameter, and the type of the expression (which may be void) corresponds to the return type of the delegate. In our example, x corresponds to parameter i, and the expression x * x corresponds to the return type int, therefore being compatible with the Transformer delegate: delegate int Transformer (int i); A lambda expression’s code can be a statement block instead of an expression. We can rewrite our example as follows: x => { return x * x; }; Lambda expressions are used most commonly with the Func and Action delegates, so you will most often see our earlier expression written as follows: Func sqr = x => x * x; Here’s an example of an expression that accepts two parameters: Func totalLength = (s1, s2) => s1.Length + s2.Length; int total = totalLength ("hello", "world"); // total is 10; Lambda expressions were introduced in C# 3.0. Explicitly Specifying Lambda Parameter Types The compiler can usually infer the type of lambda parameters contextually. When this is not the case, you must specify the type of each parameter explicitly. Consider the following expression: Func sqr = x => x * x; The compiler uses type inference to infer that x is an int. We could explicitly specify x’s type as follows: Func sqr = (int x) => x * x; Capturing Outer Variables A lambda expression can reference the local variables and parameters of the method in which it’s defined (outer variables). For example: static void Main() { int factor = 2; Func multiplier = n => n * factor; Console.WriteLine (multiplier (3)); // 6 } Outer variables referenced by a lambda expression are called captured variables. A lambda expression that captures variables is called a closure. Captured variables are evaluated when the delegate is actually invoked, not when the variables were captured: 136 | Chapter 4: Advanced C# int factor = 2; Func multiplier = n => n * factor; factor = 10; Console.WriteLine (multiplier (3)); // 30 Lambda expressions can themselves update captured variables: int seed = 0; Func natural = () => seed++; Console.WriteLine (natural()); // 0 Console.WriteLine (natural()); // 1 Console.WriteLine (seed); // 2 Captured variables have their lifetimes extended to that of the delegate. In the fol- lowing example, the local variable seed would ordinarily disappear from scope when Natural finished executing. But because seed has been captured, its lifetime is extended to that of the capturing delegate, natural: static Func Natural() { int seed = 0; return () => seed++; // Returns a closure } static void Main() { Func natural = Natural(); Console.WriteLine (natural()); // 0 Console.WriteLine (natural()); // 1 } A local variable instantiated within a lambda expression is unique per invocation of the delegate instance. If we refactor our previous example to instantiate seed within the lambda expression, we get a different (in this case, undesirable) result: static Func Natural() { return() => { int seed = 0; return seed++; }; } static void Main() { Func natural = Natural(); Console.WriteLine (natural()); // 0 Console.WriteLine (natural()); // 0 } Capturing is internally implemented by “hoisting” the captured variables into fields of a private class. When the method is called, the class is instantiated and lifetime-bound to the dele- gate instance. Lambda Expressions | 137 Advanced C# Capturing iteration variables When you capture the iteration variable of a for loop, C# treats that variable as though it was declared outside the loop. This means that the same variable is captured in each iteration. The following program writes 333 instead of writing 012: Action[] actions = new Action[3]; for (int i = 0; i < 3; i++) actions [i] = () => Console.Write (i); foreach (Action a in actions) a(); // 333 Each closure (shown in boldface) captures the same variable, i. (This actually makes sense when you consider that i is a variable whose value persists between loop iter- ations; you can even explicitly change i within the loop body if you want.) The consequence is that when the delegates are later invoked, each delegate sees i’s value at the time of invocation—which is 3. We can illustrate this better by expanding the for loop as follows: Action[] actions = new Action[3]; int i = 0; actions[0] = () => Console.Write (i); i = 1; actions[1] = () => Console.Write (i); i = 2; actions[2] = () => Console.Write (i); i = 3; foreach (Action a in actions) a(); // 333 The solution, if we want to write 012, is to assign the iteration variable to a local variable that’s scoped inside the loop: Action[] actions = new Action[3]; for (int i = 0; i < 3; i++) { int loopScopedi = i; actions [i] = () => Console.Write (loopScopedi); } foreach (Action a in actions) a(); // 012 This causes the closure to capture a different variable on each iteration. Prior to C# 5.0, foreach loops worked in the same way: Action[] actions = new Action[3]; int i = 0; foreach (char c in "abc") actions [i++] = () => Console.Write (c); 138 | Chapter 4: Advanced C# foreach (Action a in actions) a(); // ccc in C# 4.0 This caused considerable confusion: unlike with a for loop, the iteration variable in a foreach loop is immutable, and so one would expect it to be treated as local to the loop body. The good news is that it’s been fixed in C# 5.0, and the example above now writes “abc.” Technically, this is a breaking change because recompiling a C# 4.0 program in C# 5.0 could create a different result. In general, the C# team tries to avoid breaking changes; however in this case, a “break” would almost certainly indicate an undetected bug in the C# 4.0 program rather than intentional reliance on the old behavior. Anonymous Methods Anonymous methods are a C# 2.0 feature that has been mostly subsumed by C# 3.0 lambda expressions. An anonymous method is like a lambda expression, but it lacks the following features: • Implicitly typed parameters • Expression syntax (an anonymous method must always be a statement block) • The ability to compile to an expression tree, by assigning to Expression To write an anonymous method, you include the delegate keyword followed (op- tionally) by a parameter declaration and then a method body. For example, given this delegate: delegate int Transformer (int i); we could write and call an anonymous method as follows: Transformer sqr = delegate (int x) {return x * x;}; Console.WriteLine (sqr(3)); // 9 The first line is semantically equivalent to the following lambda expression: Transformer sqr = (int x) => {return x * x;}; Or simply: Transformer sqr = x => x * x; Anonymous methods capture outer variables in the same way lambda expressions do. Anonymous Methods | 139 Advanced C# A unique feature of anonymous methods is that you can omit the parameter declaration entirely—even if the delegate expects them. This can be useful in declaring events with a default empty handler: public event EventHandler Clicked = delegate { }; This avoids the need for a null check before firing the event. The following is also legal: // Notice that we omit the parameters: Clicked += delegate { Console.WriteLine ("clicked"); }; try Statements and Exceptions A try statement specifies a code block subject to error-handling or cleanup code. The try block must be followed by a catch block, a finally block, or both. The catch block executes when an error occurs in the try block. The finally block ex- ecutes after execution leaves the try block (or if present, the catch block), to perform cleanup code, whether or not an error occurred. A catch block has access to an Exception object that contains information about the error. You use a catch block to either compensate for the error or rethrow the exception. You rethrow an exception if you merely want to log the problem, or if you want to rethrow a new, higher-level exception type. A finally block adds determinism to your program: the CLR endeavors to always execute it. It’s useful for cleanup tasks such as closing network connections. A try statement looks like this: try { ... // exception may get thrown within execution of this block } catch (ExceptionA ex) { ... // handle exception of type ExceptionA } catch (ExceptionB ex) { ... // handle exception of type ExceptionB } finally { ... // cleanup code } Consider the following program: class Test { static int Calc (int x) { return 10 / x; } static void Main() 140 | Chapter 4: Advanced C# { int y = Calc (0); Console.WriteLine (y); } } Because x is zero, the runtime throws a DivideByZeroException, and our program terminates. We can prevent this by catching the exception as follows: class Test { static int Calc (int x) { return 10 / x; } static void Main() { try { int y = Calc (0); Console.WriteLine (y); } catch (DivideByZeroException ex) { Console.WriteLine ("x cannot be zero"); } Console.WriteLine ("program completed"); } } OUTPUT: x cannot be zero program completed This is a simple example to illustrate exception handling. We could deal with this particular scenario better in practice by checking explicitly for the divisor being zero before calling Calc. Exceptions are relatively expensive to handle, taking hundreds of clock cycles. When an exception is thrown, the CLR performs a test: Is execution currently within a try statement that can catch the exception? • If so, execution is passed to the compatible catch block. If the catch block suc- cessfully finishes executing, execution moves to the next statement after the try statement (if present, executing the finally block first). • If not, execution jumps back to the caller of the function, and the test is repeated (after executing any finally blocks that wrap the statement). If no function takes responsibility for the exception, an error dialog box is displayed to the user, and the program terminates. try Statements and Exceptions | 141 Advanced C# The catch Clause A catch clause specifies what type of exception to catch. This must either be System.Exception or a subclass of System.Exception. Catching System.Exception catches all possible errors. This is useful when: • Your program can potentially recover regardless of the specific exception type. • You plan to rethrow the exception (perhaps after logging it). • Your error handler is the last resort, prior to termination of the program. More typically, though, you catch specific exception types, in order to avoid having to deal with circumstances for which your handler wasn’t designed (e.g., an OutOf MemoryException). You can handle multiple exception types with multiple catch clauses (again, this example could be written with explicit argument checking rather than exception handling): class Test { static void Main (string[] args) { try { byte b = byte.Parse (args[0]); Console.WriteLine (b); } catch (IndexOutOfRangeException ex) { Console.WriteLine ("Please provide at least one argument"); } catch (FormatException ex) { Console.WriteLine ("That's not a number!"); } catch (OverflowException ex) { Console.WriteLine ("You've given me more than a byte!"); } } } Only one catch clause executes for a given exception. If you want to include a safety net to catch more general exceptions (such as System.Exception) you must put the more specific handlers first. An exception can be caught without specifying a variable, if you don’t need to access its properties: catch (StackOverflowException) // no variable { ... } 142 | Chapter 4: Advanced C# Furthermore, you can omit both the variable and the type (meaning that all excep- tions will be caught): catch { ... } In C++, it is possible (though not recommended) to throw an object that does not derive from Exception. The CLR automat- ically wraps that object in a RuntimeWrappedException class (which does derive from Exception). The finally Block A finally block always executes—whether or not an exception is thrown and whether or not the try block runs to completion. finally blocks are typically used for cleanup code. A finally block executes either: • After a catch block finishes • After control leaves the try block because of a jump statement (e.g., return or goto) • After the try block ends The only things that can defeat a finally block are an infinite loop, or the process ending abruptly. A finally block helps add determinism to a program. In the following example, the file that we open always gets closed, regardless of whether: • The try block finishes normally. • Execution returns early because the file is empty (EndOfStream). • An IOException is thrown while reading the file. static void ReadFile() { StreamReader reader = null; // In System.IO namespace try { reader = File.OpenText ("file.txt"); if (reader.EndOfStream) return; Console.WriteLine (reader.ReadToEnd()); } finally { if (reader != null) reader.Dispose(); } } In this example, we closed the file by calling Dispose on the StreamReader. Calling Dispose on an object, within a finally block, is a standard convention throughout the .NET Framework and is supported explicitly in C# through the using statement. try Statements and Exceptions | 143 Advanced C# The using statement Many classes encapsulate unmanaged resources, such as file handles, graphics han- dles, or database connections. These classes implement System.IDisposable, which defines a single parameterless method named Dispose to clean up these resources. The using statement provides an elegant syntax for calling Dispose on an IDisposa ble object within a finally block. The following: using (StreamReader reader = File.OpenText ("file.txt")) { ... } is precisely equivalent to: StreamReader reader = File.OpenText ("file.txt"); try { ... } finally { if (reader != null) ((IDisposable)reader).Dispose(); } We cover the disposal pattern in more detail in Chapter 12. Throwing Exceptions Exceptions can be thrown either by the runtime or in user code. In this example, Display throws a System.ArgumentNullException: class Test { static void Display (string name) { if (name == null) throw new ArgumentNullException ("name"); Console.WriteLine (name); } static void Main() { try { Display (null); } catch (ArgumentNullException ex) { Console.WriteLine ("Caught the exception"); } } } 144 | Chapter 4: Advanced C# Rethrowing an exception You can capture and rethrow an exception as follows: try { ... } catch (Exception ex) { // Log error ... throw; // Rethrow same exception } If we replaced throw with throw ex, the example would still work, but the StackTrace property of the newly propagated exception would no longer reflect the original error. Rethrowing in this manner lets you log an error without swallowing it. It also lets you back out of handling an exception should circumstances turn out to be outside what you expected: using System.Net; // (See Chapter 16) ... string s = null; using (WebClient wc = new WebClient()) try { s = wc.DownloadString ("http://www.albahari.com/nutshell/"); } catch (WebException ex) { if (ex.Status == WebExceptionStatus.NameResolutionFailure) Console.WriteLine ("Bad domain name"); else throw; // Can't handle other sorts of WebException, so rethrow } The other common scenario is to rethrow a more specific exception type. For example: try { ... // Parse a DateTime from XML element data } catch (FormatException ex) { throw new XmlException ("Invalid DateTime", ex); } Rethrowing a less specific exception is something you might do when crossing a trust boundary, so as not to leak technical information to potential hackers. When rethrowing a different exception, you can set the InnerException property with the original exception to aid debugging. Nearly all types of exceptions provide a constructor for this purpose (such as in our example). try Statements and Exceptions | 145 Advanced C# Key Properties of System.Exception The most important properties of System.Exception are the following: StackTrace A string representing all the methods that are called from the origin of the exception to the catch block. Message A string with a description of the error. InnerException The inner exception (if any) that caused the outer exception. This, itself, may have another InnerException. All exceptions in C# are runtime exceptions—there is no equiv- alent to Java’s compile-time checked exceptions. Common Exception Types The following exception types are used widely throughout the CLR and .NET Framework. You can throw these yourself or use them as base classes for deriving custom exception types. System.ArgumentException Thrown when a function is called with a bogus argument. This generally indi- cates a program bug. System.ArgumentNullException Subclass of ArgumentException that’s thrown when a function argument is (un- expectedly) null. System.ArgumentOutOfRangeException Subclass of ArgumentException that’s thrown when a (usually numeric) argu- ment is too big or too small. For example, this is thrown when passing a negative number into a function that accepts only positive values. System.InvalidOperationException Thrown when the state of an object is unsuitable for a method to successfully execute, regardless of any particular argument values. Examples include reading an unopened file or getting the next element from an enumerator where the underlying list has been modified partway through the iteration. System.NotSupportedException Thrown to indicate that a particular functionality is not supported. A good example is calling the Add method on a collection for which IsReadOnly returns true. System.NotImplementedException Thrown to indicate that a function has not yet been implemented. 146 | Chapter 4: Advanced C# System.ObjectDisposedException Thrown when the object upon which the function is called has been disposed. Another commonly encountered exception type is NullReferenceException. The CLR throws this exception when you attempt to access a member of an object whose value is null (indicating a bug in your code). You can throw a NullReferenceExcep tion directly (for testing purposes) as follows: throw null; The TryXXX Method Pattern When writing a method, you have a choice, when something goes wrong, to return some kind of failure code or throw an exception. In general, you throw an exception when the error is outside the normal workflow—or if you expect that the immediate caller won’t be able to cope with it. Occasionally, though, it can be best to offer both choices to the consumer. An example of this is the int type, which defines two versions of its Parse method: public int Parse (string input); public bool TryParse (string input, out int returnValue); If parsing fails, Parse throws an exception; TryParse returns false. You can implement this pattern by having the XXX method call the TryXXX method as follows: public return-type XXX (input-type input) { return-type returnValue; if (!TryXXX (input, out returnValue)) throw new YYYException (...) return returnValue; } Alternatives to Exceptions As with int.TryParse, a function can communicate failure by sending an error code back to the calling function via a return type or parameter. Although this can work with simple and predictable failures, it becomes clumsy when extended to all errors, polluting method signatures and creating unnecessary complexity and clutter. It also cannot generalize to functions that are not methods, such as operators (e.g., the division operator) or properties. An alternative is to place the error in a common place where all functions in the call stack can see it (e.g., a static method that stores the current error per thread). This, though, requires each function to participate in an error-propagation pattern that is cumbersome and, ironically, itself error-prone. try Statements and Exceptions | 147 Advanced C# Enumeration and Iterators Enumeration An enumerator is a read-only, forward-only cursor over a sequence of values. An enumerator is an object that implements either of the following interfaces: • System.Collections.IEnumerator • System.Collections.Generic.IEnumerator Technically, any object that has a method named MoveNext and a property called Current is treated as an enumerator. This relaxation was introduced in C# 1.0 to avoid the boxing/un- boxing overhead when enumerating value type elements, but was made redundant when generics were introduced in C# 2. The foreach statement iterates over an enumerable object. An enumerable object is the logical representation of a sequence. It is not itself a cursor, but an object that produces cursors over itself. An enumerable object either: • Implements IEnumerable or IEnumerable • Has a method named GetEnumerator that returns an enumerator IEnumerator and IEnumerable are defined in System.Collec tions. IEnumerator and IEnumerable are defined in System.Collections.Generic. The enumeration pattern is as follows: class Enumerator // Typically implements IEnumerator or IEnumerator { public IteratorVariableType Current { get {...} } public bool MoveNext() {...} } class Enumerable // Typically implements IEnumerable or IEnumerable { public Enumerator GetEnumerator() {...} } Here is the high-level way of iterating through the characters in the word beer using a foreach statement: foreach (char c in "beer") Console.WriteLine (c); Here is the low-level way of iterating through the characters in beer without using a foreach statement: 148 | Chapter 4: Advanced C# using (var enumerator = "beer".GetEnumerator()) while (enumerator.MoveNext()) { var element = enumerator.Current; Console.WriteLine (element); } If the enumerator implements IDisposable, the foreach statement also acts as a using statement, implicitly disposing the enumerator object. Chapter 7 explains the enumeration interfaces in further detail. Collection Initializers You can instantiate and populate an enumerable object in a single step. For example: using System.Collections.Generic; ... List list = new List {1, 2, 3}; The compiler translates this to the following: using System.Collections.Generic; ... List list = new List(); list.Add (1); list.Add (2); list.Add (3); This requires that the enumerable object implements the System.Collec tions.IEnumerable interface, and that it has an Add method that has the appropriate number of parameters for the call. Iterators Whereas a foreach statement is a consumer of an enumerator, an iterator is a pro- ducer of an enumerator. In this example, we use an iterator to return a sequence of Fibonacci numbers (where each number is the sum of the previous two): using System; using System.Collections.Generic; class Test { static void Main() { foreach (int fib in Fibs(6)) Console.Write (fib + " "); } static IEnumerable Fibs (int fibCount) { for (int i = 0, prevFib = 1, curFib = 1; i < fibCount; i++) { Enumeration and Iterators | 149 Advanced C# yield return prevFib; int newFib = prevFib+curFib; prevFib = curFib; curFib = newFib; } } } OUTPUT: 1 1 2 3 5 8 Whereas a return statement expresses “Here’s the value you asked me to return from this method,” a yield return statement expresses “Here’s the next element you asked me to yield from this enumerator.” On each yield statement, control is returned to the caller, but the callee’s state is maintained so that the method can continue executing as soon as the caller enumerates the next element. The lifetime of this state is bound to the enumerator, such that the state can be released when the caller has finished enumerating. The compiler converts iterator methods into private classes that implement IEnumerable and/or IEnumerator. The logic within the iterator block is “inverted” and spliced into the Move Next method and Current property on the compiler-written enu- merator class. This means that when you call an iterator method, all you’re doing is instantiating the compiler-written class; none of your code actually runs! Your code runs only when you start enumerating over the resultant sequence, typi- cally with a foreach statement. Iterator Semantics An iterator is a method, property, or indexer that contains one or more yield state- ments. An iterator must return one of the following four interfaces (otherwise, the compiler will generate an error): // Enumerable interfaces System.Collections.IEnumerable System.Collections.Generic.IEnumerable // Enumerator interfaces System.Collections.IEnumerator System.Collections.Generic.IEnumerator An iterator has different semantics, depending on whether it returns an enumera- ble interface or an enumerator interface. We describe this in Chapter 7. Multiple yield statements are permitted. For example: class Test { static void Main() { foreach (string s in Foo()) Console.WriteLine(s); // Prints "One","Two","Three" 150 | Chapter 4: Advanced C# } static IEnumerable Foo() { yield return "One"; yield return "Two"; yield return "Three"; } } yield break The yield break statement indicates that the iterator block should exit early, without returning more elements. We can modify Foo as follows to demonstrate: static IEnumerable Foo (bool breakEarly) { yield return "One"; yield return "Two"; if (breakEarly) yield break; yield return "Three"; } A return statement is illegal in an iterator block—you must use a yield break instead. Iterators and try/catch/finally blocks A yield return statement cannot appear in a try block that has a catch clause: IEnumerable Foo() { try { yield return "One"; } // Illegal catch { ... } } Nor can yield return appear in a catch or finally block. These restrictions are due to the fact that the compiler must translate iterators into ordinary classes with Move Next, Current, and Dispose members, and translating exception handling blocks would create excessive complexity. You can, however, yield within a try block that has (only) a finally block: IEnumerable Foo() { try { yield return "One"; } // OK finally { ... } } Enumeration and Iterators | 151 Advanced C# The code in the finally block executes when the consuming enumerator reaches the end of the sequence or is disposed. A foreach statement implicitly disposes the enumerator if you break early, making this a safe way to consume enumerators. When working with enumerators explicitly, a trap is to abandon enumeration early without disposing it, circumventing the finally block. You can avoid this risk by wrapping explicit use of enumerators in a using statement: string firstElement = null; var sequence = Foo(); using (var enumerator = sequence.GetEnumerator()) if (enumerator.MoveNext()) firstElement = enumerator.Current; Composing Sequences Iterators are highly composable. We can extend our example, this time to output even Fibonacci numbers only: using System; using System.Collections.Generic; class Test { static void Main() { foreach (int fib in EvenNumbersOnly (Fibs(6))) Console.WriteLine (fib); } static IEnumerable Fibs (int fibCount) { for (int i = 0, prevFib = 1, curFib = 1; i < fibCount; i++) { yield return prevFib; int newFib = prevFib+curFib; prevFib = curFib; curFib = newFib; } } static IEnumerable EvenNumbersOnly (IEnumerable sequence) { foreach (int x in sequence) if ((x % 2) == 0) yield return x; } } Each element is not calculated until the last moment—when requested by a Move Next() operation. Figure 4-1 shows the data requests and data output over time. 152 | Chapter 4: Advanced C# Figure 4-1. Composing sequences The composability of the iterator pattern is extremely useful in LINQ; we discuss the subject again in Chapter 8. Nullable Types Reference types can represent a nonexistent value with a null reference. Value types, however, cannot ordinarily represent null values. For example: string s = null; // OK, Reference Type int i = null; // Compile Error, Value Type cannot be null To represent null in a value type, you must use a special construct called a nullable type. A nullable type is denoted with a value type followed by the ? symbol: int? i = null; // OK, Nullable Type Console.WriteLine (i == null); // True Nullable Struct T? translates into System.Nullable. Nullable is a lightweight immutable struc- ture, having only two fields, to represent Value and HasValue. The essence of Sys tem.Nullable is very simple: public struct Nullable where T : struct { public T Value {get;} public bool HasValue {get;} public T GetValueOrDefault(); Nullable Types | 153 Advanced C# public T GetValueOrDefault (T defaultValue); ... } The code: int? i = null; Console.WriteLine (i == null); // True translates to: Nullable i = new Nullable(); Console.WriteLine (! i.HasValue); // True Attempting to retrieve Value when HasValue is false throws an InvalidOperationEx ception. GetValueOrDefault() returns Value if HasValue is true; otherwise, it returns new T() or a specified custom default value. The default value of T? is null. Implicit and explicit nullable conversions The conversion from T to T? is implicit, and from T? to T is explicit. For example: int? x = 5; // implicit int y = (int)x; // explicit The explicit cast is directly equivalent to calling the nullable object’s Value property. Hence, an InvalidOperationException is thrown if HasValue is false. Boxing and unboxing nullable values When T? is boxed, the boxed value on the heap contains T, not T?. This optimization is possible because a boxed value is a reference type that can already express null. C# also permits the unboxing of nullable types with the as operator. The result will be null if the cast fails: object o = "string"; int? x = o as int?; Console.WriteLine (x.HasValue); // False Operator Lifting The Nullable struct does not define operators such as <, >, or even ==. Despite this, the following code compiles and executes correctly: int? x = 5; int? y = 10; bool b = x < y; // true This works because the compiler borrows or “lifts” the less-than operator from the underlying value type. Semantically, it translates the preceding comparison expres- sion into this: bool b = (x.HasValue && y.HasValue) ? (x.Value < y.Value) : false; 154 | Chapter 4: Advanced C# In other words, if both x and y have values, it compares via int’s less-than operator; otherwise, it returns false. Operator lifting means you can implicitly use T’s operators on T?. You can define operators for T? in order to provide special-purpose null behavior, but in the vast majority of cases, it’s best to rely on the compiler automatically applying systematic nullable logic for you. Here are some examples: int? x = 5; int? y = null; // Equality operator examples Console.WriteLine (x == y); // False Console.WriteLine (x == null); // False Console.WriteLine (x == 5); // True Console.WriteLine (y == null); // True Console.WriteLine (y == 5); // False Console.WriteLine (y != 5); // True // Relational operator examples Console.WriteLine (x < 6); // True Console.WriteLine (y < 6); // False Console.WriteLine (y > 6); // False // All other operator examples Console.WriteLine (x + 5); // 10 Console.WriteLine (x + y); // null (prints empty line) The compiler performs null logic differently depending on the category of operator. The following sections explain these different rules. Equality operators (== and !=) Lifted equality operators handle nulls just like reference types do. This means two null values are equal: Console.WriteLine ( null == null); // True Console.WriteLine ((bool?)null == (bool?)null); // True Further: • If exactly one operand is null, the operands are unequal. • If both operands are non-null, their Values are compared. Relational operators (<, <=, >=, >) The relational operators work on the principle that it is meaningless to compare null operands. This means comparing a null value to either a null or a non-null value returns false. bool b = x < y; // Translation: bool b = (x.HasValue && y.HasValue) ? (x.Value < y.Value) : false; // b is false (assuming x is 5 and y is null) Nullable Types | 155 Advanced C# All other operators (+, −, *, /, %, &, |, ^, <<, >>, +, ++, --, !, ~) These operators return null when any of the operands are null. This pattern should be familiar to SQL users. int? c = x + y; // Translation: int? c = (x.HasValue && y.HasValue) ? (int?) (x.Value + y.Value) : null; // c is null (assuming x is 5 and y is null) An exception is when the & and | operators are applied to bool?, which we will discuss shortly. Mixing nullable and non-nullable operators You can mix and match nullable and non-nullable types (this works because there is an implicit conversion from T to T?): int? a = null; int b = 2; int? c = a + b; // c is null - equivalent to a + (int?)b bool? with & and | Operators When supplied operands of type bool? the & and | operators treat null as an unknown value. So, null | true is true, because: • If the unknown value is false, the result would be true. • If the unknown value is true, the result would be true. Similarly, null & false is false. This behavior would be familiar to SQL users. The following example enumerates other combinations: bool? n = null; bool? f = false; bool? t = true; Console.WriteLine (n | n); // (null) Console.WriteLine (n | f); // (null) Console.WriteLine (n | t); // True Console.WriteLine (n & n); // (null) Console.WriteLine (n & f); // False Console.WriteLine (n & t); // (null) Null Coalescing Operator The ?? operator is the null coalescing operator, and it can be used with both nullable types and reference types. It says “If the operand is non-null, give it to me; otherwise, give me a default value.” For example: int? x = null; int y = x ?? 5; // y is 5 156 | Chapter 4: Advanced C# int? a = null, b = 1, c = 2; Console.WriteLine (a ?? b ?? c); // 1 (first non-null value) The ?? operator is equivalent to calling GetValueOrDefault with an explicit default value, except that the expression for the default value is never evaluated if the vari- able is not null. Scenarios for Nullable Types One of the most common scenarios for nullable types is to represent unknown val- ues. This frequently occurs in database programming, where a class is mapped to a table with nullable columns. If these columns are strings (e.g., an EmailAddress column on a Customer table), there is no problem, as string is a reference type in the CLR, which can be null. However, most other SQL column types map to CLR struct types, making nullable types very useful when mapping SQL to the CLR. For example: // Maps to a Customer table in a database public class Customer { ... public decimal? AccountBalance; } A nullable type can also be used to represent the backing field of what’s sometimes called an ambient property. An ambient property, if null, returns the value of its parent. For example: public class Row { ... Grid parent; Color? color; public Color Color { get { return color ?? parent.Color; } set { color = value == parent.Color ? (Color?)null : value; } } } Alternatives to Nullable Types Before nullable types were part of the C# language (i.e., before C# 2.0), there were many strategies to deal with nullable value types, examples of which still appear in the .NET Framework for historical reasons. One of these strategies is to designate a particular non-null value as the “null value”; an example is in the string and array classes. String.IndexOf returns the magic value of −1 when the character is not found: int i = "Pink".IndexOf ('b'); Console.WriteLine (i); // −1 Nullable Types | 157 Advanced C# However, Array.IndexOf returns −1 only if the index is 0-bounded. The more general formula is that IndexOf returns 1 less than the lower bound of the array. In the next example, IndexOf returns 0 when an element is not found: // Create an array whose lower bound is 1 instead of 0: Array a = Array.CreateInstance (typeof (string), new int[] {2}, new int[] {1}); a.SetValue ("a", 1); a.SetValue ("b", 2); Console.WriteLine (Array.IndexOf (a, "c")); // 0 Nominating a “magic value” is problematic for several reasons: • It means that each value type has a different representation of null. In contrast, nullable types provide one common pattern that works for all value types. • There may be no reasonable designated value. In the previous example, −1 could not always be used. The same is true for our earlier example representing an unknown account balance. • Forgetting to test for the magic value results in an incorrect value that may go unnoticed until later in execution—when it pulls an unintended magic trick. Forgetting to test HasValue on a null value, however, throws an InvalidOpera tionException on the spot. • The ability for a value to be null is not captured in the type. Types communicate the intention of a program, allow the compiler to check for correctness, and enable a consistent set of rules enforced by the compiler. Operator Overloading Operators can be overloaded to provide more natural syntax for custom types. Operator overloading is most appropriately used for implementing custom structs that represent fairly primitive data types. For example, a custom numeric type is an excellent candidate for operator overloading. The following symbolic operators can be overloaded: + (unary) − (unary) ! ˜ ++ −− + − * / % & | ^ << >> == != > < >= <= The following operators are also overloadable: • Implicit and explicit conversions (with the implicit and explicit keywords) • true and false 158 | Chapter 4: Advanced C# The following operators are indirectly overloaded: • The compound assignment operators (e.g., +=, /=) are implicitly overridden by overriding the noncompound operators (e.g., +, /). • The conditional operators && and || are implicitly overridden by overriding the bitwise operators & and |. Operator Functions An operator is overloaded by declaring an operator function. An operator function has the following rules: • The name of the function is specified with the operator keyword followed by an operator symbol. • The operator function must be marked static and public. • The parameters of the operator function represent the operands. • The return type of an operator function represents the result of an expression. • At least one of the operands must be the type in which the operator function is declared. In the following example, we define a struct called Note representing a musical note, and then overload the + operator: public struct Note { int value; public Note (int semitonesFromA) { value = semitonesFromA; } public static Note operator + (Note x, int semitones) { return new Note (x.value + semitones); } } This overload allows us to add an int to a Note: Note B = new Note (2); Note CSharp = B + 2; Overloading an assignment operator automatically supports the corresponding compound assignment operator. In our example, since we overrode +, we can use += too: CSharp += 2; Overloading Equality and Comparison Operators Equality and comparison operators are sometimes overridden when writing structs, and in rare cases when writing classes. Special rules and obligations come with overloading the equality and comparison operators, which we explain in Chapter 6. A summary of these rules is as follows: Operator Overloading | 159 Advanced C# Pairing The C# compiler enforces operators that are logical pairs to both be defined. These operators are (== !=), (< >), and (<= >=). Equals and GetHashCode In most cases, if you overload (==) and (!=), you will usually need to override the Equals and GetHashCode methods defined on object in order to get mean- ingful behavior. The C# compiler will give a warning if you do not do this. (See “Equality Comparison” on page 254 in Chapter 6 for more details.) IComparable and IComparable If you overload (< >) and (<= >=), you should implement IComparable and IComparable. Custom Implicit and Explicit Conversions Implicit and explicit conversions are overloadable operators. These conversions are typically overloaded to make converting between strongly related types (such as numeric types) concise and natural. To convert between weakly related types, the following strategies are more suitable: • Write a constructor that has a parameter of the type to convert from. • Write ToXXX and (static) FromXXX methods to convert between types. As explained in the discussion on types, the rationale behind implicit conversions is that they are guaranteed to succeed and not lose information during the conver- sion. Conversely, an explicit conversion should be required either when runtime circumstances will determine whether the conversion will succeed or if information may be lost during the conversion. In this example, we define conversions between our musical Note type and a double (which represents the frequency in hertz of that note): ... // Convert to hertz public static implicit operator double (Note x) { return 440 * Math.Pow (2, (double) x.value / 12 ); } // Convert from hertz (accurate to the nearest semitone) public static explicit operator Note (double x) { return new Note ((int) (0.5 + 12 * (Math.Log (x/440) / Math.Log(2) ) )); } ... Note n = (Note)554.37; // explicit conversion double x = n; // implicit conversion 160 | Chapter 4: Advanced C# Following our own guidelines, this example might be better implemented with a ToFrequency method (and a static FromFre quency method) instead of implicit and explicit operators. Custom conversions are ignored by the as and is operators: Console.WriteLine (554.37 is Note); // False Note n = 554.37 as Note; // Error Overloading true and false The true and false operators are overloaded in the extremely rare case of types that are Boolean “in spirit,” but do not have a conversion to bool. An example is a type that implements three-state logic: by overloading true and false, such a type can work seamlessly with conditional statements and operators—namely, if, do, while, for, &&, ||, and ?:. The System.Data.SqlTypes.SqlBoolean struct provides this func- tionality. For example: SqlBoolean a = SqlBoolean.Null; if (a) Console.WriteLine ("True"); else if (!a) Console.WriteLine ("False"); else Console.WriteLine ("Null"); OUTPUT: Null The following code is a reimplementation of the parts of SqlBoolean necessary to demonstrate the true and false operators: public struct SqlBoolean { public static bool operator true (SqlBoolean x) { return x.m_value == True.m_value; } public static bool operator false (SqlBoolean x) { return x.m_value == False.m_value; } public static SqlBoolean operator ! (SqlBoolean x) { if (x.m_value == Null.m_value) return Null; if (x.m_value == False.m_value) return True; return False; } public static readonly SqlBoolean Null = new SqlBoolean(0); public static readonly SqlBoolean False = new SqlBoolean(1); Operator Overloading | 161 Advanced C# public static readonly SqlBoolean True = new SqlBoolean(2); private SqlBoolean (byte value) { m_value = value; } private byte m_value; } Extension Methods Extension methods allow an existing type to be extended with new methods without altering the definition of the original type. An extension method is a static method of a static class, where the this modifier is applied to the first parameter. The type of the first parameter will be the type that is extended. For example: public static class StringHelper { public static bool IsCapitalized (this string s) { if (string.IsNullOrEmpty(s)) return false; return char.IsUpper (s[0]); } } The IsCapitalized extension method can be called as though it were an instance method on a string, as follows: Console.WriteLine ("Perth".IsCapitalized()); An extension method call, when compiled, is translated back into an ordinary static method call: Console.WriteLine (StringHelper.IsCapitalized ("Perth")); The translation works as follows: arg0.Method (arg1, arg2, ...); // Extension method call StaticClass.Method (arg0, arg1, arg2, ...); // Static method call Interfaces can be extended too: public static T First (this IEnumerable sequence) { foreach (T element in sequence) return element; throw new InvalidOperationException ("No elements!"); } ... Console.WriteLine ("Seattle".First()); // S Extension methods were added in C# 3.0. Extension Method Chaining Extension methods, like instance methods, provide a tidy way to chain functions. Consider the following two functions: 162 | Chapter 4: Advanced C# public static class StringHelper { public static string Pluralize (this string s) {...} public static string Capitalize (this string s) {...} } x and y are equivalent and both evaluate to "Sausages", but x uses extension methods, whereas y uses static methods: string x = "sausage".Pluralize().Capitalize(); string y = StringHelper.Capitalize (StringHelper.Pluralize ("sausage")); Ambiguity and Resolution Namespaces An extension method cannot be accessed unless its class is in scope, typically by its namespace being imported. Consider the extension method IsCapitalized in the following example: using System; namespace Utils { public static class StringHelper { public static bool IsCapitalized (this string s) { if (string.IsNullOrEmpty(s)) return false; return char.IsUpper (s[0]); } } } To use IsCapitalized, the following application must import Utils, in order to avoid a compile-time error: namespace MyApp { using Utils; class Test { static void Main() { Console.WriteLine ("Perth".IsCapitalized()); } } } Extension methods versus instance methods Any compatible instance method will always take precedence over an extension method. In the following example, Test’s Foo method will always take precedence —even when called with an argument x of type int: Extension Methods | 163 Advanced C# class Test { public void Foo (object x) { } // This method always wins } static class Extensions { public static void Foo (this Test t, int x) { } } The only way to call the extension method in this case is via normal static syntax; in other words, Extensions.Foo(...). Extension methods versus extension methods If two extension methods have the same signature, the extension method must be called as an ordinary static method to disambiguate the method to call. If one extension method has more specific arguments, however, the more specific method takes precedence. To illustrate, consider the following two classes: static class StringHelper { public static bool IsCapitalized (this string s) {...} } static class ObjectHelper { public static bool IsCapitalized (this object s) {...} } The following code calls StringHelper’s IsCapitalized method: bool test1 = "Perth".IsCapitalized(); To call ObjectHelper’s IsCapitalized method, we must specify it explicitly: bool test2 = (ObjectHelper.IsCapitalized ("Perth")); Classes and structs are considered more specific than interfaces. Anonymous Types An anonymous type is a simple class created by the compiler on the fly to store a set of values. To create an anonymous type, use the new keyword followed by an object initializer, specifying the properties and values the type will contain. For example: var dude = new { Name = "Bob", Age = 23 }; The compiler translates this to (approximately) the following: internal class AnonymousGeneratedTypeName { private string name; // Actual field name is irrelevant private int age; // Actual field name is irrelevant public AnonymousGeneratedTypeName (string name, int age) 164 | Chapter 4: Advanced C# { this.name = name; this.age = age; } public string Name { get { return name; } } public int Age { get { return age; } } // The Equals and GetHashCode methods are overridden (see Chapter 6). // The ToString method is also overridden. } ... var dude = new AnonymousGeneratedTypeName ("Bob", 23); You must use the var keyword to reference an anonymous type, because it doesn’t have a name. The property name of an anonymous type can be inferred from an expression that is itself an identifier (or ends with one). For example: int Age = 23; var dude = new { Name = "Bob", Age, Age.ToString().Length }; is equivalent to: var dude = new { Name = "Bob", Age = Age, Length = Age.ToString().Length }; Two anonymous type instances declared within the same assembly will have the same underlying type if their elements are named and typed identically: var a1 = new { X = 2, Y = 4 }; var a2 = new { X = 2, Y = 4 }; Console.WriteLine (a1.GetType() == a2.GetType()); // True Additionally, the Equals method is overridden to perform equality comparisons: Console.WriteLine (a1 == a2); // False Console.WriteLine (a1.Equals (a2)); // True You can create arrays of anonymous types as follows: var dudes = new[] { new { Name = "Bob", Age = 30 }, new { Name = "Tom", Age = 40 } }; Anonymous types are used primarily when writing LINQ queries (see Chapter 8), and were added in C# 3.0. Dynamic Binding Dynamic binding defers binding—the process of resolving types, members, and operations—from compile time to runtime. Dynamic binding is useful when at compile time you know that a certain function, member, or operation exists, but the compiler does not. This commonly occurs when you are interoperating with dynamic Dynamic Binding | 165 Advanced C# languages (such as IronPython) and COM and in scenarios when you might other- wise use reflection. A dynamic type is declared with the contextual keyword dynamic: dynamic d = GetSomeObject(); d.Quack(); A dynamic type tells the compiler to relax. We expect the runtime type of d to have a Quack method. We just can’t prove it statically. Since d is dynamic, the compiler defers binding Quack to d until runtime. To understand what this means requires distinguishing between static binding and dynamic binding. Static Binding Versus Dynamic Binding The canonical binding example is mapping a name to a specific function when compiling an expression. To compile the following expression, the compiler needs to find the implementation of the method named Quack: d.Quack(); Let’s suppose the static type of d is Duck: Duck d = ... d.Quack(); In the simplest case, the compiler does the binding by looking for a parameterless method named Quack on Duck. Failing that, the compiler extends its search to meth- ods taking optional parameters, methods on base classes of Duck, and extension methods that take Duck as its first parameter. If no match is found, you’ll get a com- pilation error. Regardless of what method gets bound, the bottom line is that the binding is done by the compiler, and the binding utterly depends on statically knowing the types of the operands (in this case, d). This makes it static binding. Now let’s change the static type of d to object: object d = ... d.Quack(); Calling Quack gives us a compilation error, because although the value stored in d can contain a method called Quack, the compiler cannot know it since the only information it has is the type of the variable, which in this case is object. But let’s now change the static type of d to dynamic: dynamic d = ... d.Quack(); A dynamic type is like object—it’s equally nondescriptive about a type. The differ- ence is that it lets you use it in ways that aren’t known at compile time. A dynamic object binds at runtime based on its runtime type, not its compile-time type. When the compiler sees a dynamically bound expression (which in general is an expression that contains any value of type dynamic), it merely packages up the expression such that the binding can be done later at runtime. 166 | Chapter 4: Advanced C# At runtime, if a dynamic object implements IDynamicMetaObjectProvider, that in- terface is used to perform the binding. If not, binding occurs in almost the same way as it would have had the compiler known the dynamic object’s runtime type. These two alternatives are called custom binding and language binding. COM interop can be considered to use a third kind of dynamic binding (see Chapter 25). Custom Binding Custom binding occurs when a dynamic object implements IDynamicMetaObjectPro vider (IDMOP). Although you can implement IDMOP on types that you write in C#, and that is useful to do, the more common case is that you have acquired an IDMOP object from a dynamic language that is implemented in .NET on the DLR, such as IronPython or IronRuby. Objects from those languages implicitly implement IDMOP as a means by which to directly control the meanings of operations per- formed on them. We will discuss custom binders in greater detail in Chapter 20, but we will write a simple one now to demonstrate the feature: using System; using System.Dynamic; public class Test { static void Main() { dynamic d = new Duck(); d.Quack(); // Quack method was called d.Waddle(); // Waddle method was called } } public class Duck : DynamicObject { public override bool TryInvokeMember ( InvokeMemberBinder binder, object[] args, out object result) { Console.WriteLine (binder.Name + " method was called"); result = null; return true; } } The Duck class doesn’t actually have a Quack method. Instead, it uses custom binding to intercept and interpret all method calls. Dynamic Binding | 167 Advanced C# Language Binding Language binding occurs when a dynamic object does not implement IDynamicMetaObjectProvider. Language binding is useful when working around imperfectly designed types or inherent limitations in the .NET type system (we’ll explore more scenarios in Chapter 20). A typical problem when using numeric types is that they have no common interface. We have seen that methods can be bound dynamically; the same is true for operators: static dynamic Mean (dynamic x, dynamic y) { return (x + y) / 2; } static void Main() { int x = 3, y = 4; Console.WriteLine (Mean (x, y)); } The benefit is obvious—you don’t have to duplicate code for each numeric type. However, you lose static type safety, risking runtime exceptions rather than compile- time errors. Dynamic binding circumvents static type safety, but not run- time type safety. Unlike with reflection (Chapter 19), you can’t circumvent member accessibility rules with dynamic binding. By design, language runtime binding behaves as similarly as possible to static bind- ing, had the runtime types of the dynamic objects been known at compile time. In our previous example, the behavior of our program would be identical if we hard- coded Mean to work with the int type. The most notable exception in parity between static and dynamic binding is for extension methods, which we discuss in “Uncal- lable Functions” on page 172. Dynamic binding also incurs a performance hit. Because of the DLR’s caching mechanisms, however, repeated calls to the same dynamic expression are optimized—allowing you to efficiently call dynamic expressions in a loop. This optimization brings the typical overhead for a simple dynamic expression on today’s hardware down to less than 100 ns. RuntimeBinderException If a member fails to bind, a RuntimeBinderException is thrown. You can think of this like a compile-time error at runtime. dynamic d = 5; d.Hello(); // throws RuntimeBinderException 168 | Chapter 4: Advanced C# The exception is thrown because the int type has no Hello method. Runtime Representation of Dynamic There is a deep equivalence between the dynamic and object types. The runtime treats the following expression as true: typeof (dynamic) == typeof (object) This principle extends to constructed types and array types: typeof (List) == typeof (List) typeof (dynamic[]) == typeof (object[]) Like an object reference, a dynamic reference can point to an object of any type (except pointer types): dynamic x = "hello"; Console.WriteLine (x.GetType().Name); // String x = 123; // No error (despite same variable) Console.WriteLine (x.GetType().Name); // Int32 Structurally, there is no difference between an object reference and a dynamic ref- erence. A dynamic reference simply enables dynamic operations on the object it points to. You can convert from object to dynamic to perform any dynamic operation you want on an object: object o = new System.Text.StringBuilder(); dynamic d = o; d.Append ("hello"); Console.WriteLine (o); // hello Reflecting on a type exposing (public) dynamic members reveals that those members are represented as annotated objects. For example: public class Test { public dynamic Foo; } is equivalent to: public class Test { [System.Runtime.CompilerServices.DynamicAttribute] public object Foo; } This allows consumers of that type to know that Foo should be treated as dynamic, while allowing languages that don’t support dynamic binding to fall back to object. Dynamic Binding | 169 Advanced C# Dynamic Conversions The dynamic type has implicit conversions to and from all other types: int i = 7; dynamic d = i; long j = d; // No cast required (implicit conversion) For the conversion to succeed, the runtime type of the dynamic object must be implicitly convertible to the target static type. The preceding example worked because an int is implicitly convertible to a long. The following example throws a RuntimeBinderException because an int is not implicitly convertible to a short: int i = 7; dynamic d = i; short j = d; // throws RuntimeBinderException var Versus dynamic The var and dynamic types bear a superficial resemblance, but the difference is deep: var says, “Let the compiler figure out the type.” dynamic says, “Let the runtime figure out the type.” To illustrate: dynamic x = "hello"; // Static type is dynamic, runtime type is string var y = "hello"; // Static type is string, runtime type is string int i = x; // Runtime error int j = y; // Compile-time error The static type of a variable declared with var can be dynamic: dynamic x = "hello"; var y = x; // Static type of y is dynamic int z = y; // Runtime error Dynamic Expressions Fields, properties, methods, events, constructors, indexers, operators, and conver- sions can all be called dynamically. Trying to consume the result of a dynamic expression with a void return type is prohibited—just as with a statically typed expression. The difference is that the error occurs at runtime: dynamic list = new List(); var result = list.Add (5); // RuntimeBinderException thrown Expressions involving dynamic operands are typically themselves dynamic, since the effect of absent type information is cascading: dynamic x = 2; var y = x * 3; // Static type of y is dynamic 170 | Chapter 4: Advanced C# There are a couple of obvious exceptions to this rule. First, casting a dynamic expression to a static type yields a static expression: dynamic x = 2; var y = (int)x; // Static type of y is int Second, constructor invocations always yield static expressions—even when called with dynamic arguments. In this example, x is statically typed to a StringBuilder: dynamic capacity = 10; var x = new System.Text.StringBuilder (capacity); In addition, there are a few edge cases where an expression containing a dynamic argument is static, including passing an index to an array and delegate creation expressions. Dynamic Calls without Dynamic Receivers The canonical use case for dynamic involves a dynamic receiver. This means that a dynamic object is the receiver of a dynamic function call: dynamic x = ...; x.Foo(); // x is the receiver However, you can also call statically known functions with dynamic arguments. Such calls are subject to dynamic overload resolution, and can include: • Static methods • Instance constructors • Instance methods on receivers with a statically known type In the following example, the particular Foo that gets dynamically bound is depen- dent on the runtime type of the dynamic argument: class Program { static void Foo (int x) { Console.WriteLine ("1"); } static void Foo (string x) { Console.WriteLine ("2"); } static void Main() { dynamic x = 5; dynamic y = "watermelon"; Foo (x); // 1 Foo (y); // 2 } } Because a dynamic receiver is not involved, the compiler can statically perform a basic check to see whether the dynamic call will succeed. It checks that a function with the right name and number of parameters exists. If no candidate is found, you get a compile-time error. For example: class Program { Dynamic Binding | 171 Advanced C# static void Foo (int x) { Console.WriteLine ("1"); } static void Foo (string x) { Console.WriteLine ("2"); } static void Main() { dynamic x = 5; Foo (x, x); // Compiler error - wrong number of parameters Fook (x); // Compiler error - no such method name } } Static Types in Dynamic Expressions It’s obvious that dynamic types are used in dynamic binding. It’s not so obvious that static types are also used—wherever possible—in dynamic binding. Consider the following: class Program { static void Foo (object x, object y) { Console.WriteLine ("oo"); } static void Foo (object x, string y) { Console.WriteLine ("os"); } static void Foo (string x, object y) { Console.WriteLine ("so"); } static void Foo (string x, string y) { Console.WriteLine ("ss"); } static void Main() { object o = "hello"; dynamic d = "goodbye"; Foo (o, d); // os } } The call to Foo(o,d) is dynamically bound because one of its arguments, d, is dynamic. But since o is statically known, the binding—even though it occurs dy- namically—will make use of that. In this case, overload resolution will pick the second implementation of Foo due to the static type of o and the runtime type of d. In other words, the compiler is “as static as it can possibly be.” Uncallable Functions Some functions cannot be called dynamically. You cannot call: • Extension methods (via extension method syntax) • Members of an interface, if you need to cast to that interface to do so • Base members hidden by a subclass Understanding why this is so is useful in understanding dynamic binding. Dynamic binding requires two pieces of information: the name of the function to call, and the object upon which to call the function. However, in each of the three uncallable scenarios, an additional type is involved, which is known only at compile time. As of C# 5.0, there’s no way to specify these additional types dynamically. 172 | Chapter 4: Advanced C# When calling extension methods, that additional type is implicit. It’s the static class on which the extension method is defined. The compiler searches for it given the using directives in your source code. This makes extension methods compile-time– only concepts, since using directives melt away upon compilation (after they’ve done their job in the binding process in mapping simple names to namespace-qualified names). When calling members via an interface, you specify that additional type via an implicit or explicit cast. There are two scenarios where you might want to do this: when calling explicitly implemented interface members, and when calling interface members implemented in a type internal to another assembly. We can illustrate the former with the following two types: interface IFoo { void Test(); } class Foo : IFoo { void IFoo.Test() {} } To call the Test method, we must cast to the IFoo interface. This is easy with static typing: IFoo f = new Foo(); // Implicit cast to interface f.Test(); Now consider the situation with dynamic typing: IFoo f = new Foo(); dynamic d = f; d.Test(); // Exception thrown The implicit cast shown in bold tells the compiler to bind subsequent member calls on f to IFoo rather than Foo—in other words, to view that object through the lens of the IFoo interface. However, that lens is lost at runtime, so the DLR cannot com- plete the binding. The loss is illustrated as follows: Console.WriteLine (f.GetType().Name); // Foo A similar situation arises when calling a hidden base member: you must specify an additional type via either a cast or the base keyword—and that additional type is lost at runtime. Attributes You’re already familiar with the notion of attributing code elements of a program with modifiers, such as virtual or ref. These constructs are built into the language. Attributes are an extensible mechanism for adding custom information to code ele- ments (assemblies, types, members, return values, parameters, and generic type pa- rameters). This extensibility is useful for services that integrate deeply into the type system, without requiring special keywords or constructs in the C# language. A good scenario for attributes is serialization—the process of converting arbitrary objects to and from a particular format. In this scenario, an attribute on a field can specify the translation between C#’s representation of the field and the format’s representation of the field. Attributes | 173 Advanced C# Attribute Classes An attribute is defined by a class that inherits (directly or indirectly) from the abstract class System.Attribute. To attach an attribute to a code element, specify the attribute’s type name in square brackets, before the code element. For example, the following attaches the ObsoleteAttribute to the Foo class: [ObsoleteAttribute] public class Foo {...} This attribute is recognized by the compiler and will cause compiler warnings if a type or member marked obsolete is referenced. By convention, all attribute types end in the word Attribute. C# recognizes this and allows you to omit the suffix when attaching an attribute: [Obsolete] public class Foo {...} ObsoleteAttribute is a type declared in the System namespace as follows (simplified for brevity): public sealed class ObsoleteAttribute : Attribute {...} The C# language and the .NET Framework include a number of predefined attributes. We describe how to write your own attributes in Chapter 19. Named and Positional Attribute Parameters Attributes may have parameters. In the following example, we apply XmlElementAt tribute to a class. This attribute tells XML serializer (in System.Xml.Serializa tion) how an object is represented in XML and accepts several attribute parame- ters. The following attribute maps the CustomerEntity class to an XML element named Customer, belonging to the http://oreilly.com namespace: [XmlElement ("Customer", Namespace="http://oreilly.com")] public class CustomerEntity { ... } Attribute parameters fall into one of two categories: positional or named. In the preceding example, the first argument is a positional parameter; the second is a named parameter. Positional parameters correspond to parameters of the attribute type’s public constructors. Named parameters correspond to public fields or public properties on the attribute type. When specifying an attribute, you must include positional parameters that corre- spond to one of the attribute’s constructors. Named parameters are optional. In Chapter 19, we describe the valid parameter types and rules for their evaluation. Attribute Targets Implicitly, the target of an attribute is the code element it immediately precedes, which is typically a type or type member. You can also attach attributes, however, to an assembly. This requires that you explicitly specify the attribute’s target. 174 | Chapter 4: Advanced C# Here is an example of using the CLSCompliant attribute to specify CLS compliance for an entire assembly: [assembly:CLSCompliant(true)] Specifying Multiple Attributes Multiple attributes can be specified for a single code element. Each attribute can be listed either within the same pair of square brackets (separated by a comma) or in separate pairs of square brackets (or a combination of the two). The following three examples are semantically identical: [Serializable, Obsolete, CLSCompliant(false)] public class Bar {...} [Serializable] [Obsolete] [CLSCompliant(false)] public class Bar {...} [Serializable, Obsolete] [CLSCompliant(false)] public class Bar {...} Caller Info Attributes (C# 5) From C# 5, you can tag optional parameters with one of three caller info attributes, which instruct the compiler to feed information obtained from the caller’s source code into the parameter’s default value: • [CallerMemberName] applies the caller’s member name • [CallerFilePath] applies the path to caller’s source code file • [CallerLineNumber] applies the line number in caller’s source code file The Foo method in the following program demonstrates all three: using System; using System.Runtime.CompilerServices; class Program { static void Main() { Foo(); } static void Foo ( [CallerMemberName] string memberName = null, [CallerFilePath] string filePath = null, [CallerLineNumber] int lineNumber = 0) { Console.WriteLine (memberName); Console.WriteLine (filePath); Console.WriteLine (lineNumber); } } Caller Info Attributes (C# 5) | 175 Advanced C# Assuming our program resides in c:\source\test\Program.cs, the output would be: Main c:\source\test\Program.cs 8 As with standard optional parameters, the substitution is done at the calling site. Hence, our Main method is syntactic sugar for this: static void Main() { Foo ("Main", @"c:\source\test\Program.cs", 8); } Caller info attributes are useful for logging—and for implementing patterns such as firing a single change notification event whenever any property on an object changes. In fact, there’s a standard interface in the .NET Framework for this called INotify PropertyChanged (in System.ComponentModel): public interface INotifyPropertyChanged { event PropertyChangedEventHandler PropertyChanged; } public delegate void PropertyChangedEventHandler (object sender, PropertyChangedEventArgs e); public class PropertyChangedEventArgs : EventArgs { public PropertyChangedEventArgs (string propertyName); public virtual string PropertyName { get; } } Notice that PropertyChangedEventArgs requires the name of the property that changed. By applying the [CallerMemberName] attribute, however, we can implement this interface and invoke the event without ever specifying property names: public class Foo : INotifyPropertyChanged { public event PropertyChanged = delegate { } void RaisePropertyChanged ([CallerMemberName] string propertyName = null) { PropertyChanged (this, new PropertyChangedEventArgs (propertyName)); } string customerName; public string CustomerName { get { return customerName; } set { if (value == customerName) return; customerName = value; RaisePropertyChanged(); 176 | Chapter 4: Advanced C# // The compiler converts the above line to: // RaisePropertyChanged ("CustomerName"); } } } Unsafe Code and Pointers C# supports direct memory manipulation via pointers within blocks of code marked unsafe and compiled with the /unsafe compiler option. Pointer types are primarily useful for interoperability with C APIs, but may also be used for accessing memory outside the managed heap or for performance-critical hotspots. Pointer Basics For every value type or pointer type V, there is a corresponding pointer type V*. A pointer instance holds the address of a variable. Pointer types can be (unsafely) cast to any other pointer type. The main pointer operators are: Operator Meaning & The address-of operator returns a pointer to the address of a variable * The dereference operator returns the variable at the address of a pointer -> The pointer-to-member operator is a syntactic shortcut, in which x->y is equivalent to (*x).y Unsafe Code By marking a type, type member, or statement block with the unsafe keyword, you’re permitted to use pointer types and perform C++ style pointer operations on memory within that scope. Here is an example of using pointers to quickly process a bitmap: unsafe void BlueFilter (int[,] bitmap) { int length = bitmap.Length; fixed (int* b = bitmap) { int* p = b; for (int i = 0; i < length; i++) *p++ &= 0xFF; } } Unsafe code can run faster than a corresponding safe implementation. In this case, the code would have required a nested loop with array indexing and bounds check- ing. An unsafe C# method may also be faster than calling an external C function, since there is no overhead associated with leaving the managed execution environment. Unsafe Code and Pointers | 177 Advanced C# The fixed Statement The fixed statement is required to pin a managed object, such as the bitmap in the previous example. During the execution of a program, many objects are allocated and deallocated from the heap. In order to avoid unnecessary waste or fragmentation of memory, the garbage collector moves objects around. Pointing to an object is futile if its address could change while referencing it, so the fixed statement tells the garbage collector to “pin” the object and not move it around. This may have an impact on the efficiency of the runtime, so fixed blocks should be used only briefly, and heap allocation should be avoided within the fixed block. Within a fixed statement, you can get a pointer to any value type, an array of value types, or a string. In the case of arrays and strings, the pointer will actually point to the first element, which is a value type. Value types declared inline within reference types require the reference type to be pinned, as follows: class Test { int x; static void Main() { Test test = new Test(); unsafe { fixed (int* p = &test.x) // Pins test { *p = 9; } System.Console.WriteLine (test.x); } } } We describe the fixed statement further in “Mapping a Struct to Unmanaged Mem- ory” on page 979 in Chapter 25. The Pointer-to-Member Operator In addition to the & and * operators, C# also provides the C++ style -> operator, which can be used on structs: struct Test { int x; unsafe static void Main() { Test test = new Test(); Test* p = &test; p->x = 9; System.Console.WriteLine (test.x); } } 178 | Chapter 4: Advanced C# Arrays The stackalloc keyword Memory can be allocated in a block on the stack explicitly using the stackalloc keyword. Since it is allocated on the stack, its lifetime is limited to the execution of the method, just as with any other local variable (whose life hasn’t been extended by virtue of being captured by a lambda expression, iterator block, or asynchronous function). The block may use the [] operator to index into memory: int* a = stackalloc int [10]; for (int i = 0; i < 10; ++i) Console.WriteLine (a[i]); // Print raw memory Fixed-size buffers The fixed keyword has another use, which is to create fixed-size buffers within structs: unsafe struct UnsafeUnicodeString { public short Length; public fixed byte Buffer[30]; // Allocate block of 30 bytes } unsafe class UnsafeClass { UnsafeUnicodeString uus; public UnsafeClass (string s) { uus.Length = (short)s.Length; fixed (byte* p = uus.Buffer) for (int i = 0; i < s.Length; i++) p[i] = (byte) s[i]; } } class Test { static void Main() { new UnsafeClass ("Christian Troy"); } } The fixed keyword is also used in this example to pin the object on the heap that contains the buffer (which will be the instance of UnsafeClass). Hence, fixed means two different things: fixed in size, and fixed in place. The two are often used together, in that a fixed-size buffer must be fixed in place to be used. void* A void pointer (void*) makes no assumptions about the type of the underlying data and is useful for functions that deal with raw memory. An implicit conversion exists from any pointer type to void*. A void* cannot be dereferenced, and arithmetic operations cannot be performed on void pointers. Unsafe Code and Pointers | 179 Advanced C# For example: class Test { unsafe static void Main() { short[ ] a = {1,1,2,3,5,8,13,21,34,55}; fixed (short* p = a) { //sizeof returns size of value-type in bytes Zap (p, a.Length * sizeof (short)); } foreach (short x in a) System.Console.WriteLine (x); // Prints all zeros } unsafe static void Zap (void* memory, int byteCount) { byte* b = (byte*) memory; for (int i = 0; i < byteCount; i++) *b++ = 0; } } Pointers to Unmanaged Code Pointers are also useful for accessing data outside the managed heap (such as when interacting with C DLLs or COM), or when dealing with data not in the main mem- ory (such as graphics memory or a storage medium on an embedded device). Preprocessor Directives Preprocessor directives supply the compiler with additional information about regions of code. The most common preprocessor directives are the conditional di- rectives, which provide a way to include or exclude regions of code from compila- tion. For example: #define DEBUG class MyClass { int x; void Foo() { # if DEBUG Console.WriteLine ("Testing: x = {0}", x); # endif } ... } In this class, the statement in Foo is compiled as conditionally dependent upon the presence of the DEBUG symbol. If we remove the DEBUG symbol, the statement is not compiled. Preprocessor symbols can be defined within a source file (as we have 180 | Chapter 4: Advanced C# done), and they can be passed to the compiler with the /define:symbol command- line option. With the #if and #elif directives, you can use the ||, &&, and ! operators to perform or, and, and not operations on multiple symbols. The following directive instructs the compiler to include the code that follows if the TESTMODE symbol is defined and the DEBUG symbol is not defined: #if TESTMODE && !DEBUG ... Bear in mind, however, that you’re not building an ordinary C# expression, and the symbols upon which you operate have absolutely no connection to variables—static or otherwise. The #error and #warning symbols prevent accidental misuse of conditional directives by making the compiler generate a warning or error given an undesirable set of compilation symbols. Table 4-1 lists the preprocessor directives. Table 4-1. Preprocessor directives Preprocessor directive Action #define symbol Defines symbol #undef symbol Undefines symbol #if symbol [operator symbol2]... symbol to test operators are ==, !=, &&, and || followed by #else, #elif, and #endif #else Executes code to subsequent #endif #elif symbol [operator symbol2] Combines #else branch and #if test #endif Ends conditional directives #warning text text of the warning to appear in compiler output #error text text of the error to appear in compiler output #pragma warning [disable | restore] Disables/restores compiler warning(s) #line [ number ["file"] | hidden] number specifies the line in source code; file is the file- name to appear in computer output; hidden instructs debuggers to skip over code from this point until the next #line directive #region name Marks the beginning of an outline #endregion Ends an outline region Conditional Attributes An attribute decorated with the Conditional attribute will be compiled only if a given preprocessor symbol is present. For example: // file1.cs #define DEBUG using System; Preprocessor Directives | 181 Advanced C# using System.Diagnostics; [Conditional("DEBUG")] public class TestAttribute : Attribute {} // file2.cs #define DEBUG [Test] class Foo { [Test] string s; } The compiler will incorporate the [Test] attributes only if the DEBUG symbol is in scope for file2.cs. Pragma Warning The compiler generates a warning when it spots something in your code that seems unintentional. Unlike errors, warnings don’t ordinarily prevent your application from compiling. Compiler warnings can be extremely valuable in spotting bugs. Their usefulness, however, is undermined when you get false warnings. In a large application, main- taining a good signal-to-noise ratio is essential if the “real” warnings are to get noticed. To this effect, the compiler allows you to selectively suppress warnings with the #pragma warning directive. In this example, we instruct the compiler not to warn us about the field Message not being used: public class Foo { static void Main() { } #pragma warning disable 414 static string Message = "Hello"; #pragma warning restore 414 } Omitting the number in the #pragma warning directive disables or restores all warning codes. If you are thorough in applying this directive, you can compile with the /warnaser ror switch—this tells the compiler to treat any residual warnings as errors. XML Documentation A documentation comment is a piece of embedded XML that documents a type or member. A documentation comment comes immediately before a type or member declaration, and starts with three slashes: /// Cancels a running query. public void Cancel() { ... } 182 | Chapter 4: Advanced C# Multiline comments can be done either like this: /// /// Cancels a running query /// public void Cancel() { ... } or like this (notice the extra star at the start): /** Cancels a running query. */ public void Cancel() { ... } If you compile with the /doc directive, the compiler extracts and collates documen- tation comments into a single XML file. This has two main uses: • If placed in the same folder as the compiled assembly, Visual Studio automat- ically reads the XML file and uses the information to provide IntelliSense mem- ber listings to consumers of the assembly of the same name. • Third-party tools (such as Sandcastle and NDoc) can transform XML file into an HTML help file. Standard XML Documentation Tags Here are the standard XML tags that Visual Studio and documentation generators recognize: ... Indicates the tool tip that IntelliSense should display for the type or member; typically a single phrase or sentence. ... Additional text that describes the type or member. Documentation generators pick this up and merge it into the bulk of a type or member’s description. ... Explains a parameter on a method. ... Explains the return value for a method. ... Lists an exception that a method may throw (cref refers to the exception type). XML Documentation | 183 Advanced C# ... Indicates an IPermission type required by the documented type or member. ... Denotes an example (used by documentation generators). This usually contains both description text and source code (source code is typically within a or tag). ... Indicates an inline code snippet. This tag is usually used inside an block. ... Indicates a multiline code sample. This tag is usually used inside an block. ... Inserts an inline cross-reference to another type or member. HTML documen- tation generators typically convert this to a hyperlink. The compiler emits a warning if the type or member name is invalid. To refer to generic types, use curly braces; for example, cref="Foo{T,U}". ... Cross-references another type or member. Documentation generators typically write this into a separate “See Also” section at the bottom of the page. References a parameter from within a or tag. ... ... ... ... 184 | Chapter 4: Advanced C# Instructs documentation generators to emit a bulleted, numbered, or table-style list. ... Instructs documentation generators to format the contents into a separate paragraph. ... Merges an external XML file that contains documentation. The path attribute denotes an XPath query to a specific element in that file. User-Defined Tags Little is special about the predefined XML tags recognized by the C# compiler, and you are free to define your own. The only special processing done by the compiler is on the tag (in which it verifies the parameter name and that all the pa- rameters on the method are documented) and the cref attribute (in which it verifies that the attribute refers to a real type or member and expands it to a fully qualified type or member ID). The cref attribute can also be used in your own tags and is verified and expanded just as it is in the predefined , , , and tags. Type or Member Cross-References Type names and type or member cross-references are translated into IDs that uniquely define the type or member. These names are composed of a prefix that defines what the ID represents and a signature of the type or member. The member prefixes are: XML type prefix ID prefixes applied to... N Namespace T Type (class, struct, enum, interface, delegate) F Field P Property (includes indexers) M Method (includes special methods) E Event ! Error The rules describing how the signatures are generated are well documented, although fairly complex. Here is an example of a type and the IDs that are generated: // Namespaces do not have independent signatures namespace NS XML Documentation | 185 Advanced C# { /// T:NS.MyClass class MyClass { /// F:NS.MyClass.aField string aField; /// P:NS.MyClass.aProperty short aProperty {get {...} set {...}} /// T:NS.MyClass.NestedType class NestedType {...}; /// M:NS.MyClass.X() void X() {...} /// M:NS.MyClass.Y(System.Int32,System.Double@,System.Decimal@) void Y(int p1, ref double p2, out decimal p3) {...} /// M:NS.MyClass.Z(System.Char[ ],System.Single[0:,0:]) void Z(char[ ] 1, float[,] p2) {...} /// M:NS.MyClass.op_Addition(NS.MyClass,NS.MyClass) public static MyClass operator+(MyClass c1, MyClass c2) {...} /// M:NS.MyClass.op_Implicit(NS.MyClass)˜System.Int32 public static implicit operator int(MyClass c) {...} /// M:NS.MyClass.#ctor MyClass() {...} /// M:NS.MyClass.Finalize ˜MyClass() {...} /// M:NS.MyClass.#cctor static MyClass() {...} } } 186 | Chapter 4: Advanced C# 5Framework Overview Almost all the capabilities of the .NET Framework are exposed via a vast set of managed types. These types are organized into hierarchical namespaces and pack- aged into a set of assemblies, which together with the CLR comprise the .NET platform. Some of the .NET types are used directly by the CLR and are essential for the man- aged hosting environment. These types reside in an assembly called mscorlib.dll and include C#’s built-in types, as well as the basic collection classes, types for stream processing, serialization, reflection, threading, and native interoperability (“mscor- lib” is an abbreviation for Multi-language Standard Common Object Runtime Library”). At a level above this are additional types that “flesh out” the CLR-level functionality, providing features such as XML, networking, and LINQ. These reside in Sys- tem.dll, System.Xml.dll, and System.Core.dll, and together with mscorlib they pro- vide a rich programming environment upon which the rest of the Framework is built. This “core framework” largely defines the scope of the rest of this book. The remainder of the .NET Framework consists of applied APIs, most of which cover three areas of functionality: • User interface technologies • Backend technologies • Distributed system technologies Table 5-1 shows the history of compatibility between each version of C#, the CLR, and the .NET Framework. C# 5.0 targets CLR 4.5 which is unusual because it’s a patched version of CLR 4.0. This means that applications targeting CLR 4.0 will actually run on CLR 4.5 after you install the latter; hence Microsoft has taken extreme care to ensure backward compatibility. 187 Table 5-1. C#, CLR, and .NET Framework versions C# version CLR version Framework versions 1.0 1.0 1.0 1.2 1.1 1.1 2.0 2.0 2.0, 3.0 3.0 2.0 (SP1) 3.5 4.0 4.0 4.0 5.0 4.5 (Patched CLR 4.0) 4.5 This chapter skims all key areas of the .NET Framework—starting with the core types covered in this book and finishing with an overview of the applied technologies. Assemblies and namespaces in the .NET Framework cross-cut. The most extreme examples are mscorlib.dll and Sys- tem.Core.dll, both defining types in dozens of namespaces, none of which is prefixed with mscorlib or System.Core. The less obvious cases are the more confusing ones, however, such as the types in System.Security.Cryptography. Most types in this namespace reside in System.dll, except for a handful, which reside in System.Security.dll. The book’s companion website contains a complete mapping of Framework namespaces to assemblies (http://www.albahari.com/nutshell/namespacerefer ence.aspx). What’s New in .NET Framework 4.5 New features of Framework 4.5 include: • Extensive support for asynchrony through Task-returning methods • Support for the zip compression protocol (Chapter 15) • Improved HTTP support through the new HttpClient class (Chapter 16) • Performance improvements to the garbage collector and assembly resource retrieval • Support for WinRT interoperability and APIs for building Metro-style tablet apps There’s also a new TypeInfo class (Chapter 19) and the ability to specify timeouts when matching regular expression timeouts (Chapter 26). In the Parallel Computing space, there’s a specialized new library called Dataflow for building producer/consumer-style networks. There have also been improvements to the WPF, WCF and WF (Workflow Foun- dation) libraries. 188 | Chapter 5: Framework Overview Many of the core types are defined in the following assemblies: mscorlib.dll, Sys- tem.dll, and System.Core.dll. The first of these, mscorlib.dll, comprises the types required by the runtime environment itself; System.dll and System.Core.dll contain additional core types required by you as a programmer. The reason the latter two are separate is historical: when the Microsoft team introduced Framework 3.5, they made it additive insofar as it ran over the existing CLR 2.0 (just as Framework 4.5 is additive over 4.0). Therefore, almost all new core types (such as the classes sup- porting LINQ) went into a new assembly which Microsoft called System.Core.dll. What’s New in .NET Framework 4.0 Framework 4.0 added the following: • New core types: BigInteger (for arbitrarily large numbers), Complex (complex numbers), and tuples (Chapter 6) • A new SortedSet collection (Chapter 7) • Code contracts, for enabling methods to interact more reliably through mutual obligations and responsibilities (Chapter 13) • Direct support for memory-mapped files (Chapter 15) • Lazy file- and directory-I/O methods that return IEnumerable instead of arrays (Chapter 15) • The Dynamic Language Runtime (DLR), which is now part of the .NET Framework (Chapter 20) • Security transparency, which makes it easier to secure libraries in partially trusted environments (Chapter 21) • New threading constructs, including a more robust Monitor.Enter overload, Thread.Yield, new signaling classes (Barrier and CountdownEvent), and lazy initialization primitives (Chapter 22) • Parallel programming APIs for leveraging multicore processors, including Parallel LINQ (PLINQ), imperative data and task parallelism constructs, concurrent collections, and low-latency synchronization and spinning primitives (Chapter 23) • Methods for application domain resource monitoring (Chapter 24) Framework 4.0 also contained enhancements to ASP.NET, including the MVC framework and Dynamic Data, and enhancements to Entity Framework, WPF, WCF, and Workflow. In addition, it included the Managed Extensibility Frame- work library, to assist with runtime composition, discovery, and dependency injection. The CLR and Core Framework System Types The most fundamental types live directly in the System namespace. These include C#’s built-in types, the Exception base class, the Enum, Array, and Delegate base The CLR and Core Framework | 189 Framework Overview classes, and Nullable, Type, DateTime, TimeSpan, and Guid. The System namespace also includes types for performing mathematical functions (Math), generating ran- dom numbers (Random), and converting between various types (Convert and Bit Converter). Chapter 6 describes these types—as well as the interfaces that define standard pro- tocols used across the .NET Framework for such tasks as formatting (IFormatta ble) and order comparison (IComparable). The System namespace also defines the IDisposable interface and the GC class for interacting with the garbage collector. These topics are saved for Chapter 12. Text Processing The System.Text namespace contains the StringBuilder class (the editable or mutable cousin of string), and the types for working with text encodings, such as UTF-8 (Encoding and its subtypes). We cover this in Chapter 6. The System.Text.RegularExpressions namespace contains types that perform advanced pattern-based search and replace operations; these are described in Chapter 26. Collections The .NET Framework offers a variety of classes for managing collections of items. These include both list- and dictionary-based structures, and work in conjunction with a set of standard interfaces that unify their common characteristics. All collec- tion types are defined in the following namespaces, covered in Chapter 7: System.Collections // Nongeneric collections System.Collections.Generic // Generic collections System.Collections.Specialized // Strongly typed collections System.Collections.ObjectModel // Bases for your own collections System.Collections.Concurrent // Thread-safe collection (Chapter 23) Queries Language Integrated Query (LINQ) was added in Framework 3.5. LINQ allows you to perform type-safe queries over local and remote collections (e.g., SQL Server tables) and is described in Chapter 8 through Chapter 10. A big advantage of LINQ is that it presents a consistent querying API across a variety of domains. The types for resolving LINQ queries reside in these namespaces: System.Linq // LINQ to Objects and PLINQ System.Linq.Expressions // For building expressions manually System.Xml.Linq // LINQ to XML The full .NET profile also includes the following: System.Data.Linq // LINQ to SQL System.Data.Entity // LINQ to Entities (Entity Framework) (The Metro profile excludes the entire System.Data.* namespace.) 190 | Chapter 5: Framework Overview The LINQ to SQL and Entity Framework APIs leverage lower-level ADO.NET types in the System.Data namespace. XML XML is used widely within the .NET Framework, and so is supported extensively. Chapter 10 focuses entirely on LINQ to XML—a lightweight XML document object model that can be constructed and queried through LINQ. Chapter 11 describes the older W3C DOM, as well as the performant low-level reader/writer classes and the Framework’s support for XML schemas, stylesheets, and XPath. The XML name- spaces are: System.Xml // XmlReader, XmlWriter + the old W3C DOM System.Xml.Linq // The LINQ to XML DOM System.Xml.Schema // Support for XSD System.Xml.Serialization // Declarative XML serialization for .NET types The following namespaces are available in the standard .NET profiles (but not Metro): System.Xml.XPath // XPath query language System.Xml.Xsl // Stylesheet support Diagnostics and Code Contracts In Chapter 13, we cover .NET’s logging and assertion facilities and the code con- tracts system which was introduced in Framework 4.0. We also describe how to interact with other processes, write to the Windows event log, and use performance counters for monitoring. The types for this are defined in and under System.Diagnostics. Concurrency and Asynchrony Most modern applications need to deal with more than one thing happening at a time. C# 5.0 and Framework 4.5 make this easier than in the past through asyn- chronous functions and high-level constructs such as tasks and task combinators. Chapter 14 explains all of this in detail, after starting with the basics of multithread- ing. Types for working with threads and asynchronous operations are in the System.Threading and System.Threading.Tasks namespaces. Streams and I/O The Framework provides a stream-based model for low-level input/output. Streams are typically used to read and write directly to files and network connections, and can be chained or wrapped in decorator streams to add compression or encryption functionality. Chapter 15 describes .NET’s stream architecture, as well as the spe- cific support for working with files and directories, compression, isolated storage, pipes, and memory-mapped files. The .NET Stream and I/O types are defined in and under the System.IO namespace, and the WinRT types for file I/O are in and under Windows.Storage. The CLR and Core Framework | 191 Framework Overview Networking You can directly access standard network protocols such as HTTP, FTP, TCP/IP, and SMTP via the types in System.Net. In Chapter 16, we demonstrate how to com- municate using each of these protocols, starting with simple tasks such as down- loading from a web page, and finishing with using TCP/IP directly to retrieve POP3 email. Here are the namespaces we cover: System.Net System.Net.Http // HttpClient System.Net.Mail // For sending mail via SMTP System.Net.Sockets // TCP, UDP, and IP The latter two namespaces are not available to Metro applications, which must instead use WinRT types for these functions. Serialization The Framework provides several systems for saving and restoring objects to a binary or text representation. Such systems are required for distributed application tech- nologies, such as WCF, Web Services, and Remoting, and also to save and restore objects to a file. In Chapter 17, we cover all three serialization engines: the data contract serializer, the binary serializer, and the XML serializer. The types for seri- alization reside in the following namespaces: System.Runtime.Serialization System.Xml.Serialization The Metro profile excludes the binary serialization engine. Assemblies, Reflection, and Attributes The assemblies into which C# programs compile comprise executable instructions (stored as intermediate language or IL) and metadata, which describes the program’s types, members, and attributes. Through reflection, you can inspect this metadata at runtime, and do such things as dynamically invoke methods. With Reflec tion.Emit, you can construct new code on the fly. In Chapter 18, we describe the makeup of assemblies and how to sign them, use the global assembly cache and resources, and resolve file references. In Chapter 19, we cover reflection and attributes—describing how to inspect metadata, dynamically invoke functions, write custom attributes, emit new types, and parse raw IL. The types for using reflection and working with assemblies reside in the following namespaces: System System.Reflection System.Reflection.Emit (not in Metro profile) Dynamic Programming In Chapter 20, we look at some of the patterns for dynamic programming and lev- eraging the Dynamic Language Runtime, which has been a part of the CLR since 192 | Chapter 5: Framework Overview Framework 4.0. We describe how to implement the Visitor pattern, write custom dynamic objects, and interoperate with IronPython. The types for dynamic pro- gramming are in System.Dynamic. Security The .NET Framework provides its own security layer, allowing you to both sandbox other assemblies and be sandboxed yourself. In Chapter 21, we cover code access, role, and identity security, and the transparency model introduced in CLR 4.0. We then describe cryptography in the Framework, covering encryption, hashing, and data protection. The types for this are defined in: System.Security System.Security.Permissions System.Security.Policy System.Security.Cryptography Only System.Security is available in the Metro profile; cryptography is handled instead in WinRT. Advanced Threading C# 5’s asynchronous functions make concurrent programming significantly easier because they lessen the need for lower-level techniques. However, there are still times when you need signaling constructs, thread-local storage, reader/writer locks, and so on. Chapter 22 explains this in depth. Threading types are in the Sys tem.Threading namespace. Parallel Programming In Chapter 23, we cover in detail the libraries and types for leveraging multicore processors, including APIs for task parallelism, imperative data parallelism and functional parallelism (PLINQ). Application Domains The CLR provides an additional level of isolation within a process, called an appli- cation domain. In Chapter 24, we examine the properties of an application domain with which you can interact, and demonstrate how to create and use additional application domains within the same process for such purposes as unit testing. We also describe how to use Remoting to communicate with these application domains. The AppDomain type is defined in the System namespace is applicable only to non- Metro apps. Native and COM Interoperability You can interoperate with both native and COM code. Native interoperability allows you to call functions in unmanaged DLLs, register callbacks, map data structures, and interoperate with native data types. COM interoperability allows you to call The CLR and Core Framework | 193 Framework Overview COM types and expose .NET types to COM. The types that support these functions are in System.Runtime.InteropServices, and we cover them in Chapter 25. Applied Technologies User Interface Technologies The .NET Framework provides four APIs for user interface–based applications: ASP.NET (System.Web.UI) For writing thin client applications that run over a standard web browser Silverlight For providing a rich user interface inside a web browser Windows Presentation Foundation (System.Windows) For writing rich client applications Windows Forms (System.Windows.Forms) For maintaining legacy rich client applications In general, a thin client application amounts to a website; a rich client application is a program the end user must download or install on the client computer. ASP.NET Applications written using ASP.NET host under Windows IIS and can be accessed from almost any web browser. Here are the advantages of ASP.NET over rich client technologies: • There is zero deployment at the client end. • Clients can run a non-Windows platform. • Updates are easily deployed. Further, because most of what you write in an ASP.NET application runs on the server, you design your data access layer to run in the same application domain— without limiting security or scalability. In contrast, a rich client that does the same is not generally as secure or scalable. (The solution, with the rich client, is to insert a middle tier between the client and database. The middle tier runs on a remote application server [often alongside the database server] and communicates with the rich clients via WCF, Web Services, or Remoting.) In writing your web pages, you can choose between the traditional Web Forms and the new MVC (Model-View-Controller) API. Both build on the ASP.NET infra- structure. Web Forms has been part of the Framework since its inception; MVC was written much later in response to the success of Ruby on Rails and MonoRail. Although debuting in Framework 4.0, the MVC framework has benefited from having matured for some time as a public beta. It provides, in general, a better pro- gramming abstraction than Web Forms; it also allows more control over the gener- ated HTML. What you lose over Web Forms is a designer. This makes Web Forms still a good choice for web pages with predominately static content. 194 | Chapter 5: Framework Overview The limitations of ASP.NET are largely a reflection of the limitations of thin client systems in general: • A web browser interface significantly restricts what you can do. • Maintaining state on the client—or on behalf of the client—is cumbersome. You can improve the interactivity and responsiveness, however, through client-side scripting or technologies such as AJAX: a good resource for this is http://ajax.asp .net. Use of AJAX is simplified through the use of libraries such as jQuery. The types for writing ASP.NET applications are in the System.Web.UI namespace and its subnamespaces, and are packed in the System.Web.dll assembly. Silverlight Silverlight is not technically part of the main .NET Framework: it’s a separate Framework that includes a subset of the Framework’s core features—plus the ability to run as a web browser plug-in. Its graphics model is essentially a subset of WPF and this allows you to leverage existing knowledge in developing Silverlight appli- cations. Silverlight is available as a small cross-platform download for web browsers —much like Macromedia’s Flash. Flash has a far bigger installation base and so dominates in this area. For this reason, Silverlight tends to be used in fringe scenarios—corporate intranets, for example. Metro Also not part of the.NET Framework, the Windows “Metro” library is for developing tablet user interfaces in Windows 8 (see “C# and Windows Runtime” on page 5 in Chapter 1). The Metro API was inspired by WPF and uses XAML for layout. The namespaces are Windows.UI and Windows.UI.Xaml. Windows Presentation Foundation (WPF) WPF was introduced in Framework 3.0 for writing rich client applications. The benefits of WPF over its predecessor, Windows Forms, are as follows: • It supports sophisticated graphics, such as arbitrary transformations, 3D rendering, and true transparency. • Its primary measurement unit is not pixel-based, so applications display correctly at any DPI (dots per inch) setting. • It has extensive dynamic layout support, which means you can localize an application without danger of elements overlapping. • Rendering uses DirectX and is fast, taking good advantage of graphics hardware acceleration. • User interfaces can be described declaratively in XAML files that can be main- tained independently of the “code-behind” files—this helps to separate appearance from functionality. WPF’s size and complexity, however, make for a steep learning curve. Applied Technologies | 195 Framework Overview The types for writing WPF applications are in the System.Windows namespace and all subnamespaces except for System.Windows.Forms. Windows Forms Windows Forms is a rich client API that’s as old as the .NET Framework. Compared to WPF, Windows Forms is a relatively simple technology that provides most of the features you need in writing a typical Windows application. It also has significant relevancy in maintaining legacy applications. It has a number of drawbacks, though, compared to WPF: • Controls are positioned and sized in pixels, making it easy to write applications that break on clients whose DPI settings differ from the developer’s. • The API for drawing nonstandard controls is GDI+, which, although reasonably flexible, is slow in rendering large areas (and without double buffering, may flicker). • Controls lack true transparency. • Dynamic layout is difficult to get right reliably. The last point is an excellent reason to favor WPF over Windows Forms—even if you’re writing a business application that needs just a user interface and not a “user experience.” The layout elements in WPF, such as Grid, make it easy to assemble labels and text boxes such that they always align—even after language changing localization—without messy logic and without any flickering. Further, you don’t have to bow to the lowest common denominator in screen resolution—WPF layout elements have been designed from the outset to adapt properly to resizing. On the positive side, Windows Forms is relatively simple to learn and still has a wealth of support in third-party controls. The Windows Forms types are in the System.Windows.Forms (in System.Win- dows.Forms.dll) and System.Drawing (in System.Drawing.dll) namespaces. The latter also contains the GDI+ types for drawing custom controls. Backend Technologies ADO.NET ADO.NET is the managed data access API. Although the name is derived from the 1990s-era ADO (ActiveX Data Objects), the technology is completely different. ADO.NET contains two major low-level components: Provider layer The provider model defines common classes and interfaces for low-level access to database providers. These interfaces comprise connections, commands, adapters, and readers (forward-only, read-only cursors over a database). The Framework ships with native support for Microsoft SQL Server and Oracle and has OLE-DB and ODBC providers. 196 | Chapter 5: Framework Overview DataSet model A DataSet is a structured cache of data. It resembles a primitive in-memory database, which defines SQL constructs such as tables, rows, columns, rela- tionships, constraints, and views. By programming against a cache of data, you can reduce the number of trips to the server, increasing server scalability and the responsiveness of a rich-client user interface. DataSets are serializable and are designed to be sent across the wire between client and server applications. Sitting above the provider layer are two APIs that offer the ability to query databases via LINQ: • LINQ to SQL (introduced in Framework 3.5) • Entity Framework (introduced in Framework 3.5 SP1) Both technologies include object/relational mappers (ORMs), meaning they auto- matically map objects (based on classes that you define) to rows in the database. This allows you to query those objects via LINQ (instead of writing SQL select statements)—and update them without manually writing SQL insert/delete/ update statements. This cuts the volume of code in an application’s data access layer (particularly the “plumbing” code) and provides strong static type safety. These technologies also avoid the need for DataSets as receptacles of data—although DataSets still provide the unique ability to store and serialize state changes (some- thing particularly useful in multitier applications). You can use LINQ to SQL or Entity Framework in conjunction with DataSets, although the process is somewhat clumsy and DataSets are inherently ungainly. In other words, there’s no straight- forward out-of-the-box solution for writing n-tier applications with Microsoft’s ORMs as yet. LINQ to SQL is simpler and faster than Entity Framework, and tends to produce better SQL (although Entity Framework has improved in later versions). Entity Framework is more flexible in that you can create elaborate mappings between the database and the classes that you query, and offers a model that allows third-party support for databases other than SQL Server. Windows Workflow Windows Workflow is a framework for modeling and managing potentially long- running business processes. Workflow targets a standard runtime library, providing consistency and interoperability. Workflow also helps reduce coding for dynami- cally controlled decision-making trees. Windows Workflow is not strictly a backend technology—you can use it anywhere (an example is page flow, in the UI). Workflow came originally with .NET Framework 3.0, with its types defined in the System.WorkFlow namespace. Workflow was substantially revised in Framework 4.0; the new types live in and under the System.Activities namespace. Applied Technologies | 197 Framework Overview COM+ and MSMQ The Framework allows you to interoperate with COM+ for services such as dis- tributed transactions, via types in the System.EnterpriseServices namespace. It also supports MSMQ (Microsoft Message Queuing) for asynchronous, one-way messaging through types in System.Messaging. Distributed System Technologies Windows Communication Foundation (WCF) WCF is a sophisticated communications infrastructure introduced in Framework 3.0. WCF is flexible and configurable enough to make both of its predecessors— Remoting and (.ASMX) Web Services—mostly redundant. WCF, Remoting, and Web Services are all alike in that they implement the following basic model in allowing a client and server application to communicate: • On the server, you indicate what methods you’d like remote client applications to be able to call. • On the client, you specify or infer the signatures of the server methods you’d like to call. • On both the server and the client, you choose a transport and communication protocol (in WCF, this is done through a binding). • The client establishes a connection to the server. • The client calls a remote method, which executes transparently on the server. WCF further decouples the client and server through service contracts and data contracts. Conceptually, the client sends an (XML or binary) message to an endpoint on a remote service, rather than directly invoking a remote method. One of the ben- efits of this decoupling is that clients have no dependency on the .NET platform or on any proprietary communication protocols. WCF is highly configurable and provides the most extensive support for standar- dized messaging protocols, including WS-*. This lets you communicate with parties running different software—possibly on different platforms—while still supporting advanced features such as encryption. Another benefit of WCF is that you can change protocols without needing to change other aspects of your client or server applications. The types for communicating with WCF are in, and below, the System.ServiceModel namespace. Remoting and .ASMX Web Services Remoting and .ASMX Web Services are WCF’s predecessors and are almost redun- dant in WCF’s wake—although Remoting still has a niche in communicating between application domains within the same process (see Chapter 24). Remoting’s functionality is geared toward tightly coupled applications. A typical example is when the client and server are both .NET applications written by the 198 | Chapter 5: Framework Overview same company (or companies sharing common assemblies). Communication typi- cally involves exchanging potentially complex custom .NET objects that the Remoting infrastructure serializes and deserializes without needing intervention. The functionality of Web Services is geared toward loosely coupled or SOA-style applications. A typical example is a server designed to accept simple SOAP-based messages that originate from clients running a variety of software—on a variety of platforms. Web Services can only use HTTP and SOAP as transport and formatting protocols, and applications are normally hosted under IIS. The benefits of interoperability come at a performance cost—a Web Services application is typically slower, in both execution and development time, than a well-designed Remoting application. The types for Remoting are in or under System.Runtime.Remoting; the types for Web Services are under System.Web.Services. CardSpace CardSpace is a token-based authentication and identity management protocol designed to simplify password management for end users. The technology has re- ceived little attention because of the difficulty in porting tokens across machines (OpenID is a popular alternative that avoids this problem). CardSpace builds on open XML standards, and parties can participate independ- ently of Microsoft. A user can hold multiple identities, which are maintained by a third party (the identity provider). When a user wants to access a resource at site X, the user authenticates to the identity provider, which then issues a token to site X. This avoids having to provide a password directly to site X and reduces the number of identities that the user needs to maintain. WCF allows you to specify a CardSpace identity when connecting through a secure HTTP channel, through types in the System.IdentityModel.Claims and System.Iden tityModel.Policy namespaces. Applied Technologies | 199 Framework Overview 6Framework Fundamentals Many of the core facilities that you need when programming are provided not by the C# language, but by types in the .NET Framework. In this chapter, we cover the Framework’s role in fundamental programming tasks, such as virtual equality comparison, order comparison, and type conversion. We also cover the basic Frame- work types, such as String, DateTime, and Enum. The types in this section reside in the System namespace, with the following exceptions: • StringBuilder is defined in System.Text, as are the types for text encodings. • CultureInfo and associated types are defined in System.Globalization. • XmlConvert is defined in System.Xml. String and Text Handling Char A C# char represents a single Unicode character and aliases the System.Char struct. In Chapter 2, we described how to express char literals. For example: char c = 'A'; char newLine = '\n'; System.Char defines a range of static methods for working with characters, such as ToUpper, ToLower, and IsWhiteSpace. You can call these through either the Sys tem.Char type or its char alias: Console.WriteLine (System.Char.ToUpper ('c')); // C Console.WriteLine (char.IsWhiteSpace ('\t')); // True ToUpper and ToLower honor the end user’s locale, which can lead to subtle bugs. The following expression evaluates to false in Turkey: char.ToUpper ('i') == 'I' 201 because in Turkey, char.ToUpper ('i') is 'İ' (notice the dot on top!). To avoid this problem, System.Char (and System.String) also provides culture-invariant versions of ToUpper and ToLower ending with the word Invariant. These always apply English culture rules: Console.WriteLine (char.ToUpperInvariant ('i')); // I This is a shortcut for: Console.WriteLine (char.ToUpper ('i', CultureInfo.InvariantCulture)) For more on locales and culture, see “Formatting and parsing” on page 220. Most of char’s remaining static methods are related to categorizing characters and are listed in Table 6-1. Table 6-1. Static methods for categorizing characters Static method Characters included Unicode categories included IsLetter A–Z, a–z, and letters of other alphabets UpperCaseLetter LowerCaseLetter TitleCaseLetter ModifierLetter OtherLetter IsUpper Uppercase letters UpperCaseLetter IsLower Lowercase letters LowerCaseLetter IsDigit 0–9 plus digits of other alphabets DecimalDigitNumber IsLetterOrDigit Letters plus digits Sum of IsLetter and IsDigit IsNumber All digits plus Unicode fractions and Roman numeral symbols DecimalDigitNumber LetterNumber OtherNumber IsSeparator Space plus all Unicode separator characters LineSeparator ParagraphSeparator IsWhiteSpace All separators plus \n, \r, \t, \f, and \v LineSeparator ParagraphSeparator IsPunctuation Symbols used for punctuation in Western and other alphabets DashPunctuation ConnectorPunctuation InitialQuotePunctuation FinalQuotePunctuation IsSymbol Most other printable symbols MathSymbol ModifierSymbol OtherSymbol IsControl Nonprintable “control” characters below 0x20, such as \r, \n, \t, \0, and characters between 0x7F and 0x9A (None) 202 | Chapter 6: Framework Fundamentals For more granular categorization, char provides a static method called GetUnicode Category; this returns a UnicodeCategory enumeration whose members are shown in the rightmost column of Table 6-1. By explicitly casting from an integer, it’s possible to produce a char outside the allocated Unicode set. To test a character’s validity, call char.GetUnicodeCategory: if the result is UnicodeCa tegory.OtherNotAssigned, the character is invalid. A char is 16 bits wide—enough to represent any Unicode character in the Basic Multilingual Plane. To go outside this, you must use surrogate pairs: we describe the methods for doing this in “Text Encodings and Unicode” on page 211. String A C# string (== System.String) is an immutable (unchangeable) sequence of char- acters. In Chapter 2, we described how to express string literals, perform equality comparisons, and concatenate two strings. This section covers the remaining func- tions for working with strings, exposed through the static and instance members of the System.String class. Constructing strings The simplest way to construct a string is to assign a literal, as we saw in Chapter 2: string s1 = "Hello"; string s2 = "First Line\r\nSecond Line"; string s3 = @"\\server\fileshare\helloworld.cs"; To create a repeating sequence of characters you can use string’s constructor: Console.Write (new string ('*', 10)); // ********** You can also construct a string from a char array. The ToCharArray method does the reverse: char[] ca = "Hello".ToCharArray(); string s = new string (ca); // s = "Hello" string’s constructor is also overloaded to accept various (unsafe) pointer types, in order to create strings from types such as char*. Null and empty strings An empty string has a length of zero. To create an empty string, you can use either a literal or the static string.Empty field; to test for an empty string you can either perform an equality comparison or test its Length property: string empty = ""; Console.WriteLine (empty == ""); // True Console.WriteLine (empty == string.Empty); // True Console.WriteLine (empty.Length == 0); // True String and Text Handling | 203 Framework Fundamentals Because strings are reference types, they can also be null: string nullString = null; Console.WriteLine (nullString == null); // True Console.WriteLine (nullString == ""); // False Console.WriteLine (nullString.Length == 0); // NullReferenceException The static string.IsNullOrEmpty method is a useful shortcut for testing whether a given string is either null or empty. Accessing characters within a string A string’s indexer returns a single character at the given index. As with all functions that operate on strings, this is zero-indexed: string str = "abcde"; char letter = str[1]; // letter == 'b' string also implements IEnumerable, so you can foreach over its characters: foreach (char c in "123") Console.Write (c + ","); // 1,2,3, Searching within strings The simplest methods for searching within strings are Contains, StartsWith, and EndsWith. These all return true or false: Console.WriteLine ("quick brown fox".Contains ("brown")); // True Console.WriteLine ("quick brown fox".EndsWith ("fox")); // True The Contains method doesn’t offer the convenience of this overload, although you can achieve the same result with the IndexOf method. IndexOf is more powerful: it returns the first position of a given character or substring (or −1 if the substring isn’t found): Console.WriteLine ("abcde".IndexOf ("cd")); // 2 StartsWith, EndsWith and IndexOf are overloaded to let you specify a StringCompari son enum or a CultureInfo object to control case and culture sensitivity (see “Ordinal versus culture comparison” on page 207). The default is to perform a case-sensitive match using rules applicable to the current (localized) culture. The following instead performs a case-insensitive search using the invariant culture’s rules: "abcdef".StartsWith ("abc", StringComparison.InvariantCultureIgnoreCase) IndexOf is also overloaded to accept a startPosition (an index from which to begin searching), as well as a StringComparison enum: Console.WriteLine ("abcde".IndexOf ("CD", StringComparison.CurrentCultureIgnoreCase)); // 2 LastIndexOf is like IndexOf, but works backward through the string. IndexOfAny returns the first matching position of any one of a set of characters: Console.Write ("ab,cd ef".IndexOfAny (new char[] {' ', ','} )); // 2 Console.Write ("pas5w0rd".IndexOfAny ("0123456789".ToCharArray() )); // 3 204 | Chapter 6: Framework Fundamentals LastIndexOfAny does the same in the reverse direction. Manipulating strings Because String is immutable, all the methods that “manipulate” a string return a new one, leaving the original untouched (the same goes for when you reassign a string variable). Substring extracts a portion of a string: string left3 = "12345".Substring (0, 3); // left3 = "123"; string mid3 = "12345".Substring (1, 3); // mid3 = "234"; If you omit the length, you get the remainder of the string: string end3 = "12345".Substring (2); // end3 = "345"; Insert and Remove insert or remove characters at a specified position: string s1 = "helloworld".Insert (5, ", "); // s1 = "hello, world" string s2 = s1.Remove (5, 2); // s2 = "helloworld"; PadLeft and PadRight pad a string to a given length with a specified character (or a space if unspecified): Console.WriteLine ("12345".PadLeft (9, '*')); // ****12345 Console.WriteLine ("12345".PadLeft (9)); // 12345 If the input string is longer than the padding length, the original string is returned unchanged. TrimStart and TrimEnd remove specified characters from the beginning or end of a string; Trim does both. By default, these functions remove whitespace characters (including spaces, tabs, new lines, and Unicode variations of these): Console.WriteLine (" abc \t\r\n ".Trim().Length); // 3 Replace replaces all (non-overlapping) occurrences of a particular character or substring: Console.WriteLine ("to be done".Replace (" ", " | ") ); // to | be | done Console.WriteLine ("to be done".Replace (" ", "") ); // tobedone ToUpper and ToLower return upper- and lowercase versions of the input string. By default, they honor the user’s current language settings; ToUpperInvariant and ToLowerInvariant always apply English alphabet rules. Splitting and joining strings Split takes a sentence and returns an array of words: string[] words = "The quick brown fox".Split(); foreach (string word in words) Console.Write (word + "|"); // The|quick|brown|fox| By default, Split uses whitespace characters as delimiters; it’s also overloaded to accept a params array of char or string delimiters. Split also optionally accepts a String and Text Handling | 205 Framework Fundamentals StringSplitOptions enum, which has an option to remove empty entries: this is useful when words are separated by several delimiters in a row. The static Join method does the reverse of Split. It requires a delimiter and string array: string[] words = "The quick brown fox".Split(); string together = string.Join (" ", words); // The quick brown fox The static Concat method is similar to Join but accepts only a params string array and applies no separator. Concat is exactly equivalent to the + operator (the compiler, in fact, translates + to Concat): string sentence = string.Concat ("The", " quick", " brown", " fox"); string sameSentence = "The" + " quick" + " brown" + " fox"; String.Format and composite format strings The static Format method provides a convenient way to build strings that embed variables. The embedded variables (or values) can be of any type; the Format simply calls ToString on them. The master string that includes the embedded variables is called a composite format string. When calling String.Format, you provide a composite format string followed by each of the embedded variables. For example: string composite = "It's {0} degrees in {1} on this {2} morning"; string s = string.Format (composite, 35, "Perth", DateTime.Now.DayOfWeek); // s == "It's 35 degrees in Perth on this Friday morning" (And that’s Celsius!) Each number in curly braces is called a format item. The number corresponds to the argument position and is optionally followed by: • A comma and a minimum width to apply • A colon and a format string The minimum width is useful for aligning columns. If the value is negative, the data is left-aligned; otherwise, it’s right-aligned. For example: string composite = "Name={0,-20} Credit Limit={1,15:C}"; Console.WriteLine (string.Format (composite, "Mary", 500)); Console.WriteLine (string.Format (composite, "Elizabeth", 20000)); Here’s the result: Name=Mary Credit Limit= $500.00 Name=Elizabeth Credit Limit= $20,000.00 The equivalent without using string.Format is this: string s = "Name=" + "Mary".PadRight (20) + " Credit Limit=" + 500.ToString ("C").PadLeft (15); 206 | Chapter 6: Framework Fundamentals The credit limit is formatted as currency by virtue of the "C" format string. We describe format strings in detail in “Formatting and parsing” on page 220. A disadvantage of composite format strings is that it’s easier to make a mistake that goes unnoticed until runtime—such as having greater or fewer format items than values. Such a mistake is harder to make when the format items and values are together. Comparing Strings In comparing two values, the .NET Framework differentiates the concepts of equal- ity comparison and order comparison. Equality comparison tests whether two instances are semantically the same; order comparison tests which of two (if any) instances comes first when arranging them in ascending or descending sequence. Equality comparison is not a subset of order comparison; the two systems have different purposes. It’s legal, for instance, to have two unequal values in the same ordering position. We resume this topic in “Equality Comparison” on page 254. For string equality comparison, you can use the == operator or one of string’s Equals methods. The latter are more versatile because they allow you to specify options such as case insensitivity. Another difference is that == does not work reliably on strings if the variables are cast to the object type. We explain why this is so in “Equality Comparison” on page 254. For string order comparison, you can use either the CompareTo instance method or the static Compare and CompareOrdinal methods: these return a positive or negative number—or zero—depending on whether the first value comes before, after, or alongside the second. Before going into the details of each, we need to examine .NET’s underlying string comparison algorithms. Ordinal versus culture comparison There are two basic algorithms for string comparison: ordinal and culture-sensitive. Ordinal comparisons interpret characters simply as numbers (according to their numeric Unicode value); culture-sensitive comparisons interpret characters with reference to a particular alphabet. There are two special cultures: the “current cul- ture,” which is based on settings picked up from the computer’s control panel, and the “invariant culture,” which is the same on every computer (and closely matches American culture). String and Text Handling | 207 Framework Fundamentals For equality comparison, both ordinal and culture-specific algorithms are useful. For ordering, however, culture-specific comparison is nearly always preferable: to order strings alphabetically, you need an alphabet. Ordinal relies on the numeric Unicode point values, which happen to put English characters in alphabetical order —but even then not exactly as you might expect. For example, assuming case sen- sitivity, consider the strings “Atom”, “atom”, and “Zamia”. The invariant culture puts them in the following order: "Atom", "atom", "Zamia" Ordinal arranges them instead as follows: "Atom", "Zamia", "atom" This is because the invariant culture encapsulates an alphabet, which considers uppercase characters adjacent to their lowercase counterparts (AaBbCcDd...). The ordinal algorithm, however, puts all the uppercase characters first, and then all low- ercase characters (A...Z, a...z). This is essentially a throwback to the ASCII character set invented in the 1960s. String equality comparison Despite ordinal’s limitations, string’s == operator always performs ordinal case- sensitive comparison. The same goes for the instance version of string.Equals when called without arguments; this defines the “default” equality comparison behavior for the string type. The ordinal algorithm was chosen for string’s == and Equals functions because it’s both highly efficient and deterministic. String equality comparison is considered fundamental and is performed far more frequently than order comparison. A “strict” notion of equality is also consistent with the general use of the == operator. The following methods allow culture-aware or case-insensitive comparisons: public bool Equals(string value, StringComparison comparisonType); public static bool Equals (string a, string b, StringComparison comparisonType); The static version is advantageous in that it still works if one or both of the strings are null. StringComparison is an enum defined as follows: public enum StringComparison { CurrentCulture, // Case-sensitive CurrentCultureIgnoreCase, InvariantCulture, // Case-sensitive InvariantCultureIgnoreCase, Ordinal, // Case-sensitive 208 | Chapter 6: Framework Fundamentals OrdinalIgnoreCase } For example: Console.WriteLine (string.Equals ("foo", "FOO", StringComparison.OrdinalIgnoreCase)); // True Console.WriteLine ("ṻ" == "ǖ"); // False Console.WriteLine (string.Equals ("ṻ", "ǖ", StringComparison.CurrentCulture)); // ? (The result of the third example is determined by the computer’s current language settings.) String order comparison String’s CompareTo instance method performs culture-sensitive, case-sensitive order comparison. Unlike the == operator, CompareTo does not use ordinal comparison: for ordering, a culture-sensitive algorithm is much more useful. Here’s the method’s definition: public int CompareTo (string strB); The CompareTo instance method implements the generic ICom parable interface, a standard comparison protocol used across the .NET Framework. This means string’s CompareTo defines the default ordering behavior strings, in such applications as sorted collections, for instance. For more information on ICom parable, see “Order Comparison” on page 264. For other kinds of comparison, you can call the static Compare and CompareOrdinal methods: public static int Compare (string strA, string strB, StringComparison comparisonType); public static int Compare (string strA, string strB, bool ignoreCase, CultureInfo culture); public static int Compare (string strA, string strB, bool ignoreCase); public static int CompareOrdinal (string strA, string strB); The last two methods are simply shortcuts for calling the first two methods. All of the order comparison methods return a positive number, a negative number, or zero, depending on whether the first value comes after, before, or alongside the second value: Console.WriteLine ("Boston".CompareTo ("Austin")); // 1 Console.WriteLine ("Boston".CompareTo ("Boston")); // 0 Console.WriteLine ("Boston".CompareTo ("Chicago")); // −1 String and Text Handling | 209 Framework Fundamentals Console.WriteLine ("ṻ".CompareTo ("ǖ")); // 0 Console.WriteLine ("foo".CompareTo ("FOO")); // −1 The following performs a case-insensitive comparison using the current culture: Console.WriteLine (string.Compare ("foo", "FOO", true)); // 0 By supplying a CultureInfo object, you can plug in any alphabet: // CultureInfo is defined in the System.Globalization namespace CultureInfo german = CultureInfo.GetCultureInfo ("de-DE"); int i = string.Compare ("Müller", "Muller", false, german); StringBuilder The StringBuilder class (System.Text namespace) represents a mutable (editable) string. With a StringBuilder, you can Append, Insert, Remove, and Replace substrings without replacing the whole StringBuilder. StringBuilder’s constructor optionally accepts an initial string value, as well as a starting size for its internal capacity (default is 16 characters). If you go above this, StringBuilder automatically resizes its internal structures to accommodate (at a slight performance cost) up to its maximum capacity (default is int.MaxValue). A popular use of StringBuilder is to build up a long string by repeatedly calling Append. This approach is much more efficient than repeatedly concatenating ordi- nary string types: StringBuilder sb = new StringBuilder(); for (int i = 0; i < 50; i++) sb.Append (i + ","); To get the final result, call ToString(): Console.WriteLine (sb.ToString()); 0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18,19,20,21,22,23,24,25,26, 27,28,29,30,31,32,33,34,35,36,37,38,39,40,41,42,43,44,45,46,47,48,49, In our example, the expression i + "," means that we’re still repeatedly concatenating strings. However, this incurs only a small performance cost in that the strings in question are small and don’t grow with each loop iteration. For maximum perfor- mance, however, we could change the loop body to this: { sb.Append (i.ToString()); sb.Append (","); } AppendLine performs an Append that adds a new line sequence ("\r\n" in Windows). AppendFormat accepts a composite format string, just like String.Format. As well as the Insert, Remove, and Replace methods (Replace functions such as string’s Replace), StringBuilder defines a Length property and a writable indexer for getting/setting individual characters. 210 | Chapter 6: Framework Fundamentals To clear the contents of a StringBuilder, either instantiate a new one or set its Length to zero. Setting a StringBuilder’s Length to zero doesn’t shrink its internal capacity. So, if the StringBuilder previously contained one million characters, it will continue to occupy around 2 MB of memory after zeroing its Length. If you want to release the memory, you must create a new StringBuilder and allow the old one to drop out of scope (and be garbage-collected). Text Encodings and Unicode A character set is an allocation of characters, each with a numeric code or code point. There are two character sets in common use: Unicode and ASCII. Unicode has an address space of approximately one million characters, of which about 100,000 are currently allocated. Unicode covers most spoken world languages, as well as some historical languages and special symbols. The ASCII set is simply the first 128 characters of the Unicode set, which covers most of what you see on a U.S.- style keyboard. ASCII predates Unicode by 30 years and is still sometimes used for its simplicity and efficiency: each character is represented by one byte. The .NET type system is designed to work with the Unicode character set. ASCII is implicitly supported, though, by virtue of being a subset of Unicode. A text encoding maps characters from their numeric code point to a binary repre- sentation. In .NET, text encodings come into play primarily when dealing with text files or streams. When you read a text file into a string, a text encoder translates the file data from binary into the internal Unicode representation that the char and string types expect. A text encoding can restrict what characters can be represented, as well as impacting storage efficiency. There are two categories of text encoding in .NET: • Those that map Unicode characters to another character set • Those that use standard Unicode encoding schemes The first category contains legacy encodings such as IBM’s EBCDIC and 8-bit char- acter sets with extended characters in the upper-128 region that were popular prior to Unicode (identified by a code page). The ASCII encoding is also in this category: it encodes the first 128 characters and drops everything else. This category contains the nonlegacy GB18030 as well, which is the mandatory standard for applications written in China—or sold to China—since 2000. In the second category are UTF-8, UTF-16, and UTF-32 (and the obsolete UTF-7). Each differs in space efficiency. UTF-8 is the most space-efficient for most kinds of text: it uses between 1 and 4 bytes to represent each character. The first 128 characters require only a single byte, making it compatible with ASCII. UTF-8 is the most popular encoding for text files and streams (particularly on the Internet), and it is the default for stream I/O in .NET (in fact, it’s the default for almost everything that implicitly uses an encoding). String and Text Handling | 211 Framework Fundamentals UTF-16 uses one or two 16-bit words to represent each character, and is what .NET uses internally to represent characters and strings. Some programs also write files in UTF-16. UTF-32 is the least space-efficient: it maps each code point directly to 32 bits, so every character consumes 4 bytes. UTF-32 is rarely used for this reason. It does, however, make random access very easy because every character takes an equal number of bytes. Obtaining an Encoding object The Encoding class in System.Text is the common base type for classes that encap- sulate text encodings. There are several subclasses—their purpose is to encapsulate families of encodings with similar features. The easiest way to instantiate a correctly configured class is to call Encoding.GetEncoding with a standard IANA name: Encoding utf8 = Encoding.GetEncoding ("utf-8"); Encoding chinese = Encoding.GetEncoding ("GB18030"); The most common encodings can also be obtained through dedicated static prop- erties on Encoding: Encoding name Static property on Encoding UTF-8 Encoding.UTF8 UTF-16 Encoding.Unicode (not UTF16) UTF-32 Encoding.UTF32 ASCII Encoding.ASCII The static GetEncodings method returns a list of all supported encodings, with their standard IANA names: foreach (EncodingInfo info in Encoding.GetEncodings()) Console.WriteLine (info.Name); The other way to obtain an encoding is to directly instantiate an encoding class. Doing so allows you to set various options via constructor arguments, including: • Whether to throw an exception if an invalid byte sequence is encountered when decoding. The default is false. • Whether to encode/decode UTF-16/UTF-32 with the most significant bytes first (big endian) or the least significant bytes first (little endian). The default is little endian, the standard on the Windows operating system. • Whether to emit a byte-order mark (a prefix that indicates endianness). Encoding for file and stream I/O The most common application for an Encoding object is to control how text is read and written to a file or stream. For example, the following writes “Testing...” to a file called data.txt in UTF-16 encoding: System.IO.File.WriteAllText ("data.txt", "Testing...", Encoding.Unicode); 212 | Chapter 6: Framework Fundamentals If you omit the final argument, WriteAllText applies the ubiquitous UTF-8 encoding. UTF-8 is the default text encoding for all file and stream I/O. We resume this subject in Chapter 15, in “Stream Adapters” on page 621. Encoding to byte arrays You can also use an Encoding object to go to and from a byte array. The GetBytes method converts from string to byte[] with the given encoding; GetString converts from byte[] to string: byte[] utf8Bytes = System.Text.Encoding.UTF8.GetBytes ("0123456789"); byte[] utf16Bytes = System.Text.Encoding.Unicode.GetBytes ("0123456789"); byte[] utf32Bytes = System.Text.Encoding.UTF32.GetBytes ("0123456789"); Console.WriteLine (utf8Bytes.Length); // 10 Console.WriteLine (utf16Bytes.Length); // 20 Console.WriteLine (utf32Bytes.Length); // 40 string original1 = System.Text.Encoding.UTF8.GetString (utf8Bytes); string original2 = System.Text.Encoding.Unicode.GetString (utf16Bytes); string original3 = System.Text.Encoding.UTF32.GetString (utf32Bytes); Console.WriteLine (original1); // 0123456789 Console.WriteLine (original2); // 0123456789 Console.WriteLine (original3); // 0123456789 UTF-16 and surrogate pairs Recall that .NET stores characters and strings in UTF-16. Because UTF-16 requires one or two 16-bit words per character, and a char is only 16 bits in length, some Unicode characters require two chars to represent. This has a couple of consequences: • A string’s Length property may be greater than its real character count. • A single char is not always enough to fully represent a Unicode character. Most applications ignore this, because nearly all commonly used characters fit into a section of Unicode called the Basic Multilingual Plane (BMP) which requires only one 16-bit word in UTF-16. The BMP covers several dozen world languages and includes more than 30,000 Chinese characters. Excluded are characters of some ancient languages, symbols for musical notation, and some less common Chinese characters. If you need to support two-word characters, the following static methods in char convert a 32-bit code point to a string of two chars, and back again: string ConvertFromUtf32 (int utf32) int ConvertToUtf32 (char highSurrogate, char lowSurrogate) String and Text Handling | 213 Framework Fundamentals Two-word characters are called surrogates. They are easy to spot because each word is in the range 0xD800 to 0xDFFF. You can use the following static methods in char to assist: bool IsSurrogate (char c) bool IsHighSurrogate (char c) bool IsLowSurrogate (char c) bool IsSurrogatePair (char highSurrogate, char lowSurrogate) The StringInfo class in the System.Globalization namespace also provides a range of methods and properties for working with two-word characters. Characters outside the BMP typically require special fonts and have limited operat- ing system support. Dates and Times Three immutable structs in the System namespace do the job of representing dates and times: DateTime, DateTimeOffset, and TimeSpan. C# doesn’t define any special keywords that map to these types. TimeSpan A TimeSpan represents an interval of time—or a time of the day. In the latter role, it’s simply the “clock” time (without the date), which is equivalent to the time since midnight assuming no daylight saving transition. A TimeSpan has a resolution of 100 ns, has a maximum value of about 10 million days, and can be positive or negative. There are three ways to construct a TimeSpan: • Through one of the constructors • By calling one of the static From... methods • By subtracting one DateTime from another Here are the constructors: public TimeSpan (int hours, int minutes, int seconds); public TimeSpan (int days, int hours, int minutes, int seconds); public TimeSpan (int days, int hours, int minutes, int seconds, int milliseconds); public TimeSpan (long ticks); // Each tick = 100ns The static From... methods are more convenient when you want to specify an interval in just a single unit, such as minutes, hours, and so on: public static TimeSpan FromDays (double value); public static TimeSpan FromHours (double value); public static TimeSpan FromMinutes (double value); public static TimeSpan FromSeconds (double value); public static TimeSpan FromMilliseconds (double value); 214 | Chapter 6: Framework Fundamentals For example: Console.WriteLine (new TimeSpan (2, 30, 0)); // 02:30:00 Console.WriteLine (TimeSpan.FromHours (2.5)); // 02:30:00 Console.WriteLine (TimeSpan.FromHours (−2.5)); // −02:30:00 TimeSpan overloads the < and > operators, as well as the + and - operators. The following expression evaluates to a TimeSpan of 2.5 hours: TimeSpan.FromHours(2) + TimeSpan.FromMinutes(30); The next expression evaluates to one second short of 10 days: TimeSpan.FromDays(10) - TimeSpan.FromSeconds(1); // 9.23:59:59 Using this expression, we can illustrate the integer properties Days, Hours, Minutes, Seconds, and Milliseconds: TimeSpan nearlyTenDays = TimeSpan.FromDays(10) - TimeSpan.FromSeconds(1); Console.WriteLine (nearlyTenDays.Days); // 9 Console.WriteLine (nearlyTenDays.Hours); // 23 Console.WriteLine (nearlyTenDays.Minutes); // 59 Console.WriteLine (nearlyTenDays.Seconds); // 59 Console.WriteLine (nearlyTenDays.Milliseconds); // 0 In contrast, the Total... properties return values of type double describing the entire time span: Console.WriteLine (nearlyTenDays.TotalDays); // 9.99998842592593 Console.WriteLine (nearlyTenDays.TotalHours); // 239.999722222222 Console.WriteLine (nearlyTenDays.TotalMinutes); // 14399.9833333333 Console.WriteLine (nearlyTenDays.TotalSeconds); // 863999 Console.WriteLine (nearlyTenDays.TotalMilliseconds); // 863999000 The static Parse method does the opposite of ToString, converting a string to a TimeSpan. TryParse does the same, but returns false rather than throwing an excep- tion if the conversion fails. The XmlConvert class also provides TimeSpan/string con- version methods that follow standard XML formatting protocols. The default value for a TimeSpan is TimeSpan.Zero. TimeSpan can also be used to represent the time of the day (the elapsed time since midnight). To obtain the current time of day, call DateTime.Now.TimeOfDay. DateTime and DateTimeOffset DateTime and DateTimeOffset are immutable structs for representing a date, and op- tionally, a time. They have a resolution of 100 ns, and a range covering the years 0001 through 9999. DateTimeOffset was added in Framework 3.5 and is functionally similar to Date Time. Its distinguishing feature is that it also stores a UTC offset; this allows more meaningful results when comparing values across different time zones. Dates and Times | 215 Framework Fundamentals An excellent article on the rationale behind the introduction of DateTimeOffset is available on the MSDN BCL blogs. The title is “A Brief History of DateTime” by Anthony Moore. Choosing between DateTime and DateTimeOffset DateTime and DateTimeOffset differ in how they handle time zones. A DateTime incorporates a three-state flag indicating whether the DateTime is relative to: • The local time on the current computer • UTC (the modern equivalent of Greenwich Mean Time) • Unspecified A DateTimeOffset is more specific—it stores the offset from UTC as a TimeSpan: July 01 2007 03:00:00 −06:00 This influences equality comparisons, which is the main factor in choosing between DateTime and DateTimeOffset. Specifically: • DateTime ignores the three-state flag in comparisons and considers two values equal if they have the same year, month, day, hour, minute, and so on. • DateTimeOffset considers two values equal if they refer to the same point in time. Daylight saving time can make this distinction important even if your application doesn’t need to handle multiple geographic time zones. So, DateTime considers the following two values different, whereas DateTimeOffset considers them equal: July 01 2007 09:00:00 +00:00 (GMT) July 01 2007 03:00:00 −06:00 (local time, Central America) In most cases, DateTimeOffset’s equality logic is preferable. For example, in calcu- lating which of two international events is more recent, a DateTimeOffset implicitly gives the right answer. Similarly, a hacker plotting a distributed denial of service attack would reach for a DateTimeOffset! To do the same with DateTime requires standardizing on a single time zone (typically UTC) throughout your application. This is problematic for two reasons: • To be friendly to the end user, UTC DateTimes require explicit conversion to local time prior to formatting. • It’s easy to forget and incorporate a local DateTime. DateTime is better, though, at specifying a value relative to the local computer at runtime—for example, if you want to schedule an archive at each of your interna- tional offices for next Sunday, at 3 A.M. local time (when there’s least activity). Here, DateTime would be more suitable because it would respect each site’s local time. 216 | Chapter 6: Framework Fundamentals Internally, DateTimeOffset uses a short integer to store the UTC offset in minutes. It doesn’t store any regional information, so there’s nothing present to indicate whether an offset of +08:00, for instance, refers to Singapore time or Perth time. We revisit time zones and equality comparison in more depth in “Dates and Time Zones” on page 221. SQL Server 2008 introduces direct support for DateTimeOffset through a new data type of the same name. Constructing a DateTime DateTime defines constructors that accept integers for the year, month, and day— and optionally, the hour, minute, second, and millisecond: public DateTime (int year, int month, int day); public DateTime (int year, int month, int day, int hour, int minute, int second, int millisecond); If you specify only a date, the time is implicitly set to midnight (0:00). The DateTime constructors also allow you to specify a DateTimeKind—an enum with the following values: Unspecified, Local, Utc This corresponds to the three-state flag described in the preceding section. Unspeci fied is the default and it means that the DateTime is time-zone-agnostic. Local means relative to the local time zone on the current computer. A local DateTime does not include information about which particular time zone it refers to, nor, unlike Date TimeOffset, the numeric offset from UTC. A DateTime’s Kind property returns its DateTimeKind. DateTime’s constructors are also overloaded to accept a Calendar object as well—this allows you to specify a date using any of the Calendar subclasses defined in System.Globalization. For example: DateTime d = new DateTime (5767, 1, 1, new System.Globalization.HebrewCalendar()); Console.WriteLine (d); // 12/12/2006 12:00:00 AM (The formatting of the date in this example depends on your computer’s control panel settings.) A DateTime always uses the default Gregorian calendar—this exam- ple, a one-time conversion, takes place during construction. To perform computa- tions using another calendar, you must use the methods on the Calendar subclass itself. Dates and Times | 217 Framework Fundamentals You can also construct a DateTime with a single ticks value of type long, where ticks is the number of 100 ns intervals from midnight 01/01/0001. For interoperability, DateTime provides the static FromFileTime and FromFileTime Utc methods for converting from a Windows file time (specified as a long) and FromOADate for converting from an OLE automation date/time (specified as a double). To construct a DateTime from a string, call the static Parse or ParseExact method. Both methods accept optional flags and format providers; ParseExact also accepts a format string. We discuss parsing in greater detail in “Formatting and pars- ing” on page 220. Constructing a DateTimeOffset DateTimeOffset has a similar set of constructors. The difference is that you also specify a UTC offset as a TimeSpan: public DateTimeOffset (int year, int month, int day, int hour, int minute, int second, TimeSpan offset); public DateTimeOffset (int year, int month, int day, int hour, int minute, int second, int millisecond, TimeSpan offset); The TimeSpan must amount to a whole number of minutes, or an exception is thrown. DateTimeOffset also has constructors that accept a Calendar object, a long ticks value, and static Parse and ParseExact methods that accept a string. You can construct a DateTimeOffset from an existing DateTime either by using these constructors: public DateTimeOffset (DateTime dateTime); public DateTimeOffset (DateTime dateTime, TimeSpan offset); or with an implicit cast: DateTimeOffset dt = new DateTime (2000, 2, 3); The implicit cast from DateTime to DateTimeOffset is handy because most of the .NET Framework supports DateTime—not DateTimeOffset. If you don’t specify an offset, it’s inferred from the DateTime value using these rules: • If the DateTime has a DateTimeKind of Utc, the offset is zero. • If the DateTime has a DateTimeKind of Local or Unspecified (the default), the offset is taken from the current local time zone. 218 | Chapter 6: Framework Fundamentals To convert in the other direction, DateTimeOffset provides three properties that return values of type DateTime: • The UtcDateTime property returns a DateTime in UTC time. • The LocalDateTime property returns a DateTime in the current local time zone (converting it if necessary). • The DateTime property returns a DateTime in whatever zone it was specified, with a Kind of Unspecified (i.e., it returns the UTC time plus the offset). The current DateTime/DateTimeOffset Both DateTime and DateTimeOffset have a static Now property that returns the current date and time: Console.WriteLine (DateTime.Now); // 11/11/2007 1:23:45 PM Console.WriteLine (DateTimeOffset.Now); // 11/11/2007 1:23:45 PM −06:00 DateTime also provides a Today property that returns just the date portion: Console.WriteLine (DateTime.Today); // 11/11/2007 12:00:00 AM The static UtcNow property returns the current date and time in UTC: Console.WriteLine (DateTime.UtcNow); // 11/11/2007 7:23:45 AM Console.WriteLine (DateTimeOffset.UtcNow); // 11/11/2007 7:23:45 AM +00:00 The precision of all these methods depends on the operating system and is typically in the 10–20 ms region. Working with dates and times DateTime and DateTimeOffset provide a similar set of instance properties that return various date/time elements: DateTime dt = new DateTime (2000, 2, 3, 10, 20, 30); Console.WriteLine (dt.Year); // 2000 Console.WriteLine (dt.Month); // 2 Console.WriteLine (dt.Day); // 3 Console.WriteLine (dt.DayOfWeek); // Thursday Console.WriteLine (dt.DayOfYear); // 34 Console.WriteLine (dt.Hour); // 10 Console.WriteLine (dt.Minute); // 20 Console.WriteLine (dt.Second); // 30 Console.WriteLine (dt.Millisecond); // 0 Console.WriteLine (dt.Ticks); // 630851700300000000 Console.WriteLine (dt.TimeOfDay); // 10:20:30 (returns a TimeSpan) DateTimeOffset also has an Offset property of type TimeSpan. Both types provide the following instance methods to perform computations (most accept an argument of type double or int): AddYears AddMonths AddDays AddHours AddMinutes AddSeconds AddMilliseconds AddTicks Dates and Times | 219 Framework Fundamentals These all return a new DateTime or DateTimeOffset, and they take into account such things as leap years. You can pass in a negative value to subtract. The Add method adds a TimeSpan to a DateTime or DateTimeOffset. The + operator is overloaded to do the same job: TimeSpan ts = TimeSpan.FromMinutes (90); Console.WriteLine (dt.Add (ts)); Console.WriteLine (dt + ts); // same as above You can also subtract a TimeSpan from a DateTime/DateTimeOffset and subtract one DateTime/DateTimeOffset from another. The latter gives you a TimeSpan: DateTime thisYear = new DateTime (2007, 1, 1); DateTime nextYear = thisYear.AddYears (1); TimeSpan oneYear = nextYear - thisYear; Formatting and parsing Calling ToString on a DateTime formats the result as a short date (all numbers) fol- lowed by a long time (including seconds). For example: 13/02/2000 11:50:30 AM The operating system’s control panel, by default, determines such things as whether the day, month, or year comes first, the use of leading zeros, and whether 12- or 24- hour time is used. Calling ToString on a DateTimeOffset is the same, except that the offset is returned also: 3/02/2000 11:50:30 AM −06:00 The ToShortDateString and ToLongDateString methods return just the date portion. The long date format is also determined by the control panel; an example is “Sat- urday, 17 February 2007”. ToShortTimeString and ToLongTimeString return just the time portion, such as 17:10:10 (the former excludes seconds). These four methods just described are actually shortcuts to four different format strings. ToString is overloaded to accept a format string and provider, allowing you to specify a wide range of options and control how regional settings are applied. We describe this in “Formatting and parsing” on page 220. DateTimes and DateTimeOffsets can be misparsed if the culture settings differ from those in force when formatting takes place. You can avoid this problem by using ToString in conjunction with a format string that ignores culture settings (such as “o”): DateTime dt1 = DateTime.Now; string cannotBeMisparsed = dt1.ToString ("o"); DateTime dt2 = DateTime.Parse (cannotBeMisparsed); The static Parse and ParseExact methods do the reverse of ToString, converting a string to a DateTime or DateTimeOffset. The Parse method is also overloaded to accept a format provider. 220 | Chapter 6: Framework Fundamentals Null DateTime and DateTimeOffset values Because DateTime and DateTimeOffset are structs, they are not intrinsically nullable. When you need nullability, there are two ways around this: • Use a Nullable type (i.e., DateTime? or DateTimeOffset?). • Use the static field DateTime.MinValue or DateTimeOffset.MinValue (the default values for these types). A nullable type is usually the best approach because the compiler helps to prevent mistakes. DateTime.MinValue is useful for backward compatibility with code written prior to C# 2.0 (when nullable types were introduced). Calling ToUniversalTime or ToLocalTime on a DateTime.Min Value can result in it no longer being DateTime.MinValue (depending on which side of GMT you are on). If you’re right on GMT (England, outside daylight saving), the problem won’t arise at all because local and UTC times are the same. This is your compensation for the English winter! Dates and Time Zones In this section, we examine in more detail how time zones influence DateTime and DateTimeOffset. We also look at the TimeZone and TimeZoneInfo types, which provide information on time zone offsets and daylight saving time. DateTime and Time Zones DateTime is simplistic in its handling of time zones. Internally, it stores a DateTime using two pieces of information: • A 62-bit number, indicating the number of ticks since 1/1/0001 • A 2-bit enum, indicating the DateTimeKind (Unspecified, Local, or Utc) When you compare two DateTime instances, only their ticks values are compared; their DateTimeKinds are ignored: DateTime dt1 = new DateTime (2000, 1, 1, 10, 20, 30, DateTimeKind.Local); DateTime dt2 = new DateTime (2000, 1, 1, 10, 20, 30, DateTimeKind.Utc); Console.WriteLine (dt1 == dt2); // True DateTime local = DateTime.Now; DateTime utc = local.ToUniversalTime(); Console.WriteLine (local == utc); // False The instance methods ToUniversalTime/ToLocalTime convert to universal/local time. These apply the computer’s current time zone settings and return a new DateTime with a DateTimeKind of Utc or Local. No conversion happens if you call ToUniversal Time on a DateTime that’s already Utc, or ToLocalTime on a DateTime that’s already Local. You will get a conversion, however, if you call ToUniversalTime or ToLocal Time on a DateTime that’s Unspecified. Dates and Time Zones | 221 Framework Fundamentals You can construct a DateTime that differs from another only in Kind with the static DateTime.SpecifyKind method: DateTime d = new DateTime (2000, 12, 12); // Unspecified DateTime utc = DateTime.SpecifyKind (d, DateTimeKind.Utc); Console.WriteLine (utc); // 12/12/2000 12:00:00 AM DateTimeOffset and Time Zones Internally, DateTimeOffset comprises a DateTime field whose value is always in UTC, and a 16-bit integer field for the UTC offset in minutes. Comparisons look only at the (UTC) DateTime; the Offset is used primarily for formatting. The ToUniversalTime/ToLocalTime methods return a DateTimeOffset representing the same point in time, but with a UTC or local offset. Unlike with DateTime, these methods don’t affect the underlying date/time value, only the offset: DateTimeOffset local = DateTimeOffset.Now; DateTimeOffset utc = local.ToUniversalTime(); Console.WriteLine (local.Offset); // −06:00:00 (in Central America) Console.WriteLine (utc.Offset); // 00:00:00 Console.WriteLine (local == utc); // True To include the Offset in the comparison, you must use the EqualsExact method: Console.WriteLine (local.EqualsExact (utc)); // False TimeZone and TimeZoneInfo The TimeZone and TimeZoneInfo classes provide information on time zone names, UTC offsets, and daylight saving time rules. TimeZoneInfo is the more powerful of the two and was introduced in Framework 3.5. The biggest difference between the two types is that TimeZone lets you access only the current local time zone, whereas TimeZoneInfo provides access to all the world’s time zones. Further, TimeZoneInfo exposes a richer (although at times, more awk- ward) rules-based model for describing daylight saving time. TimeZone The static TimeZone.CurrentTimeZone method returns a TimeZone object based on the current local settings. The following demonstrates the result if run in Western Australia: TimeZone zone = TimeZone.CurrentTimeZone; Console.WriteLine (zone.StandardName); // W. Australia Standard Time Console.WriteLine (zone.DaylightName); // W. Australia Daylight Time The IsDaylightSavingTime and GetUtcOffset methods work as follows: DateTime dt1 = new DateTime (2008, 1, 1); DateTime dt2 = new DateTime (2008, 6, 1); Console.WriteLine (zone.IsDaylightSavingTime (dt1)); // True Console.WriteLine (zone.IsDaylightSavingTime (dt2)); // False 222 | Chapter 6: Framework Fundamentals Console.WriteLine (zone.GetUtcOffset (dt1)); // 09:00:00 Console.WriteLine (zone.GetUtcOffset (dt2)); // 08:00:00 The GetDaylightChanges method returns specific daylight saving time information for a given year: DaylightTime day = zone.GetDaylightChanges (2008); Console.WriteLine (day.Start); // 26/10/2008 2:00:00 AM (Note D/M/Y) Console.WriteLine (day.End); // 30/03/2008 3:00:00 AM Console.WriteLine (day.Delta); // 01:00:00 TimeZoneInfo The TimeZoneInfo class works in a similar manner. TimeZoneInfo.Local returns the current local time zone: TimeZoneInfo zone = TimeZoneInfo.Local; Console.WriteLine (zone.StandardName); // W. Australia Standard Time Console.WriteLine (zone.DaylightName); // W. Australia Daylight Time TimeZoneInfo also provides IsDaylightSavingTime and GetUtcOffset methods—the difference is that they accept either a DateTime or a DateTimeOffset. You can obtain a TimeZoneInfo for any of the world’s time zones by calling FindSys temTimeZoneById with the zone ID. This feature is unique to TimeZoneInfo, as is everything else that we demonstrate from this point on. We’ll stick with Western Australia for reasons that will soon become clear: TimeZoneInfo wa = TimeZoneInfo.FindSystemTimeZoneById ("W. Australia Standard Time"); Console.WriteLine (wa.Id); // W. Australia Standard Time Console.WriteLine (wa.DisplayName); // (GMT+08:00) Perth Console.WriteLine (wa.BaseUtcOffset); // 08:00:00 Console.WriteLine (wa.SupportsDaylightSavingTime); // True The Id property corresponds to the value passed to FindSystemTimeZoneById. The static GetSystemTimeZones method returns all world time zones; hence, you can list all valid zone ID strings as follows: foreach (TimeZoneInfo z in TimeZoneInfo.GetSystemTimeZones()) Console.WriteLine (z.Id); You can also create a custom time zone by calling TimeZoneInfo.CreateCustomTimeZone. Because TimeZoneInfo is immutable, you must pass in all the relevant data as method arguments. You can serialize a predefined or custom time zone to a (semi) human-readable string by calling ToSerializedString—and deserialize it by calling TimeZoneInfo.FromSerializedString. The static ConvertTime method converts a DateTime or DateTimeOffset from one time zone to another. You can include either just a destination TimeZoneInfo, or both Dates and Time Zones | 223 Framework Fundamentals source and destination TimeZoneInfo objects. You can also convert directly from or to UTC with the methods ConvertTimeFromUtc and ConvertTimeToUtc. For working with daylight saving time, TimeZoneInfo provides the following addi- tional methods: • IsInvalidTime returns true if a DateTime is within the hour (or delta) that’s skipped when the clocks move forward. • IsAmbiguousTime returns true if a DateTime or DateTimeOffset is within the hour (or delta) that’s repeated when the clocks move back. • GetAmbiguousTimeOffsets returns an array of TimeSpans representing the valid offset choices for an ambiguous DateTime or DateTimeOffset. Unlike with TimeZone, you can’t obtain simple dates from a DateZoneInfo indicating the start and end of daylight saving time. Instead, you must call GetAdjustmen tRules, which returns a declarative summary of all daylight saving rules that apply to all years. Each rule has a DateStart and DateEnd indicating the date range within which the rule is valid: foreach (TimeZoneInfo.AdjustmentRule rule in wa.GetAdjustmentRules()) Console.WriteLine ("Rule: applies from " + rule.DateStart + " to " + rule.DateEnd); Western Australia first introduced daylight saving time in 2006, midseason (and then rescinded it in 2009). This required a special rule for the first year; hence, there are two rules: Rule: applies from 1/01/2006 12:00:00 AM to 31/12/2006 12:00:00 AM Rule: applies from 1/01/2007 12:00:00 AM to 31/12/2009 12:00:00 AM Each AdjustmentRule has a DaylightDelta property of type TimeSpan (this is one hour in almost every case) and properties called DaylightTransitionStart and Daylight TransitionEnd. The latter two are of type TimeZoneInfo.TransitionTime, which has the following properties: public bool IsFixedDateRule { get; } public DayOfWeek DayOfWeek { get; } public int Week { get; } public int Day { get; } public int Month { get; } public DateTime TimeOfDay { get; } A transition time is somewhat complicated in that it needs to represent both fixed and floating dates. An example of a floating date is “the last Sunday in March.” Here are the rules for interpreting a transition time: 1. If, for an end transition, IsFixedDateRule is true, Day is 1, Month is 1, and Time OfDay is DateTime.MinValue, there is no end to daylight saving time in that year (this can happen only in the southern hemisphere, upon the initial introduction of daylight saving time to a region). 2. Otherwise, if IsFixedDateRule is true, the Month, Day, and TimeOfDay properties determine the start or end of the adjustment rule. 224 | Chapter 6: Framework Fundamentals 3. Otherwise, if IsFixedDateRule is false, the Month, DayOfWeek, Week, and TimeOf Day properties determine the start or end of the adjustment rule. In the last case, Week refers to the week of the month, with “5” meaning the last week. We can demonstrate this by enumerating the adjustment rules for our wa time zone: foreach (TimeZoneInfo.AdjustmentRule rule in wa.GetAdjustmentRules()) { Console.WriteLine ("Rule: applies from " + rule.DateStart + " to " + rule.DateEnd); Console.WriteLine (" Delta: " + rule.DaylightDelta); Console.WriteLine (" Start: " + FormatTransitionTime (rule.DaylightTransitionStart, false)); Console.WriteLine (" End: " + FormatTransitionTime (rule.DaylightTransitionEnd, true)); Console.WriteLine(); } In FormatTransitionTime, we honor the rules just described: static string FormatTransitionTime (TimeZoneInfo.TransitionTime tt, bool endTime) { if (endTime && tt.IsFixedDateRule && tt.Day == 1 && tt.Month == 1 && tt.TimeOfDay == DateTime.MinValue) return "-"; string s; if (tt.IsFixedDateRule) s = tt.Day.ToString(); else s = "The " + "first second third fourth last".Split() [tt.Week - 1] + " " + tt.DayOfWeek + " in"; return s + " " + DateTimeFormatInfo.CurrentInfo.MonthNames [tt.Month-1] + " at " + tt.TimeOfDay.TimeOfDay; } The result with Western Australia is interesting in that it demonstrates both fixed and floating date rules—as well as an absent end date: Rule: applies from 1/01/2006 12:00:00 AM to 31/12/2006 12:00:00 AM Delta: 01:00:00 Start: 3 December at 02:00:00 End: - Rule: applies from 1/01/2007 12:00:00 AM to 31/12/2009 12:00:00 AM Delta: 01:00:00 Start: The last Sunday in October at 02:00:00 End: The last Sunday in March at 03:00:00 Dates and Time Zones | 225 Framework Fundamentals Western Australia is actually unique in this regard. Here’s how we found it: from zone in TimeZoneInfo.GetSystemTimeZones() let rules = zone.GetAdjustmentRules() where rules.Any (r => r.DaylightTransitionEnd.IsFixedDateRule) && rules.Any (r => !r.DaylightTransitionEnd.IsFixedDateRule) select zone Daylight Saving Time and DateTime If you use a DateTimeOffset or a UTC DateTime, equality comparisons are unimpeded by the effects of daylight saving time. But with local DateTimes, daylight saving can be problematic. The rules can be summarized as follows: • Daylight saving impacts local time but not UTC time. • When the clocks turn back, comparisons that rely on time moving forward will break if (and only if) they use local DateTimes. • You can always reliably round-trip between UTC and local times (on the same computer)—even as the clocks turn back. The IsDaylightSavingTime tells you whether a given local DateTime is subject to day- light saving time. UTC times always return false: Console.Write (DateTime.Now.IsDaylightSavingTime()); // True or False Console.Write (DateTime.UtcNow.IsDaylightSavingTime()); // Always False Assuming dto is a DateTimeOffset, the following expression does the same: dto.LocalDateTime.IsDaylightSavingTime The end of daylight saving time presents a particular complication for algorithms that use local time. When the clocks go back, the same hour (or more precisely, Delta) repeats itself. We can demonstrate this by instantiating a DateTime right in the “twilight zone” on your computer, and then subtracting Delta (this example requires that you practice daylight saving time to be interesting!): DaylightTime changes = TimeZone.CurrentTimeZone.GetDaylightChanges (2010); TimeSpan halfDelta = new TimeSpan (changes.Delta.Ticks / 2); DateTime utc1 = changes.End.ToUniversalTime() - halfDelta; DateTime utc2 = utc1 - changes.Delta; Converting these variables to local times demonstrates why you should use UTC and not local time if your code relies on time moving forward: DateTime loc1 = utc1.ToLocalTime(); // (Pacific Standard Time) DateTime loc2 = utc2.ToLocalTime(); Console.WriteLine (loc1); // 2/11/2010 1:30:00 AM Console.WriteLine (loc2); // 2/11/2010 1:30:00 AM Console.WriteLine (loc1 == loc2); // True 226 | Chapter 6: Framework Fundamentals Despite loc1 and loc2 reporting as equal, they are different inside. DateTime reserves a special bit for indicating on which side of the twilight zone an ambiguous local date lies! This bit is ignored in comparison—as we just saw—but comes into play when you format the DateTime unambiguously: Console.Write (loc1.ToString ("o")); // 2010-11-02T02:30:00.0000000-08:00 Console.Write (loc2.ToString ("o")); // 2010-11-02T02:30:00.0000000-07:00 This bit also is read when you convert back to UTC, ensuring perfect round-tripping between local and UTC times: Console.WriteLine (loc1.ToUniversalTime() == utc1); // True Console.WriteLine (loc2.ToUniversalTime() == utc2); // True You can reliably compare any two DateTimes by first calling ToUniversalTime on each. This strategy fails if (and only if) exactly one of them has a DateTimeKind of Unspecified. This potential for failure is another reason for favoring DateTimeOff set. Formatting and Parsing Formatting means converting to a string; parsing means converting from a string. The need to format or parse arises frequently in programming, in a variety of situa- tions. Hence, the .NET Framework provides a variety of mechanisms: ToString and Parse These methods provide default functionality for many types. Format providers These manifest as additional ToString (and Parse) methods that accept a format string and/or a format provider. Format providers are highly flexible and culture- aware. The .NET Framework includes format providers for the numeric types and DateTime/DateTimeOffset. XmlConvert This is a static class with methods that format and parse while honoring XML standards. XmlConvert is also useful for general-purpose conversion when you need culture independence or you want to preempt misparsing. XmlConvert supports the numeric types, bool, DateTime, DateTimeOffset, TimeSpan, and Guid. Type converters These target designers and XAML parsers. In this section, we discuss the first two mechanisms, focusing particularly on format providers. In the section following, we describe XmlConvert and type converters, as well as other conversion mechanisms. ToString and Parse The simplest formatting mechanism is the ToString method. It gives meaningful output on all simple value types (bool, DateTime, DateTimeOffset, TimeSpan, Guid, Formatting and Parsing | 227 Framework Fundamentals and all the numeric types). For the reverse operation, each of these types defines a static Parse method. For example: string s = true.ToString(); // s = "True" bool b = bool.Parse (s); // b = true If the parsing fails, a FormatException is thrown. Many types also define a TryParse method, which returns false if the conversion fails, rather than throwing an exception: int i; bool failure = int.TryParse ("qwerty", out i); bool success = int.TryParse ("123", out i); If you anticipate an error, calling TryParse is faster and more elegant than calling Parse in an exception handling block. The Parse and TryParse methods on DateTime(Offset) and the numeric types respect local culture settings; you can change this by specifying a CultureInfo object. Spec- ifying invariant culture is often a good idea. For instance, parsing “1.234” into a double gives us 1234 in Germany: Console.WriteLine (double.Parse ("1.234")); // 1234 (In Germany) This is because in Germany, the period indicates a thousands separator rather than a decimal point. Specifying invariant culture fixes this: double x = double.Parse ("1.234", CultureInfo.InvariantCulture); The same applies when calling ToString(): string x = 1.234.ToString (CultureInfo.InvariantCulture); Format Providers Sometimes you need more control over how formatting and parsing take place. There are dozens of ways to format a DateTime(Offset), for instance. Format pro- viders allow extensive control over formatting and parsing, and are supported for numeric types and date/times. Format providers are also used by user interface con- trols for formatting and parsing. The gateway to using a format provider is IFormattable. All numeric types—and DateTime(Offset)—implement this interface: public interface IFormattable { string ToString (string format, IFormatProvider formatProvider); } The first argument is the format string; the second is the format provider. The format string provides instructions; the format provider determines how the instructions are translated. For example: NumberFormatInfo f = new NumberFormatInfo(); f.CurrencySymbol = "$$"; Console.WriteLine (3.ToString ("C", f)); // $$ 3.00 228 | Chapter 6: Framework Fundamentals Here, "C" is a format string that indicates currency, and the NumberFormatInfo object is a format provider that determines how currency—and other numeric represen- tations—are rendered. This mechanism allows for globalization. All format strings for numbers and dates are listed in “Standard Format Strings and Parsing Flags” on page 233. If you specify a null format string or provider, a default is applied. The default format provider is CultureInfo.CurrentCulture, which, unless reassigned, reflects the com- puter’s runtime control panel settings. For example, on this computer: Console.WriteLine (10.3.ToString ("C", null)); // $10.30 For convenience, most types overload ToString such that you can omit a null provider: Console.WriteLine (10.3.ToString ("C")); // $10.30 Console.WriteLine (10.3.ToString ("F4")); // 10.3000 (Fix to 4 D.P.) Calling ToString on a DateTime(Offset) or a numeric type with no arguments is equivalent to using a default format provider, with an empty format string. The .NET Framework defines three format providers (all of which implement IFormatProvider): NumberFormatInfo DateTimeFormatInfo CultureInfo All enum types are also formattable, though there’s no special IFormatProvider class. Format providers and CultureInfo Within the context of format providers, CultureInfo acts as an indirection mecha- nism for the other two format providers, returning a NumberFormatInfo or DateTime FormatInfo object applicable to the culture’s regional settings. In the following example, we request a specific culture (english language in Great Britain): CultureInfo uk = CultureInfo.GetCultureInfo ("en-GB"); Console.WriteLine (3.ToString ("C", uk)); // £3.00 This executes using the default NumberFormatInfo object applicable to the en-GB culture. Formatting and Parsing | 229 Framework Fundamentals The next example formats a DateTime with invariant culture. Invariant culture is always the same, regardless of the computer’s settings: DateTime dt = new DateTime (2000, 1, 2); CultureInfo iv = CultureInfo.InvariantCulture; Console.WriteLine (dt.ToString (iv)); // 01/02/2000 00:00:00 Console.WriteLine (dt.ToString ("d", iv)); // 01/02/2000 Invariant culture is based on American culture, with the fol- lowing differences: • The currency symbol is ☼ instead of $. • Dates and times are formatted with leading zeros (though still with the month first). • Time uses the 24-hour format rather than an AM/PM designator. Using NumberFormatInfo or DateTimeFormatInfo In the next example, we instantiate a NumberFormatInfo and change the group sep- arator from a comma to a space. We then use it to format a number to three decimal places. NumberFormatInfo f = new NumberFormatInfo (); f.NumberGroupSeparator = " "; Console.WriteLine (12345.6789.ToString ("N3", f)); // 12 345.679 The initial settings for a NumberFormatInfo or DateTimeFormatInfo are based on the invariant culture. Sometimes, however, it’s more useful to choose a different starting point. To do this, you can Clone an existing format provider: NumberFormatInfo f = (NumberFormatInfo) CultureInfo.CurrentCulture.NumberFormat.Clone(); A cloned format provider is always writable—even if the original was read-only. Composite formatting Composite format strings allow you to combine variable substitution with format strings. The static string.Format method accepts a composite format string—we illustrated this in “String and Text Handling” on page 201: string composite = "Credit={0:C}"; Console.WriteLine (string.Format (composite, 500)); // Credit=$500.00 The Console class itself overloads its Write and WriteLine methods to accept com- posite format strings, allowing us to shorten this example slightly: Console.WriteLine ("Credit={0:C}", 500); // Credit=$500.00 You can also append a composite format string to a StringBuilder (via AppendFor mat), and to a TextWriter for I/O (see Chapter 15). 230 | Chapter 6: Framework Fundamentals string.Format accepts an optional format provider. A simple application for this is to call ToString on an arbitrary object while passing in a format provider. For example: string s = string.Format (CultureInfo.InvariantCulture, "{0}", someObject); This is equivalent to: string s; if (someObject is IFormattable) s = ((IFormattable)someObject).ToString (null, CultureInfo.InvariantCulture); else if (someObject == null) s = ""; else s = someObject.ToString(); Parsing with format providers There’s no standard interface for parsing through a format provider. Instead, each participating type overloads its static Parse (and TryParse) method to accept a format provider, and optionally, a NumberStyles or DateTimeStyles enum. NumberStyles and DateTimeStyles control how parsing work: they let you specify such things as whether parentheses or a currency symbol can appear in the input string. (By default, the answer to both of these questions is no.) For example: int error = int.Parse ("(2)"); // Exception thrown int minusTwo = int.Parse ("(2)", NumberStyles.Integer | NumberStyles.AllowParentheses); // OK decimal fivePointTwo = decimal.Parse ("£5.20", NumberStyles.Currency, CultureInfo.GetCultureInfo ("en-GB")); The next section lists all NumberStyles and DateTimeStyles members—as well as the default parsing rules for each type. IFormatProvider and ICustomFormatter All format providers implement IFormatProvider: public interface IFormatProvider { object GetFormat (Type formatType); } The purpose of this method is to provide indirection—this is what allows CultureInfo to defer to an appropriate NumberFormatInfo or DateTimeInfo object to do the work. By implementing IFormatProvider—along with ICustomFormatter—you can also write your own format provider that works in conjunction with existing types. ICus tomFormatter defines a single method as follows: string Format (string format, object arg, IFormatProvider formatProvider); The following custom format provider writes numbers as words: // Program can be downloaded from http://www.albahari.com/nutshell/ Formatting and Parsing | 231 Framework Fundamentals public class WordyFormatProvider : IFormatProvider, ICustomFormatter { static readonly string[] _numberWords = "zero one two three four five six seven eight nine minus point".Split(); IFormatProvider _parent; // Allows consumers to chain format providers public WordyFormatProvider () : this (CultureInfo.CurrentCulture) { } public WordyFormatProvider (IFormatProvider parent) { _parent = parent; } public object GetFormat (Type formatType) { if (formatType == typeof (ICustomFormatter)) return this; return null; } public string Format (string format, object arg, IFormatProvider prov) { // If it's not our format string, defer to the parent provider: if (arg == null || format != "W") return string.Format (_parent, "{0:" + format + "}", arg); StringBuilder result = new StringBuilder(); string digitList = string.Format (CultureInfo.InvariantCulture, "{0}", arg); foreach (char digit in digitList) { int i = "0123456789-.".IndexOf (digit); if (i == −1) continue; if (result.Length > 0) result.Append (' '); result.Append (_numberWords[i]); } return result.ToString(); } } Notice that in the Format method we used string.Format to convert the input number to a string—with InvariantCulture. It would have been much simpler just to call ToString() on arg, but then CurrentCulture would have been used instead. The reason for needing the invariant culture is evident a few lines later: int i = "0123456789-.".IndexOf (digit); It’s critical here that the number string comprises only the characters 0123456789-. and not any internationalized versions of these. Here’s an example of using WordyFormatProvider: double n = −123.45; IFormatProvider fp = new WordyFormatProvider(); Console.WriteLine (string.Format (fp, "{0:C} in words is {0:W}", n)); // -$123.45 in words is minus one two three point four five 232 | Chapter 6: Framework Fundamentals Custom format providers can be used only in composite format strings. Standard Format Strings and Parsing Flags The standard format strings control how a numeric type or DateTime/DateTimeOff set is converted to a string. There are two kinds of format strings: Standard format strings With these, you provide general guidance. A standard format string consists of a single letter, followed, optionally, by a digit (whose meaning depends on the letter). An example is "C" or "F2". Custom format strings With these, you micromanage every character with a template. An example is "0:#.000E+00". Custom format strings are unrelated to custom format providers. Numeric Format Strings Table 6-2 lists all standard numeric format strings. Table 6-2. Standard numeric format strings Letter Meaning Sample input Result Notes G or g “General” 1.2345, "G" 0.00001, "G" 0.00001, "g" 1.2345, "G3" 12345, "G3" 1.2345 1E-05 1e-05 1.23 1.23E04 Switches to exponential notation for small or large numbers G3 limits precision to three digits in total (before + after point) F Fixed point 2345.678, "F2" 2345.6, "F2" 2345.68 2345.60 F2 rounds to two decimal places N Fixed point with group separator (“Numeric”) 2345.678, "N2" 2345.6, "N2" 2,345.68 2,345.60 As above, with group (1000s) separator (de- tails from format pro- vider) D Pad with leading zeros 123, "D5" 123, "D1" 00123 123 For integral types only D5 pads left to five digits; does not truncate E or e Force exponential notation 56789, "E" 56789, "e" 56789, "E2" 5.678900E +004 5.678900e +004 5.68E+004 Six-digit default preci- sion Standard Format Strings and Parsing Flags | 233 Framework Fundamentals Letter Meaning Sample input Result Notes C Currency 1.2, "C" 1.2, "C4" $1.20 $1.2000 C with no digit uses de- fault number of D.P. from format provider P Percent .503, "P" .503, "P0" 50.30 % 50 % Uses symbol and layout from format provider Decimal places can op- tionally be overridden X or x Hexadecimal 47, "X" 47, "x" 47, "X4" 2F 2f 002F X for uppercase hex dig- its; x for lowercase hex digits Integrals only R Round-trip 1f / 3f, "R" 0.333333343 For the float and dou ble types, R squeezes out all digits to ensure exact round-tripping Supplying no numeric format string (or a null or blank string) is equivalent to using the "G" standard format string followed by no digit. This exhibits the following behavior: • Numbers smaller than 10−4 or larger than the type’s precision are expressed in exponential (scientific) notation. • The two decimal places at the limit of float or double’s precision are rounded away to mask the inaccuracies inherent in conversion to decimal from their underlying binary form. The automatic rounding just described is usually beneficial and goes unnoticed. However, it can cause trouble if you need to round-trip a number; in other words, convert it to a string and back again (maybe repeatedly) while preserving value equality. For this reason, the "R" format string exists to circumvent this implicit rounding. Table 6-3 lists custom numeric format strings. Table 6-3. Custom numeric format strings Specifier Meaning Sample input Result Notes # Digit placeholder 12.345, ".##" 12.345, ".####" 12.35 12.345 Limits digits after D.P. 0 Zero placeholder 12.345, ".00" 12.345, ".0000" 99, "000.00" 12.35 12.3500 099.00 As above, but also pads with zeros before and after D.P. 234 | Chapter 6: Framework Fundamentals Specifier Meaning Sample input Result Notes . Decimal point Indicates D.P. Actual symbol comes from NumberFormatInfo , Group separator 1234, "#,###,###" 1234, "0,000,000" 1,234 0,001,234 Symbol comes from NumberForma tInfo , (as above) Multiplier 1000000, "#," 1000000, "#,," 1000 1 If comma is at end or before D.P., it acts as a multiplier— dividing result by 1,000, 1,000,000, etc. Percent notation 0.6, "00%" 60% First multiplies by 100 and then substitutes percent symbol obtained from NumberFormatInfo E0, e0, E+0, e +0 E-0, e-0 Exponent notation 1234, "0E0" 1234, "0E+0" 1234, "0.00E00" 1234, "0.00e00" 1E3 1E+3 1.23E03 1.23e03 \ Literal character quote 50, @"\#0" #50 Use in conjunction with an @ prefix on the string—or use \ \ 'xx''xx ' Literal string quote 50, "0 '...'" 50 ... ; Section separator 15, "#;(#);zero" 15 (If positive) −5, "#;(#);zero" (5) (If negative) 0, "#;(#);zero" zero (If zero) Any other char Literal 35.2, "$0 . 00c" $35 . 20c NumberStyles Each numeric type defines a static Parse method that accepts a NumberStyles argu- ment. NumberStyles is a flags enum that lets you determine how the string is read as it’s converted to a numeric type. It has the following combinable members: AllowLeadingWhite AllowTrailingWhite AllowLeadingSign AllowTrailingSign AllowParentheses AllowDecimalPoint Standard Format Strings and Parsing Flags | 235 Framework Fundamentals AllowThousands AllowExponent AllowCurrencySymbol AllowHexSpecifier NumberStyles also defines these composite members: None Integer Float Number HexNumber Currency Any Except for None, all composite values include AllowLeadingWhite and AllowTrailing White. Their remaining makeup is shown in Figure 6-1, with the most useful three emphasized. Figure 6-1. Composite NumberStyles When you call Parse without specifying any flags, the defaults in Figure 6-2 are applied. Figure 6-2. Default parsing flags for numeric types 236 | Chapter 6: Framework Fundamentals If you don’t want the defaults shown in Figure 6-2, you must explicitly specify NumberStyles: int thousand = int.Parse ("3E8", NumberStyles.HexNumber); int minusTwo = int.Parse ("(2)", NumberStyles.Integer | NumberStyles.AllowParentheses); double aMillion = double.Parse ("1,000,000", NumberStyles.Any); decimal threeMillion = decimal.Parse ("3e6", NumberStyles.Any); decimal fivePointTwo = decimal.Parse ("$5.20", NumberStyles.Currency); Because we didn’t specify a format provider, this example works with your local currency symbol, group separator, decimal point, and so on. The next example is hardcoded to work with the euro sign and a blank group separator for currencies: NumberFormatInfo ni = new NumberFormatInfo(); ni.CurrencySymbol = "₠"; ni.CurrencyGroupSeparator = " "; double million = double.Parse ("₠1 000 000", NumberStyles.Currency, ni); Date/Time Format Strings Format strings for DateTime/DateTimeOffset can be divided into two groups, based on whether they honor culture and format provider settings. Those that do are listed in Table 6-4; those that don’t are listed in Table 6-5. The sample output comes from formatting the following DateTime (with invariant culture, in the case of Table 6-4): new DateTime (2000, 1, 2, 17, 18, 19); Table 6-4. Culture-sensitive date/time format strings Format string Meaning Sample output d Short date 01/02/2000 D Long date Sunday, 02 January 2000 t Short time 17:18 T Long time 17:18:19 f Long date + short time Sunday, 02 January 2000 17:18 F Long date + long time Sunday, 02 January 2000 17:18:19 g Short date + short time 01/02/2000 17:18 G (default) Short date + long time 01/02/2000 17:18:19 m, M Month and day January 02 y, Y Year and month 2000 January Standard Format Strings and Parsing Flags | 237 Framework Fundamentals Table 6-5. Culture-insensitive date/time format strings Format string Meaning Sample output Notes o Round-trippable 2000-01-02T17:18:19.0000000 Will append time zone informa- tion unless DateTimeKind is Unspecified r, R RFC 1123 standard Sun, 02 Jan 2000 17:18:19 GMT You must explicitly convert to UTC with DateTime.ToUni versalTime s Sortable; ISO 8601 2000-01-02T17:18:19 Compatible with text-based sorting u “Universal” sortable 2000-01-02 17:18:19Z Similar to above; must explicitly convert to UTC U UTC Sunday, 02 January 2000 17:18:19 Long date + short time, con- verted to UTC The format strings "r", "R", and "u" emit a suffix that implies UTC; yet they don’t automatically convert a local to a UTC DateTime (so you must do the conversion yourself). Ironically, "U" automatically converts to UTC, but doesn’t write a time zone suffix! In fact, "o" is the only format specifier in the group that can write an unambiguous DateTime without intervention. DateTimeFormatInfo also supports custom format strings: these are analogous to numeric custom format strings. The list is fairly exhaustive and you can find it in the MSDN. An example of a custom format string is: yyyy-MM-dd HH:mm:ss Parsing and misparsing DateTimes Strings that put the month or day first are ambiguous and can easily be misparsed —particularly if you or any of your customers live outside the United States. This is not a problem in user interface controls because the same settings are in force when parsing as when formatting. But when writing to a file, for instance, day/month misparsing can be a real problem. There are two solutions: • Always state the same explicit culture when formatting and parsing (e.g., in- variant culture). • Format DateTime and DateTimeOffsets in a manner independent of culture. The second approach is more robust—particularly if you choose a format that puts the four-digit year first: such strings are much harder to misparse by another party. Further, strings formatted with a standards-compliant year-first format (such as "o") can parse correctly alongside locally formatted strings—rather like a “universal donor.” (Dates formatted with "s" or "u" have the further benefit of being sortable.) To illustrate, suppose we generate a culture-insensitive DateTime string s as follows: string s = DateTime.Now.ToString ("o"); 238 | Chapter 6: Framework Fundamentals The "o" format string includes milliseconds in the output. The following custom format string gives the same result as "o", but without milliseconds: yyyy-MM-ddTHH:mm:ss K We can reparse this in two ways. ParseExact demands strict compliance with the specified format string: DateTime dt1 = DateTime.ParseExact (s, "o", null); (You can achieve a similar result with XmlConvert’s ToString and ToDateTime methods.) Parse, however, implicitly accepts both the "o" format and the CurrentCulture format: DateTime dt2 = DateTime.Parse (s); This works with both DateTime and DateTimeOffset. ParseExact is usually preferable if you know the format of the string that you’re parsing. It means that if the string is incorrectly formatted, an exception will be thrown—which is usually better than risking a misparsed date. DateTimeStyles DateTimeStyles is a flags enum that provides additional instructions when calling Parse on a DateTime(Offset). Here are its members: None, AllowLeadingWhite, AllowTrailingWhite, AllowInnerWhite, AssumeLocal, AssumeUniversal, AdjustToUniversal, NoCurrentDateDefault, RoundTripKind There is also a composite member, AllowWhiteSpaces: AllowWhiteSpaces = AllowLeadingWhite | AllowTrailingWhite | AllowInnerWhite The default is None. This means that extra whitespace is normally prohibited (white- space that’s part of a standard DateTime pattern is exempt). AssumeLocal and AssumeUniversal apply if the string doesn’t have a time zone suffix (such as Z or +9:00). AdjustToUniversal still honors time zone suffixes, but then converts to UTC using the current regional settings. If you parse a string comprising a time but no date, today’s date is applied by default. If you apply the NoCurrentDateDefault flag, however, it instead uses 1st January 0001. Standard Format Strings and Parsing Flags | 239 Framework Fundamentals Enum Format Strings In “Enums” in Chapter 3, we describe formatting and parsing enum values. Table 6-6 lists each format string and the result of applying it to the following expression: Console.WriteLine (System.ConsoleColor.Red.ToString (formatString)); Table 6-6. Enum format strings Format string Meaning Sample output Notes G or g “General” Red Default F or f Treat as though Flags attribute were present Red Works on combined members even if enum has no Flags attribute D or d Decimal value 12 Retrieves underlying integral value X or x Hexadecimal value 0000000C Retrieves underlying integral value Other Conversion Mechanisms In the previous two sections, we covered format providers—.NET’s primary mech- anism for formatting and parsing. Other important conversion mechanisms are scattered through various types and namespaces. Some convert to and from string, and some do other kinds of conversions. In this section, we discuss the following topics: • The Convert class and its functions: — Real to integral conversions that round rather than truncate — Parsing numbers in base 2, 8, and 16 — Dynamic conversions — Base 64 conversions • XmlConvert and its role in formatting and parsing for XML • Type converters and their role in formatting and parsing for designers and XAML • BitConverter, for binary conversions Convert The .NET Framework calls the following types base types: • bool, char, string, System.DateTime, and System.DateTimeOffset • All of the C# numeric types The static Convert class defines methods for converting every base type to every other base type. Unfortunately, most of these methods are useless: either they throw exceptions or they are redundant alongside implicit casts. Among the clutter, how- ever, are some useful methods, listed in the following sections. 240 | Chapter 6: Framework Fundamentals All base types (explicitly) implement IConvertible, which defines methods for converting to every other base type. In most cases, the implementation of each of these methods simply calls a method in Convert. On rare occasions, it can be useful to write a method that accepts an argument of type IConvertible. Rounding real to integral conversions In Chapter 2, we saw how implicit and explicit casts allow you to convert between numeric types. In summary: • Implicit casts work for nonlossy conversions (e.g., int to double). • Explicit casts are required for lossy conversions (e.g., double to int). Casts are optimized for efficiency; hence, they truncate data that won’t fit. This can be a problem when converting from a real number to an integer, because often you want to round rather than truncate. Convert’s numerical conversion methods address just this issue; they always round: double d = 3.9; int i = Convert.ToInt32 (d); // i == 4 Convert uses banker’s rounding, which snaps midpoint values to even integers (this avoids positive or negative bias). If banker’s rounding is a problem, first call Math.Round on the real number: this accepts an additional argument that allows you to control midpoint rounding. Parsing numbers in base 2, 8, and 16 Hidden among the To(integral-type) methods are overloads that parse numbers in another base: int thirty = Convert.ToInt32 ("1E", 16); // Parse in hexadecimal uint five = Convert.ToUInt32 ("101", 2); // Parse in binary The second argument specifies the base. It can be any base you like—as long as it’s 2, 8, 10, or 16! Dynamic conversions Occasionally, you need to convert from one type to another—but you don’t know what the types are until runtime. For this, the Convert class provides a ChangeType method: public static object ChangeType (object value, Type conversionType); The source and target types must be one of the “base” types. ChangeType also accepts an optional IFormatProvider argument. Here’s an example: Type targetType = typeof (int); object source = "42"; object result = Convert.ChangeType (source, targetType); Other Conversion Mechanisms | 241 Framework Fundamentals Console.WriteLine (result); // 42 Console.WriteLine (result.GetType()); // System.Int32 An example of when this might be useful is in writing a deserializer that can work with multiple types. It can also convert any enum to its integral type (see “Enums” on page 102 in Chapter 3). A limitation of ChangeType is that you cannot specify a format string or parsing flag. Base 64 conversions Sometimes you need to include binary data such as a bitmap within a text document such as an XML file or email message. Base 64 is a ubiquitous means of encoding binary data as readable characters, using 64 characters from the ASCII set. Convert’s ToBase64String method converts from a byte array to base 64; From Base64String does the reverse. XmlConvert If you’re dealing with data that’s originated from or destined for an XML file, XmlCon vert (the System.Xml namespace) provides the most suitable methods for formatting and parsing. The methods in XmlConvert handle the nuances of XML formatting without needing special format strings. For instance, true in XML is “true” and not “True”. The .NET Framework internally uses XmlConvert extensively. XmlConvert is also good for general-purpose culture-independent serialization. The formatting methods in XmlConvert are all provided as overloaded ToString methods; the parsing methods are called ToBoolean, ToDateTime, and so on. For example: string s = XmlConvert.ToString (true); // s = "true" bool isTrue = XmlConvert.ToBoolean (s); The methods that convert to and from DateTime accept an XmlDateTimeSerializa tionMode argument. This is an enum with the following values: Unspecified, Local, Utc, RoundtripKind Local and Utc cause a conversion to take place when formatting (if the DateTime is not already in that time zone). The time zone is then appended to the string: 2010-02-22T14:08:30.9375 // Unspecified 2010-02-22T14:07:30.9375+09:00 // Local 2010-02-22T05:08:30.9375Z // Utc Unspecified strips away any time zone information embedded in the DateTime (i.e., DateTimeKind) before formatting. RoundtripKind honors the DateTime’s DateTimeKind —so when it’s reparsed, the resultant DateTime struct will be exactly as it was originally. 242 | Chapter 6: Framework Fundamentals Type Converters Type converters are designed to format and parse in design-time environments. They also parse values in XAML (Extensible Application Markup Language) documents —as used in Windows Presentation Foundation and Workflow Foundation. In the .NET Framework, there are more than 100 type converters—covering such things as colors, images, and URIs. In contrast, format providers are implemented for only a handful of simple value types. Type converters typically parse strings in a variety of ways—without needing hints. For instance, in an ASP.NET application in Visual Studio, if you assign a control a BackColor by typing "Beige" into the property window, Color’s type converter figures out that you’re referring to a color name and not an RGB string or system color. This flexibility can sometimes make type converters useful in contexts outside of design- ers and XAML documents. All type converters subclass TypeConverter in System.ComponentModel. To obtain a TypeConverter, call TypeDescriptor.GetConverter. The following obtains a Type Converter for the Color type (in the System.Drawing namespace, System.Drawing.dll): TypeConverter cc = TypeDescriptor.GetConverter (typeof (Color)); Among many other methods, TypeConverter defines methods to ConvertToString and ConvertFromString. We can call these as follows: Color beige = (Color) cc.ConvertFromString ("Beige"); Color purple = (Color) cc.ConvertFromString ("#800080"); Color window = (Color) cc.ConvertFromString ("Window"); By convention, type converters have names ending in Converter and are usually in the same namespace as the type they’re converting. A type links to its converter via a TypeConverterAttribute, allowing designers to pick up converters automatically. Type converters can also provide design-time services such as generating standard value lists for populating a drop-down list in a designer or assisting with code serialization. BitConverter Most base types can be converted to a byte array, by calling BitConverter.GetBytes: foreach (byte b in BitConverter.GetBytes (3.5)) Console.Write (b + " "); // 0 0 0 0 0 0 12 64 BitConverter also provides methods for converting in the other direction, such as ToDouble. The decimal and DateTime(Offset) types are not supported by BitConverter. You can, however, convert a decimal to an int array by calling decimal.GetBits. To go the other way around, decimal provides a constructor that accepts an int array. In the case of DateTime, you can call ToBinary on an instance—this returns a long (upon which you can then use BitConverter). The static DateTime.FromBinary method does the reverse. Other Conversion Mechanisms | 243 Framework Fundamentals Globalization There are two aspects to internationalizing an application: globalization and localization. Globalization is concerned with three tasks (in decreasing order of importance): 1. Making sure that your program doesn’t break when run in another culture 2. Respecting a local culture’s formatting rules—for instance, when displaying dates 3. Designing your program so that it picks up culture-specific data and strings from satellite assemblies that you can later write and deploy Localization means concluding that last task by writing satellite assemblies for spe- cific cultures. This can be done after writing your program—we cover the details in “Resources and Satellite Assemblies” on page 745 in Chapter 18. The .NET Framework helps you with the second task by applying culture-specific rules by default. We’ve already seen how calling ToString on a DateTime or number respects local formatting rules. Unfortunately, this makes it easy to fail the first task and have your program break because you’re expecting dates or numbers to be for- matted according to an assumed culture. The solution, as we’ve seen, is either to specify a culture (such as the invariant culture) when formatting and parsing, or to use culture-independent methods such as those in XmlConvert. Globalization Checklist We’ve already covered the important points in this chapter. Here’s a summary of the essential work required: • Understand Unicode and text encodings (see “Text Encodings and Uni- code” on page 211). • Be mindful that methods such as ToUpper and ToLower on char and string are culture-sensitive: use ToUpperInvariant/ToLowerInvariant unless you want cul- ture sensitivity. • Favor culture-independent formatting and parsing mechanisms for DateTime and DateTimeOffsets such as ToString("o") and XmlConvert. • Otherwise, specify a culture when formatting/parsing numbers or date/times (unless you want local-culture behavior). Testing You can test against different cultures by reassigning Thread’s CurrentCulture prop- erty (in System.Threading). The following changes the current culture to Turkey: Thread.CurrentThread.CurrentCulture = CultureInfo.GetCultureInfo ("tr-TR"); 244 | Chapter 6: Framework Fundamentals Turkey is a particularly good test case because: • "i".ToUpper() != "I" and "I".ToLower() != "i". • Dates are formatted as day/month/year, and with a period separator. • The decimal point indicator is a comma instead of a period. You can also experiment by changing the number and date formatting settings in the Windows Control Panel: these are reflected in the default culture (CultureInfo.CurrentCulture). CultureInfo.GetCultures() returns an array of all available cultures. Thread and CultureInfo also support a CurrentUICulture prop- erty. This is concerned more about localization: we cover this in Chapter 18. Working with Numbers Conversions We covered numeric conversions in previous chapters and sections; Table 6-7 sum- marizes all the options. Table 6-7. Summary of numeric conversions Task Functions Examples Parsing base 10 numbers Parse TryParse double d = double.Parse ("3.5"); int i; bool ok = int.TryParse ("3", out i); Parsing from base 2, 8, or 16 Convert.ToIntegral int i = Convert.ToInt32 ("1E", 16); Formatting to hexadecimal ToString ("X") string hex = 45.ToString ("X"); Lossless numeric conversion Implicit cast int i = 23; double d = d; Truncating numeric conversion Explicit cast double d = 23.5; int i = (int) d; Rounding numeric conversion (real to integral) Convert.ToIntegral double d = 23.5; int i = Convert.ToInt32 (d); Math Table 6-8 lists the members of the static Math class. The trigonometric functions accept arguments of type double; other methods such as Max are overloaded to operate on all numeric types. The Math class also defines the mathematical constants E (e) and PI. Working with Numbers | 245 Framework Fundamentals Table 6-8. Methods in the static Math class Category Methods Rounding Round, Truncate, Floor, Ceiling Maximum/minimum Max, Min Absolute value and sign Abs, Sign Square root Sqrt Raising to a power Pow, Exp Logarithm Log, Log10 Trigonometric Sin, Cos, Tan Sinh, Cosh, Tanh Asin, Acos, Atan The Round method lets you specify the number of decimal places with which to round, as well as how to handle midpoints (away from zero, or with banker’s round- ing). Floor and Ceiling round to the nearest integer: Floor always rounds down and Ceiling always rounds up—even with negative numbers. Max and Min accept only two arguments. If you have an array or sequence of numbers, use the Max and Min extension methods in System.Linq.Enumerable. BigInteger The BigInteger struct is a specialized numeric type introduced in .NET Framework 4.0. It lives in the new System.Numerics namespace in System.Numerics.dll and allows you to represent an arbitrarily large integer without any loss of precision. C# doesn’t provide native support for BigInteger, so there’s no way to represent BigInteger literals. You can, however, implicitly convert from any other integral type to a BigInteger. For instance: BigInteger twentyFive = 25; // implicit conversion from integer To represent a bigger number, such as one googol (10100), you can use one of BigIn teger’s static methods, such as Pow (raise to the power): BigInteger googol = BigInteger.Pow (10, 100); Alternatively, you can Parse a string: BigInteger googol = BigInteger.Parse ("1".PadRight (100, '0')); Calling ToString() on this prints every digit: Console.WriteLine (googol.ToString()); // 10000000000000000000000000000 00000000000000000000000000000000000000000000000000000000000000000000000 246 | Chapter 6: Framework Fundamentals You can perform potentially lossy conversions between BigInteger and the standard numeric types with the explicit cast operator: double g2 = (double) googol; // Explicit cast BigInteger g3 = (BigInteger) g2; // Explicit cast Console.WriteLine (g3); The output from this demonstrates the loss of precision: 9999999999999999673361688041166912... BigInteger overloads all the arithmetic operators including remainder (%), as well as the comparison and equality operators. You can also construct a BigInteger from a byte array. The following code generates a 32-byte random number suitable for cryptography and then assigns it to a BigIn teger: // This uses the System.Security.Cryptography namespace: RandomNumberGenerator rand = RandomNumberGenerator.Create(); byte[] bytes = new byte [32]; rand.GetBytes (bytes); var bigRandomNumber = new BigInteger (bytes); // Convert to BigInteger The advantage of storing such a number in a BigInteger over a byte array is that you get value-type semantics. Calling ToByteArray converts a BigInteger back to a byte array. Complex The Complex struct is another specialized numeric type new to Framework 4.0, and is for representing complex numbers with real and imaginary components of type double. Complex resides in the System.Numerics.dll assembly (along with BigInteger). To use Complex, instantiate the struct, specifying the real and imaginary values: var c1 = new Complex (2, 3.5); var c2 = new Complex (3, 0); There are also implicit conversions from the standard numeric types. The Complex struct exposes properties for the real and imaginary values, as well as the phase and magnitude: Console.WriteLine (c1.Real); // 2 Console.WriteLine (c1.Imaginary); // 3.5 Console.WriteLine (c1.Phase); // 1.05165021254837 Console.WriteLine (c1.Magnitude); // 4.03112887414927 You can also construct a Complex number by specifying magnitude and phase: Complex c3 = Complex.FromPolarCoordinates (1.3, 5); The standard arithmetic operators are overloaded to work on Complex numbers: Console.WriteLine (c1 + c2); // (5, 3.5) Console.WriteLine (c1 * c2); // (6, 10.5) Working with Numbers | 247 Framework Fundamentals The Complex struct exposes static methods for more advanced functions, including: • Trigonometric (Sin, Asin, Sinh, Tan, etc.) • Logarithms and exponentiations • Conjugate Random The Random class generates a pseudorandom sequence of random bytes, integers, or doubles. To use Random, you first instantiate it, optionally providing a seed to initiate the random number series. Using the same seed guarantees the same series of numbers (if run under the same CLR version), which is sometimes useful when you want reproducibility: Random r1 = new Random (1); Random r2 = new Random (1); Console.WriteLine (r1.Next (100) + ", " + r1.Next (100)); // 24, 11 Console.WriteLine (r2.Next (100) + ", " + r2.Next (100)); // 24, 11 If you don’t want reproducibility, you can construct Random with no seed—then it uses the current system time to make one up. Because the system clock has limited granularity, two Random instances created close together (typically within 10 ms) will yield the same sequence of values. A common trap is to instan- tiate a new Random object every time you need a random number, rather than reusing the same object. A good pattern is to declare a single static Random instance. In multithreaded scenarios, however, this can cause trouble because Random objects are not thread-safe. We describe a work- around in “Thread-Local Storage” in Chapter 22. Calling Next(n) generates a random integer between 0 and n−1. NextDouble generates a random double between 0 and 1. NextBytes fills a byte array with random values. Random is not considered random enough for high-security applications, such as cryptography. For this, the .NET Framework provides a cryptographically strong random number generator, in the System.Security.Cryptography namespace. Here’s how it’s used: var rand = System.Security.Cryptography.RandomNumberGenerator.Create(); byte[] bytes = new byte [32]; rand.GetBytes (bytes); // Fill the byte array with random numbers. The downside is that it’s less flexible: filling a byte array is the only means of obtaining random numbers. To obtain an integer, you must use BitConverter: byte[] bytes = new byte [4]; rand.GetBytes (bytes); int i = BitConverter.ToInt32 (bytes, 0); 248 | Chapter 6: Framework Fundamentals Enums In Chapter 3, we described C#’s enum type, and showed how to combine members, test equality, use logical operators, and perform conversions. The Framework extends C#’s support for enums through the System.Enum type. This type has two roles: • Providing type unification for all enum types • Defining static utility methods Type unification means you can implicitly cast any enum member to a System.Enum instance: enum Nut { Walnut, Hazelnut, Macadamia } enum Size { Small, Medium, Large } static void Main() { Display (Nut.Macadamia); // Nut.Macadamia Display (Size.Large); // Size.Large } static void Display (Enum value) { Console.WriteLine (value.GetType().Name + "." + value.ToString()); } The static utility methods on System.Enum are primarily related to performing con- versions and obtaining lists of members. Enum Conversions There are three ways to represent an enum value: • As an enum member • As its underlying integral value • As a string In this section, we describe how to convert between each. Enum to integral conversions Recall that an explicit cast converts between an enum member and its integral value. An explicit cast is the correct approach if you know the enum type at compile time: [Flags] public enum BorderSides { Left=1, Right=2, Top=4, Bottom=8 } ... int i = (int) BorderSides.Top; // i == 4 BorderSides side = (BorderSides) i; // side == BorderSides.Top Enums | 249 Framework Fundamentals You can cast a System.Enum instance to its integral type in the same way. The trick is to first cast to an object, and then the integral type: static int GetIntegralValue (Enum anyEnum) { return (int) (object) anyEnum; } This relies on you knowing the integral type: the method we just wrote would crash if passed an enum whose integral type was long. To write a method that works with an enum of any integral type, you can take one of three approaches. The first is to call Convert.ToDecimal: static decimal GetAnyIntegralValue (Enum anyEnum) { return Convert.ToDecimal (anyEnum); } This works because every integral type (including ulong) can be converted to decimal without loss of information. The second approach is to call Enum.GetUnderlying Type in order to obtain the enum’s integral type, and then call Convert.ChangeType: static object GetBoxedIntegralValue (Enum anyEnum) { Type integralType = Enum.GetUnderlyingType (anyEnum.GetType()); return Convert.ChangeType (anyEnum, integralType); } This preserves the original integral type, as the following example shows: object result = GetBoxedIntegralValue (BorderSides.Top); Console.WriteLine (result); // 4 Console.WriteLine (result.GetType()); // System.Int32 Our GetBoxedIntegralType method in fact performs no value conversion; rather, it reboxes the same value in another type. It translates an integral value in enum-type clothing to an integral value in integral-type clothing. We describe this further in “How Enums Work.” The third approach is to call Format or ToString specifying the "d" or "D" format string. This gives you the enum’s integral value as a string, and it is useful when writing custom serialization formatters: static string GetIntegralValueAsString (Enum anyEnum) { return anyEnum.ToString ("D"); // returns something like "4" } Integral to enum conversions Enum.ToObject converts an integral value to an enum instance of the given type: object bs = Enum.ToObject (typeof (BorderSides), 3); Console.WriteLine (bs); // Left, Right 250 | Chapter 6: Framework Fundamentals This is the dynamic equivalent of this: BorderSides bs = (BorderSides) 3; ToObject is overloaded to accept all integral types, as well as object. (The latter works with any boxed integral type.) String conversions To convert an enum to a string, you can either call the static Enum.Format method or call ToString on the instance. Each method accepts a format string, which can be "G" for default formatting behavior, "D" to emit the underlying integral value as a string, "X" for the same in hexadecimal, or "F" to format combined members of an enum without the Flags attribute. We listed examples of these in “Standard Format Strings and Parsing Flags.” Enum.Parse converts a string to an enum. It accepts the enum type and a string that can include multiple members: BorderSides leftRight = (BorderSides) Enum.Parse (typeof (BorderSides), "Left, Right"); An optional third argument lets you perform case-insensitive parsing. An ArgumentException is thrown if the member is not found. Enumerating Enum Values Enum.GetValues returns an array comprising all members of a particular enum type: foreach (Enum value in Enum.GetValues (typeof (BorderSides))) Console.WriteLine (value); Composite members such as LeftRight = Left | Right are included too. Enum.GetNames performs the same function, but returns an array of strings. Internally, the CLR implements GetValues and GetNames by reflecting over the fields in the enum’s type. The results are cached for efficiency. How Enums Work The semantics of enums are enforced largely by the compiler. In the CLR, there’s no runtime difference between an enum instance (when unboxed) and its underlying integral value. Further, an enum definition in the CLR is merely a subtype of Sys tem.Enum with static integral-type fields for each member. This makes the ordinary use of an enum highly efficient, with a runtime cost matching that of integral constants. Enums | 251 Framework Fundamentals The downside of this strategy is that enums can provide static but not strong type safety. We saw an example of this in Chapter 3: public enum BorderSides { Left=1, Right=2, Top=4, Bottom=8 } ... BorderSides b = BorderSides.Left; b += 1234; // No error! When the compiler is unable to perform validation (as in this example), there’s no backup from the runtime to throw an exception. What we said about there being no runtime difference between an enum instance and its integral value might seem at odds with the following: [Flags] public enum BorderSides { Left=1, Right=2, Top=4, Bottom=8 } ... Console.WriteLine (BorderSides.Right.ToString()); // Right Console.WriteLine (BorderSides.Right.GetType().Name); // BorderSides Given the nature of an enum instance at runtime, you’d expect this to print 2 and Int32! The reason for its behavior is down to some more compile-time trickery. C# explicitly boxes an enum instance before calling its virtual methods—such as ToString or GetType. And when an enum instance is boxed, it gains a runtime wrap- ping that references its enum type. Tuples Framework 4.0 introduced a new set of generic classes for holding a set of differently typed elements. These are called tuples: public class Tuple public class Tuple public class Tuple public class Tuple public class Tuple public class Tuple public class Tuple public class Tuple Each has read-only properties called Item1, Item2, and so on (one for each type parameter). You can instantiate a tuple either via its constructor: var t = new Tuple (123, "Hello"); or via the static helper method Tuple.Create: Tuple t = Tuple.Create (123, "Hello"); The latter leverages generic type inference. You can combine this with implicit typ- ing: var t = Tuple.Create (123, "Hello"); You can then access the properties as follows (notice that each is statically typed): 252 | Chapter 6: Framework Fundamentals Console.WriteLine (t.Item1 * 2); // 246 Console.WriteLine (t.Item2.ToUpper()); // HELLO Tuples are convenient in returning more than one value from a method—or creating collections of value pairs (we’ll cover collections in the following chapter). An alternative to tuples is to use an object array. However, you then lose static type safety, incur the cost of boxing/unboxing for value types, and require clumsy casts that cannot be validated by the compiler: object[] items = { 123, "Hello" }; Console.WriteLine ( ((int) items[0]) * 2 ); // 246 Console.WriteLine ( ((string) items[1]).ToUpper() ); // HELLO Comparing Tuples Tuples are classes (and therefore reference types). In keeping with this comparing two distinct instances with the equality operator returns false. However, the Equals method is overridden to compare each individual element instead: var t1 = Tuple.Create (123, "Hello"); var t2 = Tuple.Create (123, "Hello"); Console.WriteLine (t1 == t2); // False Console.WriteLine (t1.Equals (t2)); // True You can also pass in a custom equality comparer (by virtue of tuples implementing IStructuralEquatable). We cover equality and order comparison later in this chapter. The Guid Struct The Guid struct represents a globally unique identifier: a 16-byte value that, when generated, is almost certainly unique in the world. Guids are often used for keys of various sorts—in applications and databases. There are 2128 or 3.4 × 1038 unique Guids. The static Guid.NewGuid method generates a unique Guid: Guid g = Guid.NewGuid (); Console.WriteLine (g.ToString()); // 0d57629c-7d6e-4847-97cb-9e2fc25083fe To instantiate an existing value, you use one of the constructors. The two most useful constructors are: public Guid (byte[] b); // Accepts a 16-byte array public Guid (string g); // Accepts a formatted string When represented as a string, a Guid is formatted as a 32-digit hexadecimal number, with optional hyphens after the 8th, 12th, 16th, and 20th digits. The whole string can also be optionally wrapped in brackets or braces: Guid g1 = new Guid ("{0d57629c-7d6e-4847-97cb-9e2fc25083fe}"); Guid g2 = new Guid ("0d57629c7d6e484797cb9e2fc25083fe"); Console.WriteLine (g1 == g2); // True The Guid Struct | 253 Framework Fundamentals Being a struct, a Guid honors value-type semantics; hence, the equality operator works in the preceding example. The ToByteArray method converts a Guid to a byte array. The static Guid.Empty property returns an empty Guid (all zeros). This is often used in place of null. Equality Comparison Until now, we’ve assumed that the == and != operators are all there is to equality comparison. The issue of equality, however, is more complex and subtler, sometimes requiring the use of additional methods and interfaces. This section explores the standard C# and .NET protocols for equality, focusing particularly on two questions: • When are == and != adequate—and inadequate—for equality comparison, and what are the alternatives? • How and when should you customize a type’s equality logic? But before exploring the details of equality protocols and how to customize them, we must first look at the preliminary concept of value versus referential equality. Value Versus Referential Equality There are two kinds of equality: Value equality Two values are equivalent in some sense. Referential equality Two references refer to exactly the same object. By default: • Value types use value equality. • Reference types use referential equality. Value types, in fact, can only use value equality (unless boxed). A simple demon- stration of value equality is to compare two numbers: int x = 5, y = 5; Console.WriteLine (x == y); // True (by virtue of value equality) A more elaborate demonstration is to compare two DateTimeOffset structs. The fol- lowing prints True because the two DateTimeOffsets refer to the same point in time and so are considered equivalent: var dt1 = new DateTimeOffset (2010, 1, 1, 1, 1, 1, TimeSpan.FromHours(8)); var dt2 = new DateTimeOffset (2010, 1, 1, 2, 1, 1, TimeSpan.FromHours(9)); Console.WriteLine (dt1 == dt2); // True 254 | Chapter 6: Framework Fundamentals DateTimeOffset is a struct whose equality semantics have been tweaked. By default, structs exhibit a special kind of value equality called structural equality, where two values are consid- ered equal if all of their members are equal. (You can see this by creating a struct and calling its Equals method; more on this later.) Reference types exhibit referential equality by default. In the following example, f1 and f2 are not equal—despite their objects having identical content: class Foo { public int X; } ... Foo f1 = new Foo { X = 5 }; Foo f2 = new Foo { X = 5 }; Console.WriteLine (f1 == f2); // False In contrast, f3 and f1 are equal because they reference the same object: Foo f3 = f1; Console.WriteLine (f1 == f3); // True We’ll explain later in this section how reference types can be customized to exhibit value equality. An example of this is the Uri class in the System namespace: Uri uri1 = new Uri ("http://www.linqpad.net"); Uri uri2 = new Uri ("http://www.linqpad.net"); Console.WriteLine (uri1 == uri2); // True Standard Equality Protocols There are three standard protocols that types can implement for equality comparison: • The == and != operators • The virtual Equals method in object • The IEquatable interface In addition, there are the pluggable protocols and the IStructuralEquatable interface which we describe in Chapter 7. == and != We’ve already seen in many examples how the standard == and != operators perform equality/inequality comparisons. The subtleties with == and != arise because they are operators, and so are statically resolved (in fact, they are implemented as static functions). So, when you use == or !=, C# makes a compile-time decision as to which type will perform the comparison, and no virtual behavior comes into play. This is normally desirable. In the following example, the compiler hard-wires == to the int type because x and y are both int: int x = 5; int y = 5; Console.WriteLine (x == y); // True Equality Comparison | 255 Framework Fundamentals But in the next example, the compiler wires the == operator to the object type: object x = 5; object y = 5; Console.WriteLine (x == y); // False Because object is a class (and so a reference type), object’s == operator uses refer- ential equality to compare x and y. The result is false, because x and y each refer to different boxed objects on the heap. The virtual Object.Equals method To correctly equate x and y in the preceding example, we can use the virtual Equals method. Equals is defined in System.Object, and so is available to all types: object x = 5; object y = 5; Console.WriteLine (x.Equals (y)); // True Equals is resolved at runtime—according to the object’s actual type. In this case, it calls Int32’s Equals method, which applies value equality to the operands, returning true. With reference types, Equals performs referential equality comparison by default; with structs, Equals performs structural comparison by calling Equals on each of its fields. Why the Complexity? You might wonder why the designers of C# didn’t avoid the problem by making == virtual, and so functionally identical to Equals. There are three reasons for this: • If the first operand is null, Equals fails with a NullReferenceException; a static operator does not. • Because the == operator is statically resolved, it executes extremely quickly. This means that you can write computationally intensive code without penalty—and without needing to learn another language such as C++. • Sometimes it can be useful to have == and Equals apply different definitions of equality. We describe this scenario later in this section. Essentially, the complexity of the design reflects the complexity of the situation: the concept of equality covers a multitude of scenarios. Hence, Equals is suitable for equating two objects in a type-agnostic fashion. The following method equates two objects of any type: public static bool AreEqual (object obj1, object obj2) { return obj1.Equals (obj2); } There is one case, however, in which this fails. If the first argument is null, you get a NullReferenceException. Here’s the fix: public static bool AreEqual (object obj1, object obj2) { 256 | Chapter 6: Framework Fundamentals if (obj1 == null) return obj2 == null; return obj1.Equals (obj2); } The static object.Equals method The object class provides a static helper method that does the work of AreEqual in the preceding example. Its name is Equals—just like the virtual method—but there’s no conflict because it accepts two arguments: public static bool Equals (object objA, object objB) This provides a null-safe equality comparison algorithm for when the types are unknown at compile time. For example: object x = 3, y = 3; Console.WriteLine (object.Equals (x, y)); // True x = null; Console.WriteLine (object.Equals (x, y)); // False y = null; Console.WriteLine (object.Equals (x, y)); // True A useful application is when writing generic types. The following code will not compile if object.Equals is replaced with the == or != operator: class Test { T _value; public void SetValue (T newValue) { if (!object.Equals (newValue, _value)) { _value = newValue; OnValueChanged(); } } protected virtual void OnValueChanged() { ... } } Operators are prohibited here because the compiler cannot bind to the static method of an unknown type. A more elaborate way to implement this comparison is with the EqualityComparer class. This has the advantage of avoiding boxing: if (!EqualityComparer.Default.Equals (newValue, _value)) We discuss EqualityComparer in more detail in Chapter 7 (see “Plugging in Equality and Order” on page 312). Equality Comparison | 257 Framework Fundamentals The static object.ReferenceEquals method Occasionally, you need to force referential equality comparison. The static object.ReferenceEquals method does just this: class Widget { ... } class Test { static void Main() { Widget w1 = new Widget(); Widget w2 = new Widget(); Console.WriteLine (object.ReferenceEquals (w1, w2)); // False } } You might want to do this because it’s possible for Widget to override the virtual Equals method, such that w1.Equals(w2) would return true. Further, it’s possible for Widget to overload the == operator so that w1==w2 would also return true. In such cases, calling object.ReferenceEquals guarantees normal referential equality semantics. Another way to force referential equality comparison is to cast the values to object and then apply the == operator. The IEquatable interface A consequence of calling object.Equals is that it forces boxing on value types. This is undesirable in highly performance-sensitive scenarios because boxing is relatively expensive compared to the actual comparison. A solution was introduced in C# 2.0, with the IEquatable interface: public interface IEquatable { bool Equals (T other); } The idea is that IEquatable, when implemented, gives the same result as calling object’s virtual Equals method—but more quickly. Most basic .NET types imple- ment IEquatable. You can use IEquatable as a constraint in a generic type: class Test where T : IEquatable { public bool IsEqual (T a, T b) { return a.Equals (b); // No boxing with generic T } } 258 | Chapter 6: Framework Fundamentals If we remove the generic constraint, the class would still compile, but a.Equals(b) would instead bind to the slower object.Equals (slower assuming T was a value type). When Equals and == are not equal We said earlier that it’s sometimes useful for == and Equals to apply different defi- nitions of equality. For example: double x = double.NaN; Console.WriteLine (x == x); // False Console.WriteLine (x.Equals (x)); // True The double type’s == operator enforces that one NaN can never equal anything else —even another NaN. This is most natural from a mathematical perspective, and it reflects the underlying CPU behavior. The Equals method, however, is obliged to apply reflexive equality; in other words: x.Equals (x) must always return true. Collections and dictionaries rely on Equals behaving this way; otherwise, they could not find an item they previously stored. Having Equals and == apply different definitions of equality is actually quite rare with value types. A more common scenario is with reference types, and happens when the author customizes Equals so that it performs value equality while leaving == to perform (default) referential equality. The StringBuilder class does exactly this: var sb1 = new StringBuilder ("foo"); var sb2 = new StringBuilder ("foo"); Console.WriteLine (sb1 == sb2); // False (referential equality) Console.WriteLine (sb1.Equals (sb2)); // True (value equality) Let’s now look at how to customize equality. Equality and Custom Types Recall default equality comparison behavior: • Value types use value equality. • Reference types use referential equality. Further: • A struct’s Equals method applies structural value equality by default (i.e., it compares each field in the struct). Sometimes it makes sense to override this behavior when writing a type. There are two cases for doing so: • To change the meaning of equality • To speed up equality comparisons for structs Equality Comparison | 259 Framework Fundamentals Changing the meaning of equality Changing the meaning of equality makes sense when the default behavior of == and Equals is unnatural for your type and is not what a consumer would expect. An ex- ample is DateTimeOffset, a struct with two private fields: a UTC DateTime and a numeric integer offset. If you were writing this type, you’d probably want to ensure that equality comparisons considered only the UTC DateTime field and not the offset field. Another example is numeric types that support NaN values such as float and double. If you were implementing such types yourself, you’d want to ensure that NaN- comparison logic was supported in equality comparisons. With classes, it’s sometimes more natural to offer value equality as the default instead of referential equality. This is often the case with small classes that hold a simple piece of data—such as System.Uri (or System.String). Speeding up equality comparisons with structs The default structural equality comparison algorithm for structs is relatively slow. Taking over this process by overriding Equals can improve performance by a factor of five. Overloading the == operator and implementing IEquatable allows unboxed equality comparisons, and this can speed things up by a factor of five again. Overriding equality semantics for reference types doesn’t ben- efit performance. The default algorithm for referential equality comparison is already very fast because it simply compares two 32- or 64-bit references. There’s actually another, rather peculiar case for customizing equality, and that’s to improve a struct’s hashing algorithm for better performance in a hashtable. This comes of the fact that equality comparison and hashing are joined at the hip. We’ll examine hashing in a moment. How to override equality semantics Here is a summary of the steps: 1. Override GetHashCode() and Equals(). 2. (Optionally) overload != and ==. 3. (Optionally) implement IEquatable. Overriding GetHashCode It might seem odd that System.Object—with its small footprint of members— defines a method with a specialized and narrow purpose. GetHashCode is a virtual method in Object that fits this description—it exists primarily for the benefit of just the following two types: System.Collections.Hashtable System.Collections.Generic.Dictionary 260 | Chapter 6: Framework Fundamentals These are hashtables—collections where each element has a key used for storage and retrieval. A hashtable applies a very specific strategy for efficiently allocating ele- ments based on their key. This requires that each key have an Int32 number, or hash code. The hash code need not be unique for each key, but should be as varied as possible for good hashtable performance. Hashtables are considered important enough that GetHashCode is defined in System.Object—so that every type can emit a hash code. We describe hashtables in detail in “Dictionaries” in Chapter 7. Both reference and value types have default implementations of GetHashCode, mean- ing you don’t need to override this method—unless you override Equals. (And if you override GetHashCode, you will almost certainly want to also override Equals.) Here are the other rules for overriding object.GetHashCode: • It must return the same value on two objects for which Equals returns true (hence, GetHashCode and Equals are overridden together). • It must not throw exceptions. • It must return the same value if called repeatedly on the same object (unless the object has changed). For maximum performance in hashtables, GetHashCode should be written so as to minimize the likelihood of two different values returning the same hashcode. This gives rise to the third reason for overriding Equals and GetHashCode on structs, which is to provide a more efficient hashing algorithm than the default. The default implementation for structs is at the discretion of the runtime and may be based on every field in the struct. In contrast, the default GetHashCode implementation for classes is based on an inter- nal object token, which is unique for each instance in the CLR’s current implementation. If an object’s hashcode changes after it’s been added as a key to a dictionary, the object will no longer be accessible in the dic- tionary. You can preempt this by basing hashcode calculations on immutable fields. A complete example illustrating how to override GetHashCode is listed shortly. Overriding Equals The axioms for object.Equals are as follows: • An object cannot equal null (unless it’s a nullable type). • Equality is reflexive (an object equals itself). Equality Comparison | 261 Framework Fundamentals • Equality is commutative (if a.Equals(b), then b.Equals(a)). • Equality is transitive (if a.Equals(b) and b.Equals(c), then a.Equals(c)). • Equality operations are repeatable and reliable (they don’t throw exceptions). Overloading == and != In addition to overriding Equals, you can optionally overload the equality and inequality operators. This is nearly always done with structs, because the consequence of not doing so is that the == and != operators will simply not work on your type. With classes, there are two ways to proceed: • Leave == and != alone—so that they apply referential equality. • Overload == and != in line with Equals. The first approach is most common with custom types—especially mutable types. It ensures that your type follows the expectation that == and != should exhibit ref- erential equality with reference types and this avoids confusing consumers. We saw an example earlier: var sb1 = new StringBuilder ("foo"); var sb2 = new StringBuilder ("foo"); Console.WriteLine (sb1 == sb2); // False (referential equality) Console.WriteLine (sb1.Equals (sb2)); // True (value equality) The second approach makes sense with types for which a consumer would never want referential equality. These are typically immutable—such as the string and System.Uri classes—and are sometimes good candidates for structs. Although it’s possible to overload != such that it means some- thing other than !(==), this is almost never done in practice, except in cases such as comparing float.NaN. Implementing IEquatable For completeness, it’s also good to implement IEquatable when overriding Equals. Its results should always match those of the overridden object’s Equals method. Implementing IEquatable comes at no programming cost if you struc- ture your Equals method implementation, as in the following example. An example: The Area struct Imagine we need a struct to represent an area whose width and height are inter- changeable. In other words, 5 × 10 is equal to 10 × 5. (Such a type would be suitable in an algorithm that arranges rectangular shapes.) Here’s the complete code: public struct Area : IEquatable { 262 | Chapter 6: Framework Fundamentals public readonly int Measure1; public readonly int Measure2; public Area (int m1, int m2) { Measure1 = Math.Min (m1, m2); Measure2 = Math.Max (m1, m2); } public override bool Equals (object other) { if (!(other is Area)) return false; return Equals ((Area) other); // Calls method below } public bool Equals (Area other) // Implements IEquatable { return Measure1 == other.Measure1 && Measure2 == other.Measure2; } public override int GetHashCode() { return Measure2 * 31 + Measure1; // 31 = some prime number } public static bool operator == (Area a1, Area a2) { return a1.Equals (a2); } public static bool operator != (Area a1, Area a2) { return !a1.Equals (a2); } } Here’s another way to implement the Equals method, leveraging nullable types: Area? otherArea = other as Area?; return otherArea.HasValue && Equals (otherArea.Value); In implementing GetHashCode, we’ve helped to improve the likelihood of uniqueness by multiplying the larger measure by some prime number (ignoring any overflow) before adding the two together. When there are more than two fields, the following pattern, suggested by Josh Bloch, gives good results while being performant: int hash = 17; // 17 = some prime number hash = hash * 31 + field1.GetHashCode(); // 31 = another prime number hash = hash * 31 + field2.GetHashCode(); hash = hash * 31 + field3.GetHashCode(); ... return hash; Equality Comparison | 263 Framework Fundamentals (See http://albahari.com/hashprimes for a link to a discussion on primes and hashcodes.) Here’s a demo of the Area struct: Area a1 = new Area (5, 10); Area a2 = new Area (10, 5); Console.WriteLine (a1.Equals (a2)); // True Console.WriteLine (a1 == a2); // True Pluggable equality comparers If you want a type to take on different equality semantics just for a particular sce- nario, you can use a pluggable IEqualityComparer. This is particularly useful in con- junction with the standard collection classes, and we describe it in the following chapter, in “Plugging in Equality and Order” on page 312. Order Comparison As well as defining standard protocols for equality, C# and .NET define standard protocols for determining the order of one object relative to another. The basic pro- tocols are: • The IComparable interfaces (IComparable and IComparable) • The > and < operators The IComparable interfaces are used by general-purpose sorting algorithms. In the following example, the static Array.Sort method works because System.String implements the IComparable interfaces: string[] colors = { "Green", "Red", "Blue" }; Array.Sort (colors); foreach (string c in colors) Console.Write (c + " "); // Blue Green Red The < and > operators are more specialized, and they are intended mostly for numeric types. Because they are statically resolved, they can translate to highly efficient byte- code, suitable for computationally intensive algorithms. The .NET Framework also provides pluggable ordering protocols, via the ICom parer interfaces. We describe these in the final section of Chapter 7. IComparable The IComparable interfaces are defined as follows: public interface IComparable { int CompareTo (object other); } public interface IComparable { int CompareTo (T other); } The two interfaces represent the same functionality. With value types, the generic type-safe interface is faster than the nongeneric interface. In both cases, the CompareTo method works as follows: • If a comes after b, a.CompareTo(b) returns a positive number. • If a is the same as b, a.CompareTo(b) returns 0. 264 | Chapter 6: Framework Fundamentals • If a comes before b, a.CompareTo(b) returns a negative number. For example: Console.WriteLine ("Beck".CompareTo ("Anne")); // 1 Console.WriteLine ("Beck".CompareTo ("Beck")); // 0 Console.WriteLine ("Beck".CompareTo ("Chris")); // −1 Most of the base types implement both IComparable interfaces. These interfaces are also sometimes implemented when writing custom types. An example is given shortly. IComparable versus Equals Consider a type that both overrides Equals and implements the IComparable inter- faces. You’d expect that when Equals returns true, CompareTo should return 0. And you’d be right. But here’s the catch: When Equals returns false CompareTo can return what it likes (as long as it’s internally consistent)! In other words, equality can be “fussier” than comparison, but not vice versa (violate this and sorting algorithms will break). So, CompareTo can say “All objects are equal” while Equals says “But some are more equal than others!” A great example of this is System.String. String’s Equals method and == operator use ordinal comparison, which compares the Unicode point values of each character. Its CompareTo method, however, uses a less fussy culture-dependent comparison. On most computers, for instance, the strings “ṻ” and “ǖ” are different according to Equals, but the same according to CompareTo. In Chapter 7, we discuss the pluggable ordering protocol, IComparer, which allows you to specify an alternative ordering algorithm when sorting or instantiating a sor- ted collection. A custom IComparer can further extend the gap between CompareTo and Equals—a case-insensitive string comparer, for instance, will return 0 when comparing "A" and "a". The reverse rule still applies, however: CompareTo can never be fussier than Equals. When implementing the IComparable interfaces in a custom type, you can avoid running afoul of this rule by writing the first line of CompareTo as follows: if (Equals (other)) return 0; After that, it can return what it likes, as long as it’s consistent! < and > Some types define < and > operators. For instance: bool after2010 = DateTime.Now > new DateTime (2010, 1, 1); Order Comparison | 265 Framework Fundamentals You can expect the < and > operators, when implemented, to be functionally con- sistent with the IComparable interfaces. This is standard practice across the .NET Framework. It’s also standard practice to implement the IComparable interfaces whenever < and > are overloaded, although the reverse is not true. In fact, most .NET types that implement IComparable do not overload < and >. This differs from the situation with equality, where it’s normal to overload == when overriding Equals. Typically, > and < are overloaded only when: • A type has a strong intrinsic concept of “greater than” and “less than” (versus IComparable’s broader concepts of “comes before” and “comes after”). • There is only one way or context in which to perform the comparison. • The result is invariant across cultures. System.String doesn’t satisfy the last point: the results of string comparisons can vary according to language. Hence, string doesn’t support the > and < operators: bool error = "Beck" > "Anne"; // Compile-time error Implementing the IComparable Interfaces In the following struct, representing a musical note, we implement the ICompara ble interfaces, as well as overloading the < and > operators. For completeness, we also override Equals/GetHashCode and overload == and !=. public struct Note : IComparable, IEquatable, IComparable { int _semitonesFromA; public int SemitonesFromA { get { return _semitonesFromA; } } public Note (int semitonesFromA) { _semitonesFromA = semitonesFromA; } public int CompareTo (Note other) // Generic IComparable { if (Equals (other)) return 0; // Fail-safe check return _semitonesFromA.CompareTo (other._semitonesFromA); } int IComparable.CompareTo (object other) // Nongeneric IComparable { if (!(other is Note)) throw new InvalidOperationException ("CompareTo: Not a note"); return CompareTo ((Note) other); } public static bool operator < (Note n1, Note n2) { return n1.CompareTo (n2) < 0; } 266 | Chapter 6: Framework Fundamentals public static bool operator > (Note n1, Note n2) { return n1.CompareTo (n2) > 0; } public bool Equals (Note other) // for IEquatable { return _semitonesFromA == other._semitonesFromA; } public override bool Equals (object other) { if (!(other is Note)) return false; return Equals ((Note) other); } public override int GetHashCode() { return _semitonesFromA.GetHashCode(); } public static bool operator == (Note n1, Note n2) { return n1.Equals (n2); } public static bool operator != (Note n1, Note n2) { return !(n1 == n2); } } Utility Classes Console The static Console class handles standard input/output for console-based applica- tions. In a command-line (Console) application, the input comes from the keyboard via Read, ReadKey, and ReadLine, and the output goes to the text window via Write and WriteLine. You can control the window’s position and dimensions with the properties WindowLeft, WindowTop, WindowHeight, and WindowWidth. You can also change the BackgroundColor and ForegroundColor properties and manipulate the cursor with the CursorLeft, CursorTop, and CursorSize properties: Console.WindowWidth = Console.LargestWindowWidth; Console.ForegroundColor = ConsoleColor.Green; Console.Write ("test... 50%"); Console.CursorLeft -= 3; Console.Write ("90%"); // test... 90% The Write and WriteLine methods are overloaded to accept a composite format string (see String.Format in “String and Text Handling” on page 201). However, neither Utility Classes | 267 Framework Fundamentals method accepts a format provider, so you’re stuck with CultureInfo.CurrentCul ture. (The workaround, of course, is to explicitly call string.Format.) The Console.Out property returns a TextWriter. Passing Console.Out to a method that expects a TextWriter is a useful way to get that method to write to the Con sole for diagnostic purposes. You can also redirect the Console’s input and output streams via the SetIn and SetOut methods: // First save existing output writer: System.IO.TextWriter oldOut = Console.Out; // Redirect the console's output to a file: using (System.IO.TextWriter w = System.IO.File.CreateText ("e:\\output.txt")) { Console.SetOut (w); Console.WriteLine ("Hello world"); } // Restore standard console output Console.SetOut (oldOut); // Open the output.txt file in Notepad: System.Diagnostics.Process.Start ("e:\\output.txt"); In Chapter 15, we describe how streams and text writers work. In a Visual Studio WPF and Windows Forms applications, the Console’s output is automatically redirected to Visual Studio’s output window (in debug mode). This can make Con sole.Write useful for diagnostic purposes; although in most cases the Debug and Trace classes in the System.Diagnostics namespace are more appropriate (see Chapter 13). Environment The static System.Environment class provides a range of useful properties: Files and folders CurrentDirectory, SystemDirectory, CommandLine Computer and operating system MachineName, ProcessorCount, OSVersion, NewLine User logon UserName, UserInteractive, UserDomainName Diagnostics TickCount, StackTrace, WorkingSet, Version You can obtain additional folders by calling GetFolderPath; we describe this in “File and Directory Operations” on page 632 in Chapter 15. 268 | Chapter 6: Framework Fundamentals You can access OS environment variables (what you see when you type “set” at the command prompt) with the following three methods: GetEnvironmentVariable, GetEnvironmentVariables, and SetEnvironmentVariable. The ExitCode property lets you set the return code, for when your program is called from a command or batch file, and the FailFast method terminates a program immediately, without performing cleanup. The Metro profile exposes the Environment class but offers just a limited number of members (ProcessorCount, NewLine, and FailFast). Process The Process class in System.Diagnostics allows you to launch a new process. For security reasons, the Process class is not available in the Metro profile, and you cannot start arbitrary processes. Instead, you must use the Windows.System.Launcher class to “launch” a URI or file to which you have access, e.g.: Launcher.LaunchUriAsync (new Uri ("http://albahari.com")); var file = await KnownFolders.DocumentsLibrary .GetFileAsync ("foo.txt"); Launcher.LaunchFileAsync (file); This opens the URI or file, using whatever program is associated with the URI scheme or file extension. Your program must be in the foreground for this to work. The static Process.Start method has a number of overloads; the simplest accepts a simple filename with optional arguments: Process.Start ("notepad.exe"); Process.Start ("notepad.exe", "e:\\file.txt"); You can also specify just a filename, and the registered program for its extension will be launched: Process.Start ("e:\\file.txt"); The most flexible overload accepts a ProcessStartInfo instance. With this, you can capture and redirect the launched process’s input, output, and error output (if you set UseShellExecute to false). The following captures the output of calling ipconfig: ProcessStartInfo psi = new ProcessStartInfo { FileName = "cmd.exe", Arguments = "/c ipconfig /all", RedirectStandardOutput = true, UseShellExecute = false }; Process p = Process.Start (psi); string result = p.StandardOutput.ReadToEnd(); Console.WriteLine (result); Utility Classes | 269 Framework Fundamentals You can do the same to invoke the csc compiler, if you set Filename to the following: psi.FileName = System.IO.Path.Combine ( System.Runtime.InteropServices.RuntimeEnvironment.GetRuntimeDirectory(), "csc.exe"); If you don’t redirect output, Process.Start executes the program in parallel to the caller. If you want to wait for the new process to complete, you can call WaitFor Exit on the Process object, with an optional timeout. The Process class also allows you to query and interact with other processes running on the computer (see Chapter 13). 270 | Chapter 6: Framework Fundamentals 7Collections The .NET Framework provides a standard set of types for storing and managing collections of objects. These include resizable lists, linked lists, sorted and unsorted dictionaries as well as arrays. Of these, only arrays form part of the C# language; the remaining collections are just classes you instantiate like any other. The types in the Framework for collections can be divided into the following categories: • Interfaces that define standard collection protocols • Ready-to-use collection classes (lists, dictionaries, etc.) • Base classes for writing application-specific collections This chapter covers each of these categories, with an additional section on the types used in determining element equality and order. The collection namespaces are as follows: Namespace Contains System.Collections Nongeneric collection classes and interfaces System.Collections.Specialized Strongly typed nongeneric collection classes System.Collections.Generic Generic collection classes and interfaces System.Collections.ObjectModel Proxies and bases for custom collections System.Collections.Concurrent Thread-safe collections (see Chapter 23) Enumeration In computing, there are many different kinds of collections ranging from simple data structures, such as arrays or linked lists, to more complex ones, such as red/black trees and hashtables. Although the internal implementation and external charac- teristics of these data structures vary widely, the ability to traverse the contents of the collection is an almost universal need. The Framework supports this need via a 271 pair of interfaces (IEnumerable, IEnumerator, and their generic counterparts) that allow different data structures to expose a common traversal API. These are part of a larger set of collection interfaces illustrated in Figure 7-1. Figure 7-1. Collection interfaces IEnumerable and IEnumerator The IEnumerator interface defines the basic low-level protocol by which elements in a collection are traversed—or enumerated—in a forward-only manner. Its declara- tion is as follows: public interface IEnumerator { bool MoveNext(); object Current { get; } void Reset(); } MoveNext advances the current element or “cursor” to the next position, returning false if there are no more elements in the collection. Current returns the element at the current position (usually cast from object to a more specific type). MoveNext must be called before retrieving the first element—this is to allow for an empty collection. The Reset method, if implemented, moves back to the start, allowing the collection to be enumerated again. Reset exists mainly for COM interop: calling it directly is generally avoided because it’s not universally supported (and is unnecessary in that it’s usually just as easy to instantiate a new enumerator.) Collections do not usually implement enumerators; instead, they provide enumera- tors, via the interface IEnumerable: public interface IEnumerable { IEnumerator GetEnumerator(); } 272 | Chapter 7: Collections By defining a single method retuning an enumerator, IEnumerable provides flexibility in that the iteration logic can be farmed off to another class. Moreover, it means that several consumers can enumerate the collection at once without interfering with each other. IEnumerable can be thought of as “IEnumeratorProvider,” and it is the most basic interface that collection classes implement. The following example illustrates low-level use of IEnumerable and IEnumerator: string s = "Hello"; // Because string implements IEnumerable, we can call GetEnumerator(): IEnumerator rator = s.GetEnumerator(); while (rator.MoveNext()) { char c = (char) rator.Current; Console.Write (c + "."); } // Output: H.e.l.l.o. However, it’s rare to call methods on enumerators directly in this manner, because C# provides a syntactic shortcut: the foreach statement. Here’s the same example rewritten using foreach: string s = "Hello"; // The String class implements IEnumerable foreach (char c in s) Console.Write (c + "."); IEnumerable and IEnumerator IEnumerator and IEnumerable are nearly always implemented in conjunction with their extended generic versions: public interface IEnumerator : IEnumerator, IDisposable { T Current { get; } } public interface IEnumerable : IEnumerable { IEnumerator GetEnumerator(); } By defining a typed version of Current and GetEnumerator, these interfaces strengthen static type safety, avoid the overhead of boxing with value-type elements, and are more convenient to the consumer. Arrays automatically implement IEnumera ble (where T is the member type of the array). Thanks to the improved static type safety, calling the following method with an array of characters will generate a compile-time error: void Test (IEnumerable numbers) { ... } Enumeration | 273 Collections It’s a standard practice for collection classes to publicly expose IEnumerable, while “hiding” the nongeneric IEnumerable through explicit interface implementa- tion. This is so that if you directly call GetEnumerator(), you get back the type-safe generic IEnumerator. There are times, though, when this rule is broken for reasons of backward compatibility (generics did not exist prior to C# 2.0). A good example is arrays—these must return the nongeneric (the nice way of putting it is “classic”) IEnumerator to avoid breaking earlier code. In order to get a generic IEnumera tor, you must cast to expose the explicit interface: int[] data = { 1, 2, 3 }; var rator = ((IEnumerable )data).GetEnumerator(); Fortunately, you rarely need to write this sort of code, thanks to the foreach statement. IEnumerable and IDisposable IEnumerator inherits from IDisposable. This allows enumerators to hold refer- ences to resources such as database connections—and ensure that those resources are released when enumeration is complete (or abandoned partway through). The foreach statement recognizes this detail and translates this: foreach (var element in somethingEnumerable) { ... } into the logical equivalent of this: using (var rator = somethingEnumerable.GetEnumerator()) while (rator.MoveNext()) { var element = rator.Current; ... } The using block ensures disposal—more on IDisposable in Chapter 12. When to Use the Nongeneric Interfaces Given the extra type safety of the generic collection interfaces such as IEnumera ble, the question arises: do you ever need to use the nongeneric IEnumerable (or ICollection or IList)? In the case of IEnumerable, you must implement this interface in conjunction with IEnumerable—because the latter derives from the former. However, it’s very rare that you actually implement these interfaces from scratch: in nearly all cases, you can take the higher-level approach of using iterator methods, Collection, and LINQ. So, what about as a consumer? In nearly all cases, you can manage entirely with the generic interfaces. The nongeneric interfaces are still occasionally useful, though, in their ability to provide type unification for collections across all element types. The following method, for instance, counts elements in any collection recursively: public static int Count (IEnumerable e) { 274 | Chapter 7: Collections int count = 0; foreach (object element in e) { var subCollection = element as IEnumerable; if (subCollection != null) count += Count (subCollection); else count++; } return count; } Because C# offers covariance with generic interfaces, it might seem valid to have this method instead accept IEnumerable. This, however, would fail with value-type elements and with legacy collections that don’t implement IEnumera ble— an example is ControlCollection in Windows Forms. (On a slight tangent, you might have noticed a potential bug in our example: cyclic references will cause infinite recursion and crash the method. We could fix this most easily with the use of a HashSet (see “HashSet and Sorted- Set” on page 297.) Implementing the Enumeration Interfaces You might want to implement IEnumerable or IEnumerable for one or more of the following reasons: • To support the foreach statement • To interoperate with anything expecting a standard collection • To meet the requirements of a more sophisticated collection interface • To support collection initializers To implement IEnumerable/IEnumerable, you must provide an enumerator. You can do this in one of three ways: • If the class is “wrapping” another collection, by returning the wrapped collec- tion’s enumerator • Via an iterator using yield return • By instantiating your own IEnumerator/IEnumerator implementation You can also subclass an existing collection: Collection is designed just for this purpose (see “Customizable Collections and Proxies” on page 306). Yet another approach is to use the LINQ query operators that we’ll cover in the next chapter. Returning another collection’s enumerator is just a matter of calling GetEnumera tor on the inner collection. However, this is viable only in the simplest scenarios, where the items in the inner collection are exactly what are required. A more flexible approach is to write an iterator, using C#’s yield return statement. An iterator is a Enumeration | 275 Collections C# language feature that assists in writing collections, in the same way the foreach statement assists in consuming collections. An iterator automatically han- dles the implementation of IEnumerable and IEnumerator—or their generic versions. Here’s a simple example: public class MyCollection : IEnumerable { int[] data = { 1, 2, 3 }; public IEnumerator GetEnumerator() { foreach (int i in data) yield return i; } } Notice the “black magic”: GetEnumerator doesn’t appear to return an enumerator at all! Upon parsing the yield return statement, the compiler writes a hidden nested enumerator class behind the scenes, and then refactors GetEnumerator to instantiate and return that class. Iterators are powerful and simple (and are used extensively in the implementation of LINQ-to-Object’s standard query operators). Keeping with this approach, we can also implement the generic interface IEnumerable: public class MyGenCollection : IEnumerable { int[] data = { 1, 2, 3 }; public IEnumerator GetEnumerator() { foreach (int i in data) yield return i; } IEnumerator IEnumerable.GetEnumerator() // Explicit implementation { // keeps it hidden. return GetEnumerator(); } } Because IEnumerable inherits from IEnumerable, we must implement both the generic and the nongeneric versions of GetEnumerator. In accordance with standard practice, we’ve implemented the nongeneric version explicitly. It can simply call the generic GetEnumerator because IEnumerator inherits from IEnumerator. The class we’ve just written would be suitable as a basis from which to write a more sophisticated collection. However, if we need nothing above a simple IEnumera ble implementation, the yield return statement allows for an easier variation. Rather than writing a class, you can move the iteration logic into a method returning a generic IEnumerable and let the compiler take care of the rest. Here’s an example: public class Test { 276 | Chapter 7: Collections public static IEnumerable GetSomeIntegers() { yield return 1; yield return 2; yield return 3; } } Here’s our method in use: foreach (int i in Test.GetSomeIntegers()) Console.WriteLine (i); // Output 1 2 3 The final approach in writing GetEnumerator is to write a class that implements IEnumerator directly. This is exactly what the compiler does behind the scenes, in resolving iterators. (Fortunately, it’s rare that you’ll need to go this far yourself.) The following example defines a collection that’s hardcoded to contain the integers 1, 2, and 3: public class MyIntList : IEnumerable { int[] data = { 1, 2, 3 }; public IEnumerator GetEnumerator() { return new Enumerator (this); } class Enumerator : IEnumerator // Define an inner class { // for the enumerator. MyIntList collection; int currentIndex = −1; public Enumerator (MyIntList collection) { this.collection = collection; } public object Current { get { if (currentIndex == −1) throw new InvalidOperationException ("Enumeration not started!"); if (currentIndex == collection.data.Length) throw new InvalidOperationException ("Past end of list!"); return collection.data [currentIndex]; } } public bool MoveNext() Enumeration | 277 Collections { if (currentIndex >= collection.data.Length - 1) return false; return ++currentIndex < collection.data.Length; } public void Reset() { currentIndex = −1; } } } Implementing Reset is optional—you can instead throw a Not SupportedException. Note that the first call to MoveNext should move to the first (and not the second) item in the list. To get on par with an iterator in functionality, we must also implement IEnumera tor. Here’s an example with bounds checking omitted for brevity: class MyIntList : IEnumerable { int[] data = { 1, 2, 3 }; // The generic enumerator is compatible with both IEnumerable and // IEnumerable. We implement the nongeneric GetEnumerator method // explicitly to avoid a naming conflict. public IEnumerator GetEnumerator() { return new Enumerator(this); } IEnumerator IEnumerable.GetEnumerator() { return new Enumerator(this); } class Enumerator : IEnumerator { int currentIndex = −1; MyIntList collection; public Enumerator (MyIntList collection) { this.collection = collection; } public int Current { get { return collection.data [currentIndex]; } } object IEnumerator.Current { get { return Current; } } public bool MoveNext() { return ++currentIndex < collection.data.Length; } public void Reset() { currentIndex = −1; } // Given we don't need a Dispose method, it's good practice to // implement it explicitly, so it's hidden from the public interface. void IDisposable.Dispose() {} 278 | Chapter 7: Collections } } The example with generics is faster because IEnumerator.Current doesn’t require casting from int to object, and so avoids the overhead of boxing. The ICollection and IList Interfaces Although the enumeration interfaces provide a protocol for forward-only iteration over a collection, they don’t provide a mechanism to determine the size of the collection, access a member by index, search, or modify the collection. For such functionality, the .NET Framework defines the ICollection, IList, and IDiction ary interfaces. Each comes in both generic and nongeneric versions; however, the nongeneric versions exist mostly for legacy support. The inheritance hierarchy for these interfaces was shown in Figure 7-1. The easiest way to summarize them is as follows: IEnumerable (and IEnumerable) Provides minimum functionality (enumeration only) ICollection (and ICollection) Provides medium functionality (e.g., the Count property) IList /IDictionary and their nongeneric versions Provide maximum functionality (including “random” access by index/key) It’s rare that you’ll need to implement any of these interfaces. In nearly all cases when you need to write a collection class, you can instead subclass Collection (see “Customizable Collec- tions and Proxies” on page 306). LINQ provides yet another option that covers many scenarios. The generic and nongeneric versions differ in ways over and above what you might expect, particularly in the case of ICollection. The reasons for this are mostly his- torical: because generics came later, the generic interfaces were developed with the benefit of hindsight, leading to a different (and better) choice of members. For this reason, ICollection does not extend ICollection, IList does not extend IList, and IDictionary does not extend IDictionary. Of course, a collection class itself is free to implement both versions of an interface if beneficial (which, often, it is). Another, subtler reason for IList not extending IList is that casting to IList would then return an interface with both Add(T) and Add(object) members. This would effectively defeat static type safety, because you could call Add with an object of any type. The ICollection and IList Interfaces | 279 Collections This section covers ICollection, IList, and their nongeneric versions; “Dic- tionaries” covers the dictionary interfaces. There is no consistent rationale in the way the words collection and list are applied throughout the .NET Framework. For instance, since IList is a more functional version of ICollec tion, you might expect the class List to be correspond- ingly more functional than the class Collection. This is not the case. It’s best to consider the terms collection and list as broadly synonymous, except when a specific type is involved. ICollection and ICollection ICollection is the standard interface for countable collections of objects. It pro- vides the ability to determine the size of a collection (Count), determine whether an item exists in the collection (Contains), copy the collection into an array (ToArray), and determine whether the collection is read-only (IsReadOnly). For writable col- lections, you can also Add, Remove, and Clear items from the collection. And since it extends IEnumerable, it can also be traversed via the foreach statement. public interface ICollection : IEnumerable, IEnumerable { int Count { get; } bool Contains (T item); void CopyTo (T[] array, int arrayIndex); bool IsReadOnly { get; } void Add(T item); bool Remove (T item); void Clear(); } The nongeneric ICollection is similar in providing a countable collection, but doesn’t provide functionality for altering the list or checking for element membership: public interface ICollection : IEnumerable { int Count { get; } bool IsSynchronized { get; } object SyncRoot { get; } void CopyTo (Array array, int index); } The nongeneric interface also defines properties to assist with synchronization (Chapter 14)—these were dumped in the generic version because thread safety is no longer considered intrinsic to the collection. Both interfaces are fairly straightforward to implement. If implementing a read-only ICollection, the Add, Remove, and Clear methods should throw a NotSupportedEx ception. 280 | Chapter 7: Collections These interfaces are usually implemented in conjunction with either the IList or the IDictionary interface. IList and IList IList is the standard interface for collections indexable by position. In addition to the functionality inherited from ICollection and IEnumerable, it provides the ability to read or write an element by position (via an indexer) and insert/remove by position: public interface IList : ICollection, IEnumerable, IEnumerable { T this [int index] { get; set; } int IndexOf (T item); void Insert (int index, T item); void RemoveAt (int index); } The IndexOf methods perform a linear search on the list, returning −1 if the specified item is not found. The nongeneric version of IList has more members because it inherits less from ICollection: public interface IList : ICollection, IEnumerable { object this [int index] { get; set } bool IsFixedSize { get; } bool IsReadOnly { get; } int Add (object value); void Clear(); bool Contains (object value); int IndexOf (object value); void Insert (int index, object value); void Remove (object value); void RemoveAt (int index); } The Add method on the nongeneric IList interface returns an integer—this is the index of the newly added item. In contrast, the Add method on ICollection has a void return type. The general-purpose List class is the quintessential implementation of both IList and IList. C# arrays also implement both the generic and nongeneric ILists (although the methods that add or remove elements are hidden via explicit interface implementation and throw a NotSupportedException if called). An ArgumentException is thrown if you try to access a multidi- mensional array via IList’s indexer. This is a trap when writing methods such as the following: public object FirstOrNull (IList list) { if (list == null || list.Count == 0) return null; The ICollection and IList Interfaces | 281 Collections return list[0]; } This might appear bulletproof, but it will throw an exception if called with a multidimensional array. You can test for a multi- dimensional array at runtime with this expression (more on this in Chapter 19): list.GetType().IsArray && list.GetType().GetArrayRank()>1 IReadOnlyList In order to interoperate with read-only Windows Runtime collections, Framework 4.5 introduces a new collection interface called IReadOnlyList. This interface is useful in and of itself, and can be considered a cut-down version of IList, exposing just the members required for read-only operations on lists: public interface IReadOnlyList : IEnumerable, IEnumerable { int Count { get; } T this[int index] { get; } } Because its type parameter is used only in output positions, it’s marked as covariant. This allows a list of cats, for instance, to be treated as a read-only list of animals. In contrast, T is not marked as covariant with IList, because T is used in both input and output positions. IReadOnlyList represents a read-only view of a list. It doesn’t necessarily imply that the underlying implementation is read- only. It would be logical for IList to derive from IReadOnlyList. However, Micro- soft was unable to make this change because doing so would require moving mem- bers from IList to IReadOnlyList, which would introduce a breaking change into CLR 4.5 (consumers would need to re-compile their programs to avoid runtime errors). Instead, implementers of IList need to manually add IReadOnlyList to their list of implemented interfaces. IReadOnlyList maps to the Windows Runtime type IVectorView. The Array Class The Array class is the implicit base class for all single and multidimensional arrays, and it is one of the most fundamental types implementing the standard collection interfaces. The Array class provides type unification, so a common set of methods is available to all arrays, regardless of their declaration or underlying element type. Since arrays are so fundamental, C# provides explicit syntax for their declaration and initialization, described in Chapters 2 and 3. When an array is declared using 282 | Chapter 7: Collections C#’s syntax, the CLR implicitly subtypes the Array class—synthesizing a pseudo- type appropriate to the array’s dimensions and element types. This pseudotype im- plements the typed generic collection interfaces, such as IList. The CLR also treats array types specially upon construction, assigning them a con- tiguous space in memory. This makes indexing into arrays highly efficient, but pre- vents them from being resized later on. Array implements the collection interfaces up to IList in both their generic and nongeneric forms. IList itself is implemented explicitly, though, to keep Array’s public interface clean of methods such as Add or Remove, which throw an exception on fixed-length collections such as arrays. The Array class does actually offer a static Resize method, although this works by creating a new array and then copying over each element. As well as being inefficient, references to the array else- where in the program will still point to the original version. A better solution for resizable collections is to use the List class (described in the following section). An array can contain value type or reference type elements. Value type elements are stored in place in the array, so an array of three long integers (each 8 bytes) will occupy 24 bytes of contiguous memory. A reference type element, however, occupies only as much space in the array as a reference (4 bytes in a 32-bit environment or 8 bytes in a 64-bit environment). Figure 7-2 illustrates the effect, in memory, of the following program. StringBuilder[] builders = new StringBuilder [5]; builders [0] = new StringBuilder ("builder1"); builders [1] = new StringBuilder ("builder2"); builders [2] = new StringBuilder ("builder3"); long[] numbers = new long [3]; numbers [0] = 12345; numbers [1] = 54321; Because Array is a class, arrays are always (themselves) reference types—regardless of the array’s element type. This means that the statement arrayB = arrayA results in two variables that reference the same array. Similarly, two distinct arrays will always fail an equality test—unless you use a custom equality comparer. Framework 4.0 introduced one for the purpose of comparing elements in arrays or tuples which you can access via the StructuralComparisons type: object[] a1 = { "string", 123, true }; object[] a2 = { "string", 123, true }; Console.WriteLine (a1 == a2); // False Console.WriteLine (a1.Equals (a2)); // False IStructuralEquatable se1 = a1; Console.WriteLine (se1.Equals (a2, StructuralComparisons.StructuralEqualityComparer)); // True Arrays can be duplicated with the Clone method: arrayB = arrayA.Clone(). How- ever, this results in a shallow clone, meaning that only the memory represented by the array itself is copied. If the array contains value type objects, the values The Array Class | 283 Collections themselves are copied; if the array contains reference type objects, just the references are copied (resulting in two arrays whose members reference the same objects). Figure 7-3 demonstrates the effect of adding the following code to our example: StringBuilder[] builders2 = builders; StringBuilder[] shallowClone = (StringBuilder[]) builders.Clone(); Figure 7-3. Shallow-cloning an array To create a deep copy—where reference type subobjects are duplicated—you must loop through the array and clone each element manually. The same rules apply to other .NET collection types. Figure 7-2. Arrays in memory 284 | Chapter 7: Collections Although Array is designed primarily for use with 32-bit indexers, it also has limited support for 64-bit indexers (allowing an array to theoretically address up to 264 elements) via several methods that accept both Int32 and Int64 parameters. These overloads are useless in practice, because the CLR does not permit any object— including arrays—to exceed 2GB in size (whether running on a 32- or 64-bit environment). Many of the methods on the Array class that you expect to be instance methods are in fact static methods. This is an odd design decision, and means you should check for both static and instance methods when looking for a method on Array. Construction and Indexing The easiest way to create and index arrays is through C#’s language constructs: int[] myArray = { 1, 2, 3 }; int first = myArray [0]; int last = myArray [myArray.Length - 1]; Alternatively, you can instantiate an array dynamically by calling Array.CreateIn stance. This allows you to specify element type and rank (number of dimensions) at runtime—as well as allowing nonzero-based arrays through specifying a lower bound. Nonzero-based arrays are not CLS (Common Language Specification)- compliant. The static GetValue and SetValue methods let you access elements in a dynamically created array (they also work on ordinary arrays): // Create a string array 2 elements in length: Array a = Array.CreateInstance (typeof(string), 2); a.SetValue ("hi", 0); // → a[0] = "hi"; a.SetValue ("there", 1); // → a[1] = "there"; string s = (string) a.GetValue (0); // → s = a[0]; // We can also cast to a C# array as follows: string[] cSharpArray = (string[]) a; string s2 = cSharpArray [0]; Zero-indexed arrays created dynamically can be cast to a C# array of a matching or compatible type (compatible by standard array-variance rules). For example, if Apple subclasses Fruit, Apple[] can be cast to Fruit[]. This leads to the issue of why object[] was not used as the unifying array type rather the Array class. The answer is that object[] is incompatible with both multidimensional and value-type arrays (and non-zero-based arrays). An int[] array cannot be cast to object[]. Hence, we require the Array class for full type unification. GetValue and SetValue also work on compiler-created arrays, and they are useful when writing methods that can deal with an array of any type and rank. For multi- dimensional arrays, they accept an array of indexers: public object GetValue (params int[] indices) public void SetValue (object value, params int[] indices) The Array Class | 285 Collections The following method prints the first element of any array, regardless of rank: void WriteFirstValue (Array a) { Console.Write (a.Rank + "-dimensional; "); // The indexers array will automatically initialize to all zeros, so // passing it into GetValue or SetValue will get/set the zero-based // (i.e., first) element in the array. int[] indexers = new int[a.Rank]; Console.WriteLine ("First value is " + a.GetValue (indexers)); } void Demo() { int[] oneD = { 1, 2, 3 }; int[,] twoD = { {5,6}, {8,9} }; WriteFirstValue (oneD); // 1-dimensional; first value is 1 WriteFirstValue (twoD); // 2-dimensional; first value is 5 } For working with arrays of unknown type but known rank, generics provide an easier and more efficient solution: void WriteFirstValue (T[] array) { Console.WriteLine (array[0]); } SetValue throws an exception if the element is of an incompatible type for the array. When an array is instantiated, whether via language syntax or Array.CreateIn stance, its elements are automatically initialized. For arrays with reference type elements, this means writing nulls; for arrays with value type elements, this means calling the value type’s default constructor (effectively “zeroing” the members). The Array class also provides this functionality on demand via the Clear method: public static void Clear (Array array, int index, int length); This method doesn’t affect the size of the array. This is in contrast to the usual use of Clear (such as in ICollection.Clear) where the collection is reduced to zero elements. Enumeration Arrays are easily enumerated with a foreach statement: int[] myArray = { 1, 2, 3}; foreach (int val in myArray) Console.WriteLine (val); 286 | Chapter 7: Collections You can also enumerate using the static Array.ForEach method, defined as follows: public static void ForEach (T[] array, Action action); This uses an Action delegate, with this signature: public delegate void Action (T obj); Here’s the first example rewritten with Array.ForEach: Array.ForEach (new[] { 1, 2, 3 }, Console.WriteLine); Length and Rank Array provides the following methods and properties for querying length and rank: public int GetLength (int dimension); public long GetLongLength (int dimension); public int Length { get; } public long LongLength { get; } public int GetLowerBound (int dimension); public int GetUpperBound (int dimension); public int Rank { get; } // Returns number of dimensions in array GetLength and GetLongLength return the length for a given dimension (0 for a single- dimensional array), and Length and LongLength return the total number of elements in the array—all dimensions included. GetLowerBound and GetUpperBound are useful with nonzero indexed arrays. GetUpper Bound returns the same result as adding GetLowerBound to GetLength for any given dimension. Searching The Array class offers a range of methods for finding elements within a one- dimensional array: BinarySearch methods For rapidly searching a sorted array for a particular item IndexOf / LastIndex methods For searching unsorted arrays for a particular item Find / FindLast / FindIndex / FindLastIndex / FindAll / Exists / TrueForAll For searching unsorted arrays for item(s) that satisfy a given Predicate None of the array searching methods throws an exception if the specified value is not found. Instead, if an item is not found, methods returning an integer return −1 (assuming a zero-indexed array), and methods returning a generic type return the type’s default value (e.g., 0 for an int, or null for a string ). The binary search methods are fast, but they work only on sorted arrays and require that the elements be compared for order, rather than simply equality. To this effect, the binary search methods can accept an IComparer or IComparer object to The Array Class | 287 Collections arbitrate on ordering decisions (see the section “Plugging in Equality and Or- der” on page 312 later in this chapter). This must be consistent with any comparer used in originally sorting the array. If no comparer is provided, the type’s default ordering algorithm will be applied, based on its implementation of IComparable / IComparable. The IndexOf and LastIndexOf methods perform a simple enumeration over the array, returning the position of the first (or last) element that matches the given value. The predicate-based searching methods allow a method delegate or lambda expres- sion to arbitrate on whether a given element is a “match.” A predicate is simply a delegate accepting an object and returning true or false: public delegate bool Predicate (T object); In the following example, we search an array of strings for a name containing the letter “a”: static void Main() { string[] names = { "Rodney", "Jack", "Jill" }; string match = Array.Find (names, ContainsA); Console.WriteLine (match); // Jack } static bool ContainsA (string name) { return name.Contains ("a"); } Here’s the same code shortened with an anonymous method: string[] names = { "Rodney", "Jack", "Jill" }; string match = Array.Find (names, delegate (string name) { return name.Contains ("a"); } ); A lambda expression shortens it further: string[] names = { "Rodney", "Jack", "Jill" }; string match = Array.Find (names, n => n.Contains ("a")); // Jack FindAll returns an array of all items satisfying the predicate. In fact, it’s equivalent to Enumerable.Where in the System.Linq namespace, except that FindAll returns an array of matching items rather than an IEnumerable of the same. Exists returns true if any array member satisfies the given predicate, and is equiv- alent to Any in System.Linq.Enumerable. TrueForAll returns true if all items satisfy the predicate, and is equivalent to All in System.Linq.Enumerable. Sorting Array has the following built-in sorting methods: // For sorting a single array: public static void Sort (T[] array); public static void Sort (Array array); // For sorting a pair of arrays: 288 | Chapter 7: Collections public static void Sort (TKey[] keys, TValue[] items); public static void Sort (Array keys, Array items); Each of these methods is additionally overloaded to also accept: int index // Starting index at which to begin sorting int length // Number of elements to sort IComparer comparer // Object making ordering decisions Comparison comparison // Delegate making ordering decisions The following illustrates the simplest use of Sort: int[] numbers = { 3, 2, 1 }; Array.Sort (numbers); // Array is now { 1, 2, 3 } The methods accepting a pair of arrays work by rearranging the items of each array in tandem, basing the ordering decisions on the first array. In the next example, both the numbers and their corresponding words are sorted into numerical order: int[] numbers = { 3, 2, 1 }; string[] words = { "three", "two", "one" }; Array.Sort (numbers, words); // numbers array is now { 1, 2, 3 } // words array is now { "one", "two", "three" } Array.Sort requires that the elements in the array implement IComparable (see the section “Order Comparison” on page 264 in Chapter 6). This means that most built- in C# types (such as integers, as in the preceding example) can be sorted. If the elements are not intrinsically comparable, or you want to override the default ordering, you must provide Sort with a custom comparison provider that reports on the relative position of two elements. There are ways to do this: • Via a helper object that implements IComparer /IComparer (see the section “Plugging in Equality and Order” on page 312 later in this chapter) • Via a Comparison delegate: public delegate int Comparison (T x, T y); The Comparison delegate follows the same semantics as IComparer.CompareTo: if x comes before y, a negative integer is returned; if x comes after y, a positive integer is returned; if x and y have the same sorting position, 0 is returned. In the following example, we sort an array of integers such that the odd numbers come first: int[] numbers = { 1, 2, 3, 4, 5 }; Array.Sort (numbers, (x, y) => x % 2 == y % 2 ? 0 : x % 2 == 1 ? −1 : 1); // numbers array is now { 3, 5, 1, 2, 4 } The Array Class | 289 Collections As an alternative to calling Sort, you can use LINQ’s OrderBy and ThenBy operators. Unlike Array.Sort, the LINQ operators don’t alter the original array, instead emitting the sorted result in a fresh IEnumerable sequence. Reversing Elements The following Array methods reverse the order of all—or a portion of—elements in an array: public static void Reverse (Array array); public static void Reverse (Array array, int index, int length); Copying Array provides four methods to perform shallow copying: Clone, CopyTo, Copy and ConstrainedCopy. The former two are instance methods; the latter two are static methods. The Clone method returns a whole new (shallow-copied) array. The CopyTo and Copy methods copy a contiguous subset of the array. Copying a multidimensional rectangular array requires you to map the multidimensional index to a linear index. For example, the middle square (position[1,1]) in a 3 × 3 array is represented with the index 4, from the calculation: 1*3 + 1. The source and destination ranges can overlap without causing a problem. ConstrainedCopy performs an atomic operation: if all of the requested elements can- not be successfully copied (due to a type error, for instance), the operation is rolled back. Array also provides a AsReadOnly method which returns a wrapper that prevents elements from being reassigned. Converting and Resizing Array.ConvertAll creates and returns a new array of element type TOutput, calling the supplied Converter delegate to copy over the elements. Converter is defined as follows: public delegate TOutput Converter (TInput input) The following converts an array of floats to an array of integers: float[] reals = { 1.3f, 1.5f, 1.8f }; int[] wholes = Array.ConvertAll (reals, r => Convert.ToInt32 (r)); // wholes array is { 1, 2, 2 } The Resize method works by creating a new array and copying over the elements, returning the new array via the reference parameter. However, any references to the original array in other objects will remain unchanged. 290 | Chapter 7: Collections The System.Linq namespace offers an additional buffet of extension methods suitable for array conversion. These meth- ods return an IEnumerable, which you can convert back to an array via Enumerable’s ToArray method. Lists, Queues, Stacks, and Sets The Framework provides a basic set of concrete collection classes that implement the interfaces described in this chapter. This section concentrates on the list- like collections (versus the dictionary-like collections covered in “Dictionar- ies” on page 299). As with the interfaces we discussed previously, you usually have a choice of generic or nongeneric versions of each type. In terms of flexibility and performance, the generic classes win, making their nongeneric counterparts redun- dant except for backward compatibility. This differs from the situation with collec- tion interfaces, where the nongeneric versions are still occasionally useful. Of the classes described in this section, the generic List class is the most commonly used. List and ArrayList The generic List and nongeneric ArrayList classes provide a dynamically sized array of objects and are among the most commonly used of the collection classes. Array List implements IList, whereas List implements both IList and IList (and the new read-only version, IReadOnlyList). Unlike with arrays, all interfaces are implemented publicly, and methods such as Add and Remove are exposed and work as you would expect. Internally, List and ArrayList work by maintaining an internal array of objects, replaced with a larger array upon reaching capacity. Appending elements is efficient (since there is usually a free slot at the end), but inserting elements can be slow (since all elements after the insertion point have to be shifted to make a free slot). As with arrays, searching is efficient if the BinarySearch method is used on a list that has been sorted, but is otherwise inefficient because each item must be individually checked. List is up to several times faster than ArrayList if T is a value type, because List avoids the overhead of boxing and un- boxing elements. List and ArrayList provide constructors that accept an existing collection of elements: these copy each element from the existing collection into the new List or ArrayList: public class List : IList, IReadOnlyList { public List (); public List (IEnumerable collection); Lists, Queues, Stacks, and Sets | 291 Collections public List (int capacity); // Add+Insert public void Add (T item); public void AddRange (IEnumerable collection); public void Insert (int index, T item); public void InsertRange (int index, IEnumerable collection); // Remove public bool Remove (T item); public void RemoveAt (int index); public void RemoveRange (int index, int count); public int RemoveAll (Predicate match); // Indexing public T this [int index] { get; set; } public List GetRange (int index, int count); public Enumerator GetEnumerator(); // Exporting, copying and converting: public T[] ToArray(); public void CopyTo (T[] array); public void CopyTo (T[] array, int arrayIndex); public void CopyTo (int index, T[] array, int arrayIndex, int count); public ReadOnlyCollection AsReadOnly(); public List ConvertAll (Converter converter); // Other: public void Reverse(); // Reverses order of elements in list. public int Capacity { get;set; } // Forces expansion of internal array. public void TrimExcess(); // Trims internal array back to size. public void Clear(); // Removes all elements, so Count=0. } public delegate TOutput Converter (TInput input); In addition to these members, List provides instance versions of all of Array’s searching and sorting methods. The following code demonstrates List’s properties and methods. See “The Array Class” on page 282 for examples on searching and sorting. List words = new List(); // New string-typed list words.Add ("melon"); words.Add ("avocado"); words.AddRange (new[] { "banana", "plum" } ); words.Insert (0, "lemon"); // Insert at start words.InsertRange (0, new[] { "peach", "nashi" }); // Insert at start words.Remove ("melon"); words.RemoveAt (3); // Remove the 4th element words.RemoveRange (0, 2); // Remove first 2 elements // Remove all strings starting in 'n': words.RemoveAll (s => s.StartsWith ("n")); 292 | Chapter 7: Collections Console.WriteLine (words [0]); // first word Console.WriteLine (words [words.Count - 1]); // last word foreach (string s in words) Console.WriteLine (s); // all words List subset = words.GetRange (1, 2); // 2nd->3rd words string[] wordsArray = words.ToArray(); // Creates a new typed array // Copy first two elements to the end of an existing array: string[] existing = new string [1000]; words.CopyTo (0, existing, 998, 2); List upperCastWords = words.ConvertAll (s => s.ToUpper()); List lengths = words.ConvertAll (s => s.Length); The nongeneric ArrayList class is used mainly for backward compatibility with Framework 1.x code and requires clumsy casts—as the following example demonstrates: ArrayList al = new ArrayList(); al.Add ("hello"); string first = (string) al [0]; string[] strArr = (string[]) al.ToArray (typeof (string)); Such casts cannot be verified by the compiler; the following compiles successfully but then fails at runtime: int first = (int) al [0]; // Runtime exception An ArrayList is functionally similar to List. Both are useful when you need a list of mixed-type elements that share no common base type (other than object). A possible advantage of choosing an ArrayList, in this case, would be if you need to deal with the list using reflection (Chapter 19). Reflection is easier with a nongeneric ArrayList than a List. If you import the System.Linq namespace, you can convert an ArrayList to a generic List by calling Cast and then ToList: ArrayList al = new ArrayList(); al.AddRange (new[] { 1, 5, 9 } ); List list = al.Cast().ToList(); Cast and ToList are extension methods in the System.Linq.Enumerable class. LinkedList LinkedList is a generic doubly linked list (see Figure 7-4). A doubly linked list is a chain of nodes in which each references the node before, the node after, and the actual element. Its main benefit is that an element can always be inserted efficiently anywhere in the list, since it just involves creating a new node and updating a few references. However, finding where to insert the node in the first place can be slow Lists, Queues, Stacks, and Sets | 293 Collections as there’s no intrinsic mechanism to index directly into a linked list; each node must be traversed, and binary-chop searches are not possible. Figure 7-4. LinkedList LinkedList implements IEnumerable and ICollection (and their nongene- ric versions), but not IList since access by index is not supported. List nodes are implemented via the following class: public sealed class LinkedListNode { public LinkedList List { get; } public LinkedListNode Next { get; } public LinkedListNode Previous { get; } public T Value { get; set; } } When adding a node, you can specify its position either relative to another node or at the start/end of the list. LinkedList provides the following methods for this: public void AddFirst(LinkedListNode node); public LinkedListNode AddFirst (T value); public void AddLast (LinkedListNode node); public LinkedListNode AddLast (T value); public void AddAfter (LinkedListNode node, LinkedListNode newNode); public LinkedListNode AddAfter (LinkedListNode node, T value); public void AddBefore (LinkedListNode node, LinkedListNode newNode); public LinkedListNode AddBefore (LinkedListNode node, T value); Similar methods are provided to remove elements: public void Clear(); public void RemoveFirst(); 294 | Chapter 7: Collections public void RemoveLast(); public bool Remove (T value); public void Remove (LinkedListNode node); LinkedList has internal fields to keep track of the number of elements in the list, as well as the head and tail of the list. These are exposed in the following public properties: public int Count { get; } // Fast public LinkedListNode First { get; } // Fast public LinkedListNode Last { get; } // Fast LinkedList also supports the following searching methods (each requiring that the list be internally enumerated): public bool Contains (T value); public LinkedListNode Find (T value); public LinkedListNode FindLast (T value); Finally, LinkedList supports copying to an array for indexed processing and obtaining an enumerator to support the foreach statement: public void CopyTo (T[] array, int index); public Enumerator GetEnumerator(); Here’s a demonstration on the use of LinkedList: var tune = new LinkedList(); tune.AddFirst ("do"); // do tune.AddLast ("so"); // do - so tune.AddAfter (tune.First, "re"); // do - re- so tune.AddAfter (tune.First.Next, "mi"); // do - re - mi- so tune.AddBefore (tune.Last, "fa"); // do - re - mi - fa- so tune.RemoveFirst(); // re - mi - fa - so tune.RemoveLast(); // re - mi - fa LinkedListNode miNode = tune.Find ("mi"); tune.Remove (miNode); // re - fa tune.AddFirst (miNode); // mi- re - fa foreach (string s in tune) Console.WriteLine (s); Queue and Queue Queue and Queue are first-in first-out (FIFO) data structures, providing methods to Enqueue (add an item to the tail of the queue) and Dequeue (retrieve and remove the item at the head of the queue). A Peek method is also provided to return the element at the head of the queue without removing it, and a Count property (useful in checking that elements are present before dequeuing). Lists, Queues, Stacks, and Sets | 295 Collections Although queues are enumerable, they do not implement IList/IList, since members cannot be accessed directly by index. A ToArray method is provided, how- ever, for copying the elements to an array where they can be randomly accessed: public class Queue : IEnumerable, ICollection, IEnumerable { public Queue(); public Queue (IEnumerable collection); // Copies existing elements public Queue (int capacity); // To lessen auto-resizing public void Clear(); public bool Contains (T item); public void CopyTo (T[] array, int arrayIndex); public int Count { get; } public T Dequeue(); public void Enqueue (T item); public Enumerator GetEnumerator(); // To support foreach public T Peek(); public T[] ToArray(); public void TrimExcess(); } The following is an example of using Queue: var q = new Queue(); q.Enqueue (10); q.Enqueue (20); int[] data = q.ToArray(); // Exports to an array Console.WriteLine (q.Count); // "2" Console.WriteLine (q.Peek()); // "10" Console.WriteLine (q.Dequeue()); // "10" Console.WriteLine (q.Dequeue()); // "20" Console.WriteLine (q.Dequeue()); // throws an exception (queue empty) Queues are implemented internally using an array that’s resized as required—much like the generic List class. The queue maintains indexes that point directly to the head and tail elements; therefore, enqueuing and dequeuing are extremely quick operations (except when an internal resize is required). Stack and Stack Stack and Stack are last-in first-out (LIFO) data structures, providing methods to Push (add an item to the top of the stack) and Pop (retrieve and remove an element from the top of the stack). A nondestructive Peek method is also provided, as is a Count property and a ToArray method for exporting the data for random access: public class Stack : IEnumerable, ICollection, IEnumerable { public Stack(); public Stack (IEnumerable collection); // Copies existing elements public Stack (int capacity); // Lessens auto-resizing public void Clear(); public bool Contains (T item); public void CopyTo (T[] array, int arrayIndex); public int Count { get; } public Enumerator GetEnumerator(); // To support foreach public T Peek(); 296 | Chapter 7: Collections public T Pop(); public void Push (T item); public T[] ToArray(); public void TrimExcess(); } The following example demonstrates Stack: var s = new Stack(); s.Push (1); // Stack = 1 s.Push (2); // Stack = 1,2 s.Push (3); // Stack = 1,2,3 Console.WriteLine (s.Count); // Prints 3 Console.WriteLine (s.Peek()); // Prints 3, Stack = 1,2,3 Console.WriteLine (s.Pop()); // Prints 3, Stack = 1,2 Console.WriteLine (s.Pop()); // Prints 2, Stack = 1 Console.WriteLine (s.Pop()); // Prints 1, Stack = Console.WriteLine (s.Pop()); // throws exception Stacks are implemented internally with an array that’s resized as required, as with Queue and List. BitArray A BitArray is a dynamically sized collection of compacted bool values. It is more memory-efficient than both a simple array of bool and a generic List of bool, because it uses only one bit for each value, whereas the bool type otherwise occupies one byte for each value. BitArray’s indexer reads and writes individual bits: var bits = new BitArray(2); bits[1] = true; There are four bitwise operator methods (And, Or, Xor and Not). All but the last accept another BitArray: bits.Xor (bits); // Bitwise exclusive-OR bits with itself Console.WriteLine (bits[1]); // False HashSet and SortedSet HashSet and SortedSet are generic collections new to Framework 3.5 and 4.0, respectively. Both have the following distinguishing features: • Their Contains methods execute quickly using a hash-based lookup. • They do not store duplicate elements and silently ignore requests to add duplicates. • You cannot access an element by position. SortedSet keeps elements in order whereas HashSet does not. Lists, Queues, Stacks, and Sets | 297 Collections The commonality of these types is captured by the interface ISet. For historical reasons, HashSet lives in System.Core.dll (whereas SortedSet and ISet live in System.dll). HashSet is implemented with a hashtable that stores just keys; SortedSet is implemented with a red/black tree. Both collections implement ICollection and offer methods that you would expect, such as Contains, Add and Remove. In addition, there’s a predicate-based re- moval method called RemoveWhere. The following constructs a HashSet from an existing collection, tests for mem- bership, and then enumerates the collection (notice the absence of duplicates): var letters = new HashSet ("the quick brown fox"); Console.WriteLine (letters.Contains ('t')); // true Console.WriteLine (letters.Contains ('j')); // false foreach (char c in letters) Console.Write (c); // the quickbrownfx (The reason we can pass a string into HashSet’s constructor is because string implements IEnumerable.) The really interesting methods are the set operations. The following set operations are destructive, in that they modify the set: public void UnionWith (IEnumerable other); // Adds public void IntersectWith (IEnumerable other); // Removes public void ExceptWith (IEnumerable other); // Removes public void SymmetricExceptWith (IEnumerable other); // Removes whereas the following methods simply query the set and so are non-destructive: public bool IsSubsetOf (IEnumerable other); public bool IsProperSubsetOf (IEnumerable other); public bool IsSupersetOf (IEnumerable other); public bool IsProperSupersetOf (IEnumerable other); public bool Overlaps (IEnumerable other); public bool SetEquals (IEnumerable other); UnionWith adds all the elements in the second set to the original set (excluding duplicates). IntersectWith removes the elements that are not in both sets. We can extract all the vowels from our set of characters as follows: var letters = new HashSet ("the quick brown fox"); letters.IntersectWith ("aeiou"); foreach (char c in letters) Console.Write (c); // euio ExceptWith removes the specified elements from the source set. Here, we strip all vowels from the set: var letters = new HashSet ("the quick brown fox"); letters.ExceptWith ("aeiou"); foreach (char c in letters) Console.Write (c); // th qckbrwnfx 298 | Chapter 7: Collections SymmetricExceptWith removes all but the elements that are unique to one set or the other: var letters = new HashSet ("the quick brown fox"); letters.SymmetricExceptWith ("the lazy brown fox"); foreach (char c in letters) Console.Write (c); // quicklazy Note that because HashSet and SortedSet implement IEnumerable, you can use another type of set (or collection) as the argument to any of the set operation methods. SortedSet offers all the members of HashSet, plus the following: public virtual SortedSet GetViewBetween (T lowerValue, T upperValue) public IEnumerable Reverse() public T Min { get; } public T Max { get; } SortedSet also accepts an optional IComparer in its constructor (rather than an equality comparer). Here’s an example of loading the same letters into a SortedSet: var letters = new SortedSet ("the quick brown fox"); foreach (char c in letters) Console.Write (c); // bcefhiknoqrtuwx Following on from this, we can obtain the letters between f and j as follows: foreach (char c in letters.GetViewBetween ('f', 'j')) Console.Write (c); // fhi Dictionaries A dictionary is a collection in which each element is a key/value pair. Dictionaries are most commonly used for lookups and sorted lists. The Framework defines a standard protocol for dictionaries, via the interfaces IDic tionary and IDictionary , as well as a set of general-purpose dictio- nary classes. The classes each differ in the following regard: • Whether or not items are stored in sorted sequence • Whether or not items can be accessed by position (index) as well as by key • Whether generic or nongeneric • Whether it’s fast or slow to retrieve items by key from a large dictionary Table 7-1 summarizes each of the dictionary classes and how they differ in these respects. The performance times are in milliseconds, to perform 50,000 operations on a dictionary with integer keys and values, on a 1.5 GHz PC. (The differences in performance between generic and nongeneric counterparts using the same under- lying collection structure are due to boxing, and show up only with value-type elements.) Dictionaries | 299 Collections Table 7-1. Dictionary classes Type Internal structure Retrieve by index? Memory overhead (avg. bytes per item) Speed: random insertion Speed: sequential insertion Speed: retrieval by key Unsorted Dictionary Hashtable No 22 30 30 20 Hashtable Hashtable No 38 50 50 30 ListDictionary Linked list No 36 50,000 50,000 50,000 OrderedDictionary Hashtable + array Yes 59 70 70 40 Sorted SortedDictionary Red/black tree No 20 130 100 120 SortedList 2xArray Yes 2 3,300 30 40 SortedList 2xArray Yes 27 4,500 100 180 In Big-O notation, retrieval time by key is: • O(1) for Hashtable, Dictionary, and OrderedDictionary • O(log n) for SortedDictionary and SortedList • O(n) for ListDictionary (and nondictionary types such as List) where n is the number of elements in the collection. IDictionary IDictionary defines the standard protocol for all key/value-based col- lections. It extends ICollection by adding methods and properties to access elements based on a key of arbitrary type: public interface IDictionary : ICollection >, IEnumerable { bool ContainsKey (TKey key); bool TryGetValue (TKey key, out TValue value); void Add (TKey key, TValue value); bool Remove (TKey key); TValue this [TKey key] { get; set; } // Main indexer - by key ICollection Keys { get; } // Returns just keys ICollection Values { get; } // Returns just values } 300 | Chapter 7: Collections From Framework 4.5, there’s also an interface called IReadOnly Dictionary which defines the read-only subset of dictionary members. This maps to the Windows Runtime type IMapView, and was introduced primarily for that reason. To add an item to a dictionary, you either call Add or use the index’s set accessor— the latter adds an item to the dictionary if the key is not already present (or updates the item if it is present). Duplicate keys are forbidden in all dictionary implementa- tions, so calling Add twice with the same key throws an exception. To retrieve an item from a dictionary, use either the indexer or the TryGetValue method. If the key doesn’t exist, the indexer throws an exception whereas TryGet Value returns false. You can test for membership explicitly by calling ContainsKey; however, this incurs the cost of two lookups if you then subsequently retrieve the item. Enumerating directly over an IDictionary returns a sequence of Key ValuePair structs: public struct KeyValuePair { public TKey Key { get; } public TValue Value { get; } } You can enumerate over just the keys or values via the dictionary’s Keys/Values properties. We demonstrate the use of this interface with the generic Dictionary class in the following section. IDictionary The nongeneric IDictionary interface is the same in principle as IDictionary, apart from two important functional differences. It’s important to be aware of these differences, because IDictionary appears in legacy code (including the .NET Framework itself in places): • Retrieving a nonexistent key via the indexer returns null (rather than throwing an exception). • Contains tests for membership rather than ContainsKey. Enumerating over a nongeneric IDictionary returns a sequence of DictionaryEn try structs: public struct DictionaryEntry { public object Key { get; set; } public object Value { get; set; } } Dictionaries | 301 Collections Dictionary and Hashtable The generic Dictionary class is one of the most commonly used collections (along with the List collection). It uses a hashtable data structure to store keys and values, and it is fast and efficient. The nongeneric version of Dictionary is called Hashtable; there is no nongeneric class called Dictionary. When we refer simply to Dictionary, we mean the generic Dictio nary class. Dictionary implements both the generic and nongeneric IDictionary interfaces, the generic IDictionary being exposed publicly. Dictionary is, in fact, a “textbook” implementation of the generic IDictionary. Here’s how to use it: var d = new Dictionary(); d.Add("One", 1); d["Two"] = 2; // adds to dictionary because "two" is not already present d["Two"] = 22; // updates dictionary because "two" is now present d["Three"] = 3; Console.WriteLine (d["Two"]); // Prints "22" Console.WriteLine (d.ContainsKey ("One")); // true (fast operation) Console.WriteLine (d.ContainsValue (3)); // true (slow operation) int val = 0; if (!d.TryGetValue ("onE", out val)) Console.WriteLine ("No val"); // "No val" (case sensitive) // Three different ways to enumerate the dictionary: foreach (KeyValuePair kv in d) // One ; 1 Console.WriteLine (kv.Key + "; " + kv.Value); // Two ; 22 // Three ; 3 foreach (string s in d.Keys) Console.Write (s); // OneTwoThree Console.WriteLine(); foreach (int i in d.Values) Console.Write (i); // 1223 Its underlying hashtable works by converting each element’s key into an integer hashcode—a pseudo-unique value—and then applying an algorithm to convert the hashcode into a hash key. This hash key is used internally to determine which “bucket” an entry belongs to. If the bucket contains more than one value, a linear search is performed on the bucket. A good hash function does not strive to return strictly unique hashcodes (which would usually be impossible); it strives to return hashcodes that are evenly distributed across the 32-bit integer space. This avoids the scenario of ending up with a few very large (and inefficient) buckets. A dictionary can work with keys of any type, providing it’s able to determine equality between keys and obtain hashcodes. By default, equality is determined via the key’s 302 | Chapter 7: Collections object.Equals method, and the pseudo-unique hashcode is obtained via the key’s GetHashCode method. This behavior can be changed, either by overriding these methods or by providing an IEqualityComparer object when constructing the dictionary. A common application of this is to specify a case-insensitive equality comparer when using string keys: var d = new Dictionary (StringComparer.OrdinalIgnoreCase); We discuss this further in “Plugging in Equality and Order” on page 312. As with many other types of collections, the performance of a dictionary can be improved slightly by specifying the collection’s expected size in the constructor, avoiding or lessening the need for internal resizing operations. The nongeneric version is named Hashtable and is functionally similar apart from differences stemming from it exposing the nongeneric IDictionary interface discussed previously. The downside to Dictionary and Hashtable is that the items are not sorted. Further- more, the original order in which the items were added is not retained. As with all dictionaries, duplicate keys are not allowed. When the generic collections were introduced in Framework 2.0, the CLR team chose to name them according to what they represent (Dictionary, List) rather than how they are internally implemented (Hashtable, ArrayList). While this is good because it gives them the freedom to later change the imple- mentation, it also means that the performance contract (often the most important criteria in choosing one kind of collection over another) is no longer captured in the name. OrderedDictionary An OrderedDictionary is a nongeneric dictionary that maintains elements in the same order that they were added. With an OrderedDictionary, you can access elements both by index and by key. An OrderedDictionary is not a sorted dictionary. An OrderedDictionary is a combination of a Hashtable and an ArrayList. This means it has all the functionality of a Hashtable, plus functions such as RemoveAt, as well as an integer indexer. It also exposes Keys and Values properties that return elements in their original order. This class was introduced in .NET 2.0, yet peculiarly, there’s no generic version. Dictionaries | 303 Collections ListDictionary and HybridDictionary ListDictionary uses a singly linked list to store the underlying data. It doesn’t pro- vide sorting, although it does preserve the original entry order of the items. ListDic tionary is extremely slow with large lists. Its only real “claim to fame” is its efficiency with very small lists (fewer than 10 items). HybridDictionary is a ListDictionary that automatically converts to a Hashtable upon reaching a certain size, to address ListDictionary’s problems with perfor- mance. The idea is to get a low memory footprint when the dictionary is small, and good performance when the dictionary is large. However, given the overhead in converting from one to the other—and the fact that a Dictionary is not excessively heavy or slow in either scenario—you wouldn’t suffer unreasonably by using a Dictionary to begin with. Both classes come only in nongeneric form. Sorted Dictionaries The Framework provides two dictionary classes internally structured such that their content is always sorted by key: • SortedDictionary • SortedList1 (In this section, we will abbreviate to <,>.) SortedDictionary<,> uses a red/black tree: a data structure designed to perform consistently well in any insertion or retrieval scenario. SortedList<,> is implemented internally with an ordered array pair, providing fast retrieval (via a binary-chop search) but poor insertion performance (because existing values have to be shifted to make room for a new entry). SortedDictionary<,> is much faster than SortedList<,> at inserting elements in a random sequence (particularly with large lists). SortedList<,>, however, has an extra ability: to access items by index as well as by key. With a sorted list, you can go directly to the nth element in the sorting sequence (via the indexer on the Keys/ Values properties). To do the same with a SortedDictionary<,>, you must manually enumerate over n items. (Alternatively, you could write a class that combines a sorted dictionary with a list class.) None of the three collections allows duplicate keys (as is the case with all dictionaries). 1. There’s also a functionally identical nongeneric version of this called SortedList. 304 | Chapter 7: Collections The following example uses reflection to load all the methods defined in Sys tem.Object into a sorted list keyed by name, and then enumerates their keys and values: // MethodInfo is in the System.Reflection namespace var sorted = new SortedList (); foreach (MethodInfo m in typeof (object).GetMethods()) sorted [m.Name] = m; foreach (string name in sorted.Keys) Console.WriteLine (name); foreach (MethodInfo m in sorted.Values) Console.WriteLine (m.Name + " returns a " + m.ReturnType); Here’s the result of the first enumeration: Equals GetHashCode GetType ReferenceEquals ToString Here’s the result of the second enumeration: Equals returns a System.Boolean GetHashCode returns a System.Int32 GetType returns a System.Type ReferenceEquals returns a System.Boolean ToString returns a System.String Notice that we populated the dictionary through its indexer. If we instead used the Add method, it would throw an exception because the object class upon which we’re reflecting overloads the Equals method, and you can’t add the same key twice to a dictionary. By using the indexer, the later entry overwrites the earlier entry, pre- venting this error. You can store multiple members of the same key by making each value element a list: SortedList > Extending our example, the following retrieves the MethodInfo whose key is "GetHash Code", just as with an ordinary dictionary: Console.WriteLine (sorted ["GetHashCode"]); // Int32 GetHashCode() So far, everything we’ve done would also work with a SortedDictionary<,>. The following two lines, however, which retrieve the last key and value, work only with a sorted list: Console.WriteLine (sorted.Keys [sorted.Count - 1]); // ToString Console.WriteLine (sorted.Values[sorted.Count - 1].IsVirtual); // True Dictionaries | 305 Collections Customizable Collections and Proxies The collection classes discussed in previous sections are convenient in that they can be directly instantiated, but they don’t allow you to control what happens when an item is added to or removed from the collection. With strongly typed collections in an application, you sometimes need this control—for instance: • To fire an event when an item is added or removed • To update properties because of the added or removed item • To detect an “illegal” add/remove operation and throw an exception (for ex- ample, if the operation violates a business rule) The .NET Framework provides collection classes for this exact purpose, in the System.Collections.ObjectModel namespace. These are essentially proxies or wrap- pers that implement IList or IDictionary<,> by forwarding the methods through to an underlying collection. Each Add, Remove, or Clear operation is routed via a virtual method that acts as a “gateway” when overridden. Customizable collection classes are commonly used for publicly exposed collections; for instance, a collection of controls exposed publicly on a System.Windows.Form class. Collection and CollectionBase Collection class is a customizable wrapper for List. As well as implementing IList and IList, it defines four additional virtual meth- ods and a protected property as follows: public class Collection : IList, ICollection, IEnumerable, IList, ICollection, IEnumerable { // ... protected virtual void ClearItems(); protected virtual void InsertItem (int index, T item); protected virtual void RemoveItem (int index); protected virtual void SetItem (int index, T item); protected IList Items { get; } } The virtual methods provide the gateway by which you can “hook in” to change or enhance the list’s normal behavior. The protected Items property allows the imple- menter to directly access the “inner list”—this is used to make changes internally without the virtual methods firing. The virtual methods need not be overridden; they can be left alone until there’s a requirement to alter the list’s default behavior. The following example demonstrates the typical “skeleton” use of Collection: public class Animal { 306 | Chapter 7: Collections public string Name; public int Popularity; public Animal (string name, int popularity) { Name = name; Popularity = popularity; } } public class AnimalCollection : Collection { // AnimalCollection is already a fully functioning list of animals. // No extra code is required. } public class Zoo // The class that will expose AnimalCollection. { // This would typically have additional members. public readonly AnimalCollection Animals = new AnimalCollection(); } class Program { static void Main() { Zoo zoo = new Zoo(); zoo.Animals.Add (new Animal ("Kangaroo", 10)); zoo.Animals.Add (new Animal ("Mr Sea Lion", 20)); foreach (Animal a in zoo.Animals) Console.WriteLine (a.Name); } } As it stands, AnimalCollection is no more functional than a simple List; its role is to provide a base for future extension. To illustrate, we’ll now add a Zoo property to Animal, so it can reference the Zoo in which it lives and override each of the virtual methods in Collection to maintain that property automatically: public class Animal { public string Name; public int Popularity; public Zoo Zoo { get; internal set; } public Animal(string name, int popularity) { Name = name; Popularity = popularity; } } public class AnimalCollection : Collection { Zoo zoo; public AnimalCollection (Zoo zoo) { this.zoo = zoo; } protected override void InsertItem (int index, Animal item) { base.InsertItem (index, item); Customizable Collections and Proxies | 307 Collections item.Zoo = zoo; } protected override void SetItem (int index, Animal item) { base.SetItem (index, item); item.Zoo = zoo; } protected override void RemoveItem (int index) { this [index].Zoo = null; base.RemoveItem (index); } protected override void ClearItems() { foreach (Animal a in this) a.Zoo = null; base.ClearItems(); } } public class Zoo { public readonly AnimalCollection Animals; public Zoo() { Animals = new AnimalCollection (this); } } Collection also has a constructor accepting an existing IList. Unlike with other collection classes, the supplied list is proxied rather than copied, meaning that subsequent changes will be reflected in the wrapping Collection (although without Collection’s virtual methods firing). Conversely, changes made via the Collection will change the underlying list. CollectionBase CollectionBase is the nongeneric version of Collection introduced in Framework 1.0. This provides most of the same features as Collection, but is clumsier to use. Instead of the template methods InsertItem, RemoveItem, SetItem, and ClearItem, CollectionBase has “hook” methods that double the number of methods required: OnInsert, OnInsertComplete, OnSet, OnSetComplete, OnRemove, OnRemoveComplete, OnClear, and OnClearComplete. Because CollectionBase is nongeneric, you must also implement typed methods when subclassing it—at a minimum, a typed indexer and Add method. KeyedCollection and DictionaryBase KeyedCollection subclasses Collection. It both adds and sub- tracts functionality. What it adds is the ability to access items by key, much like with a dictionary. What it subtracts is the ability to proxy your own inner list. A keyed collection has some resemblance to an OrderedDictionary in that it com- bines a linear list with a hashtable. However, unlike OrderedDictionary, it doesn’t implement IDictionary and doesn’t support the concept of a key/value pair. Keys are obtained instead from the items themselves: via the abstract GetKeyForItem 308 | Chapter 7: Collections method. This means enumerating a keyed collection is just like enumerating an ordinary list. KeyedCollection is best thought of as Collection plus fast lookup by key. Because it subclasses Collection<>, a keyed collection inherits all of Collection<>’s functionality, except for the ability to specify an existing list in construction. The additional members it defines are as follows: public abstract class KeyedCollection : Collection // ... protected abstract TKey GetKeyForItem(TItem item); protected void ChangeItemKey(TItem item, TKey newKey); // Fast lookup by key - this is in addition to lookup by index. public TItem this[TKey key] { get; } protected IDictionary Dictionary { get; } } GetKeyForItem is what the implementer overrides to obtain an item’s key from the underlying object. The ChangeItemKey method must be called if the item’s key prop- erty changes, in order to update the internal dictionary. The Dictionary property returns the internal dictionary used to implement the lookup, which is created when the first item is added. This behavior can be changed by specifying a creation thresh- old in the constructor, delaying the internal dictionary from being created until the threshold is reached (in the interim, a linear search is performed if an item is requested by key). A good reason not to specify a creation threshold is that having a valid dictionary can be useful in obtaining an ICollection<> of keys, via the Dic tionary’s Keys property. This collection can then be passed on to a public property. The most common use for KeyedCollection<,> is in providing a collection of items accessible both by index and by name. To demonstrate this, we’ll revisit the zoo, this time implementing AnimalCollection as a KeyedCollection: public class Animal { string name; public string Name { get { return name; } set { if (Zoo != null) Zoo.Animals.NotifyNameChange (this, value); name = value; } } public int Popularity; public Zoo Zoo { get; internal set; } public Animal (string name, int popularity) { Name = name; Popularity = popularity; Customizable Collections and Proxies | 309 Collections } } public class AnimalCollection : KeyedCollection { Zoo zoo; public AnimalCollection (Zoo zoo) { this.zoo = zoo; } internal void NotifyNameChange (Animal a, string newName) { this.ChangeItemKey (a, newName); } protected override string GetKeyForItem (Animal item) { return item.Name; } // The following methods would be implemented as in the previous example protected override void InsertItem (int index, Animal item)... protected override void SetItem (int index, Animal item)... protected override void RemoveItem (int index)... protected override void ClearItems()... } public class Zoo { public readonly AnimalCollection Animals; public Zoo() { Animals = new AnimalCollection (this); } } class Program { static void Main() { Zoo zoo = new Zoo(); zoo.Animals.Add (new Animal ("Kangaroo", 10)); zoo.Animals.Add (new Animal ("Mr Sea Lion", 20)); Console.WriteLine (zoo.Animals [0].Popularity); // 10 Console.WriteLine (zoo.Animals ["Mr Sea Lion"].Popularity); // 20 zoo.Animals ["Kangaroo"].Name = "Mr Roo"; Console.WriteLine (zoo.Animals ["Mr Roo"].Popularity); // 10 } } DictionaryBase The nongeneric version of KeyedCollection is called DictionaryBase. This legacy class takes very different in its approach: it implements IDictionary and uses clumsy hook methods like CollectionBase : OnInsert, OnInsertComplete, OnSet, OnSetCom plete, OnRemove, OnRemoveComplete, OnClear, and OnClearComplete (and additionally, OnGet). The primary advantage of implementing IDictionary over taking the Keyed Collection approach is that you don’t need to subclass it in order to obtain keys. But since the very purpose of DictionaryBase is to be subclassed, it’s no advantage 310 | Chapter 7: Collections at all. The improved model in KeyedCollection is almost certainly due to the fact that it was written some years later, with the benefit of hindsight. DictionaryBase is best considered useful for backward compatibility. ReadOnlyCollection ReadOnlyCollection is a wrapper, or proxy, that provides a read-only view of a collection. This is useful in allowing a class to publicly expose read-only access to a collection that the class can still update internally. A read-only collection accepts the input collection in its constructor, to which it maintains a permanent reference. It doesn’t take a static copy of the input collection, so subsequent changes to the input collection are visible through the read-only wrapper. To illustrate, suppose your class wants to provide read-only public access to a list of strings called Names: public class Test { public List Names { get; private set; } } This does only half the job. Although other types cannot reassign the Names property, they can still call Add, Remove, or Clear on the list. The ReadOnlyCollection class resolves this: public class Test { List names; public ReadOnlyCollection Names { get; private set; } public Test() { names = new List(); Names = new ReadOnlyCollection (names); } public void AddInternally() { names.Add ("test"); } } Now, only members within the Test class can alter the list of names: Test t = new Test(); Console.WriteLine (t.Names.Count); // 0 t.AddInternally(); Console.WriteLine (t.Names.Count); // 1 t.Names.Add ("test"); // Compiler error ((IList) t.Names).Add ("test"); // NotSupportedException Customizable Collections and Proxies | 311 Collections Plugging in Equality and Order In the sections “Equality Comparison” on page 254 and “Order Compari- son” on page 264 in Chapter 6, we described the standard .NET protocols that make a type equatable, hashable, and comparable. A type that implements these protocols can function correctly in a dictionary or sorted list “out of the box.” More specifically: • A type for which Equals and GetHashCode return meaningful results can be used as a key in a Dictionary or Hashtable. • A type that implements IComparable /IComparable can be used as a key in any of the sorted dictionaries or lists. A type’s default equating or comparison implementation typically reflects what is most “natural” for that type. Sometimes, however, the default behavior is not what you want. You might need a dictionary whose string -type key is treated case-insensitively. Or you might want a sorted list of customers, sorted by each customer’s postcode. For this reason, the .NET Framework also defines a matching set of “plug-in” protocols. The plug-in protocols achieve two things: • They allow you to switch in alternative equating or comparison behavior. • They allow you to use a dictionary or sorted collection with a key type that’s not intrinsically equatable or comparable. The plug-in protocols consist of the following interfaces: IEqualityComparer and IEqualityComparer • Performs plug-in equality comparison and hashing • Recognized by Hashtable and Dictionary IComparer and IComparer • Performs plug-in order comparison • Recognized by the sorted dictionaries and collections; also, Array.Sort Each interface comes in both generic and nongeneric forms. The IEqualityComparer interfaces also have a default implementation in a class called EqualityComparer. In addition, in Framework 4.0 we got two new interfaces called IStructuralEquata ble and IStructuralComparable which allow the option of structural comparisons on such as classes and arrays. IEqualityComparer and EqualityComparer An equality comparer switches in nondefault equality and hashing behavior, pri- marily for the Dictionary and Hashtable classes. 312 | Chapter 7: Collections Recall the requirements of a hashtable-based dictionary. It needs answers to two questions for any given key: • Is it the same as another? • What is its integer hashcode? An equality comparer answers these questions by implementing the IEqualityCom parer interfaces: public interface IEqualityComparer { bool Equals (T x, T y); int GetHashCode (T obj); } public interface IEqualityComparer // Nongeneric version { bool Equals (object x, object y); int GetHashCode (object obj); } To write a custom comparer, you implement one or both of these interfaces (implementing both gives maximum interoperability). As this is somewhat tedious, an alternative is to subclass the abstract EqualityComparer class, defined as follows: public abstract class EqualityComparer : IEqualityComparer, IEqualityComparer { public abstract bool Equals (T x, T y); public abstract int GetHashCode (T obj); bool IEqualityComparer.Equals (object x, object y); int IEqualityComparer.GetHashCode (object obj); public static EqualityComparer Default { get; } } EqualityComparer implements both interfaces; your job is simply to override the two abstract methods. The semantics for Equals and GetHashCode follow the same rules for object.Equals and object.GetHashCode, described in Chapter 6. In the following example, we define a Customer class with two fields, and then write an equality comparer that matches both the first and last names: public class Customer { public string LastName; public string FirstName; public Customer (string last, string first) { LastName = last; FirstName = first; } } Plugging in Equality and Order | 313 Collections public class LastFirstEqComparer : EqualityComparer { public override bool Equals (Customer x, Customer y) { return x.LastName == y.LastName && x.FirstName == y.FirstName; } public override int GetHashCode (Customer obj) { return (obj.LastName + ";" + obj.FirstName).GetHashCode(); } } To illustrate how this works, we’ll create two customers: Customer c1 = new Customer ("Bloggs", "Joe"); Customer c2 = new Customer ("Bloggs", "Joe"); Because we’ve not overridden object.Equals, normal reference type equality semantics apply: Console.WriteLine (c1 == c2); // False Console.WriteLine (c1.Equals (c2)); // False The same default equality semantics apply when using these customers in a Dictio nary without specifying an equality comparer: var d = new Dictionary(); d [c1] = "Joe"; Console.WriteLine (d.ContainsKey (c2)); // False Now with the custom equality comparer: var eqComparer = new LastFirstEqComparer(); var d = new Dictionary (eqComparer); d [c1] = "Joe"; Console.WriteLine (d.ContainsKey (c2)); // True In this example, we would have to be careful not to change the customer’s First Name or LastName while it was in use in the dictionary. Otherwise, its hashcode would change and the Dictionary would break. EqualityComparer.Default Calling EqualityComparer.Default returns a general-purpose equality comparer that can be used as an alternative to the static object.Equals method. The advantage is that it first checks if T implements IEquatable and if so, calls that implemen- tation instead, avoiding the boxing overhead. This is particularly useful in generic methods: static bool Foo (T x, T y) { bool same = EqualityComparer.Default.Equals (x, y); ... 314 | Chapter 7: Collections IComparer and Comparer Comparers are used to switch in custom ordering logic for sorted dictionaries and collections. Note that a comparer is useless to the unsorted dictionaries such as Dictionary and Hashtable—these require an IEqualityComparer to get hashcodes. Similarly, an equality comparer is useless for sorted dictionaries and collections. Here are the IComparer interface definitions: public interface IComparer { int Compare(object x, object y); } public interface IComparer { int Compare(T x, T y); } As with equality comparers, there’s an abstract class you can subtype instead of implementing the interfaces: public abstract class Comparer : IComparer, IComparer { public static Comparer Default { get; } public abstract int Compare (T x, T y); // Implemented by you int IComparer.Compare (object x, object y); // Implemented for you } The following example illustrates a class that describes a wish, and a comparer that sorts wishes by priority: class Wish { public string Name; public int Priority; public Wish (string name, int priority) { Name = name; Priority = priority; } } class PriorityComparer : Comparer { public override int Compare (Wish x, Wish y) { if (object.Equals (x, y)) return 0; // Fail-safe check return x.Priority.CompareTo (y.Priority); } } Plugging in Equality and Order | 315 Collections The object.Equals check ensures that we can never contradict the Equals method. Calling the static object.Equals method in this case is better than calling x.Equals because it still works if x is null! Here’s how our PriorityComparer is used to sort a List: var wishList = new List(); wishList.Add (new Wish ("Peace", 2)); wishList.Add (new Wish ("Wealth", 3)); wishList.Add (new Wish ("Love", 2)); wishList.Add (new Wish ("3 more wishes", 1)); wishList.Sort (new PriorityComparer()); foreach (Wish w in wishList) Console.Write (w.Name + " | "); // OUTPUT: 3 more wishes | Love | Peace | Wealth | In the next example, SurnameComparer allows you to sort surname strings in an order suitable for a phonebook listing: class SurnameComparer : Comparer { string Normalize (string s) { s = s.Trim().ToUpper(); if (s.StartsWith ("MC")) s = "MAC" + s.Substring (2); return s; } public override int Compare (string x, string y) { return Normalize (x).CompareTo (Normalize (y)); } } Here’s SurnameComparer in use in a sorted dictionary: var dic = new SortedDictionary (new SurnameComparer()); dic.Add ("MacPhail", "second!"); dic.Add ("MacWilliam", "third!"); dic.Add ("McDonald", "first!"); foreach (string s in dic.Values) Console.Write (s + " "); // first! second! third! StringComparer StringComparer is a predefined plug-in class for equating and comparing strings, allowing you to specify language and case sensitivity. StringComparer implements both IEqualityComparer and IComparer (and their generic versions), so it can be used with any type of dictionary or sorted collection: // CultureInfo is defined in System.Globalization public abstract class StringComparer : IComparer, IComparer , IEqualityComparer, IEqualityComparer 316 | Chapter 7: Collections { public abstract int Compare (string x, string y); public abstract bool Equals (string x, string y); public abstract int GetHashCode (string obj); public static StringComparer Create (CultureInfo culture, bool ignoreCase); public static StringComparer CurrentCulture { get; } public static StringComparer CurrentCultureIgnoreCase { get; } public static StringComparer InvariantCulture { get; } public static StringComparer InvariantCultureIgnoreCase { get; } public static StringComparer Ordinal { get; } public static StringComparer OrdinalIgnoreCase { get; } } Because StringComparer is abstract, you obtain instances via its static methods and properties. StringComparer.Ordinal mirrors the default behavior for string equality comparison and StringComparer.CurrentCulture for order comparison. In the following example, an ordinal case-insensitive dictionary is created, such that dict["Joe"] and dict["JOE"] mean the same thing: var dict = new Dictionary (StringComparer.OrdinalIgnoreCase); In the next example, an array of names is sorted, using Australian English: string[] names = { "Tom", "HARRY", "sheila" }; CultureInfo ci = new CultureInfo ("en-AU"); Array.Sort (names, StringComparer.Create (ci, false)); The final example is a culture-aware version of the SurnameComparer we wrote in the previous section (to compare names suitable for a phonebook listing): class SurnameComparer : Comparer { StringComparer strCmp; public SurnameComparer (CultureInfo ci) { // Create a case-sensitive, culture-sensitive string comparer strCmp = StringComparer.Create (ci, false); } string Normalize (string s) { s = s.Trim(); if (s.ToUpper().StartsWith ("MC")) s = "MAC" + s.Substring (2); return s; } public override int Compare (string x, string y) { // Directly call Compare on our culture-aware StringComparer return strCmp.Compare (Normalize (x), Normalize (y)); } } Plugging in Equality and Order | 317 Collections IStructuralEquatable and IStructuralComparable As we said in the previous chapter, structs implement structural comparison by default: two structs are equal if all of their fields are equal. Sometimes, however, structural equality and order comparison are useful as plug-in options on other types as well—such as arrays and tuples. Framework 4.0 introduced two new interfaces to help with this: public interface IStructuralEquatable { bool Equals (object other, IEqualityComparer comparer); int GetHashCode (IEqualityComparer comparer); } public interface IStructuralComparable { int CompareTo (object other, IComparer comparer); } The IEqualityComparer/IComparer that you pass in are applied to each individual element in the composite object. We can demonstrate this using arrays and tuples, which implement these interfaces: in the following example, we compare two arrays for equality: first using the default Equals method, then using IStructuralEquata ble’s version: int[] a1 = { 1, 2, 3 }; int[] a2 = { 1, 2, 3 }; IStructuralEquatable se1 = a1; Console.Write (a1.Equals (a2)); // False Console.Write (se1.Equals (a2, EqualityComparer.Default)); // True Here’s another example: string[] a1 = "the quick brown fox".Split(); string[] a2 = "THE QUICK BROWN FOX".Split(); IStructuralEquatable se1 = a1; bool isTrue = se1.Equals (a2, StringComparer.InvariantCultureIgnoreCase); Tuples work in the same way: var t1 = Tuple.Create (1, "foo"); var t2 = Tuple.Create (1, "FOO"); IStructuralEquatable se1 = t1; bool isTrue = se1.Equals (t2, StringComparer.InvariantCultureIgnoreCase); IStructuralComparable sc1 = t1; int zero = sc1.CompareTo (t2, StringComparer.InvariantCultureIgnoreCase); The difference with tuples, though, is that their default equality and order compar- ison implementations also apply structural comparisons: var t1 = Tuple.Create (1, "FOO"); var t2 = Tuple.Create (1, "FOO"); Console.WriteLine (t1.Equals (t2)); // True 318 | Chapter 7: Collections 8LINQ Queries LINQ, or Language Integrated Query, is a set of language and framework features for writing structured type-safe queries over local object collections and remote data sources. LINQ was introduced in C# 3.0 and Framework 3.5. LINQ enables you to query any collection implementing IEnumerable, whether an array, list, or XML DOM, as well as remote data sources, such as tables in SQL Server. LINQ offers the benefits of both compile-time type checking and dynamic query composition. This chapter describes the LINQ architecture and the fundamentals of writing quer- ies. All core types are defined in the System.Linq and System.Linq.Expressions namespaces. The examples in this and the following two chapters are pre- loaded into an interactive querying tool called LINQPad. You can download LINQPad from http://www.linqpad.net. Getting Started The basic units of data in LINQ are sequences and elements. A sequence is any object that implements IEnumerable and an element is each item in the sequence. In the following example, names is a sequence, and "Tom", "Dick", and "Harry" are elements: string[] names = { "Tom", "Dick", "Harry" }; We call this a local sequence because it represents a local collection of objects in memory. A query operator is a method that transforms a sequence. A typical query operator accepts an input sequence and emits a transformed output sequence. In the Enumera ble class in System.Linq, there are around 40 query operators—all implemented as static extension methods. These are called standard query operators. 319 Queries that operate over local sequences are called local quer- ies or LINQ-to-objects queries. LINQ also supports sequences that can be dynamically fed from a remote data source such as a SQL Server. These sequences additionally implement the IQueryable interface and are supported through a matching set of standard query operators in the Queryable class. We discuss this further in the section “Interpreted Queries” on page 347 in this chapter. A query is an expression that, when enumerated, transforms sequences with query operators. The simplest query comprises one input sequence and one operator. For instance, we can apply the Where operator on a simple array to extract those whose length is at least four characters as follows: string[] names = { "Tom", "Dick", "Harry" }; IEnumerable filteredNames = System.Linq.Enumerable.Where (names, n => n.Length >= 4); foreach (string n in filteredNames) Console.WriteLine (n); Dick Harry Because the standard query operators are implemented as extension methods, we can call Where directly on names—as though it were an instance method: IEnumerable filteredNames = names.Where (n => n.Length >= 4); For this to compile, you must import the System.Linq namespace. Here’s a complete example: using System; usign System.Collections.Generic; using System.Linq; class LinqDemo { static void Main() { string[] names = { "Tom", "Dick", "Harry" }; IEnumerable filteredNames = names.Where (n => n.Length >= 4); foreach (string name in filteredNames) Console.WriteLine (name); } } Dick Harry 320 | Chapter 8: LINQ Queries We could further shorten our code by implicitly typing filteredNames: var filteredNames = names.Where (n => n.Length >= 4); This can hinder readability, however, particularly outside of an IDE, where there are no tool tips to help. In this chapter, we avoid implicitly typing query results except when it’s mandatory (as we’ll see later, in the section “Projection Strategies” on page 345), or when a query’s type is irrelevant to an example. Most query operators accept a lambda expression as an argument. The lambda ex- pression helps guide and shape the query. In our example, the lambda expression is as follows: n => n.Length >= 4 The input argument corresponds to an input element. In this case, the input argu- ment n represents each name in the array and is of type string. The Where operator requires that the lambda expression return a bool value, which if true, indicates that the element should be included in the output sequence. Here’s its signature: public static IEnumerable Where (this IEnumerable source, Func predicate) The following query retrieves all names that contain the letter “a”: IEnumerable filteredNames = names.Where (n => n.Contains ("a")); foreach (string name in filteredNames) Console.WriteLine (name); // Harry So far, we’ve built queries using extension methods and lambda expressions. As we’ll see shortly, this strategy is highly composable in that it allows the chaining of query operators. In the book, we refer to this as fluent syntax.1 C# also provides another syntax for writing queries, called query expression syntax. Here’s our preceding query written as a query expression: IEnumerable filteredNames = from n in names where n.Contains ("a") select n; Fluent syntax and query syntax are complementary. In the following two sections, we explore each in more detail. Fluent Syntax Fluent syntax is the most flexible and fundamental. In this section, we describe how to chain query operators to form more complex queries—and show why extension 1. The term is based on Eric Evans & Martin Fowler’s work on fluent interfaces. Fluent Syntax | 321 LINQ Queries methods are important to this process. We also describe how to formulate lambda expressions for a query operator and introduce several new query operators. Chaining Query Operators In the preceding section, we showed two simple queries, each comprising a single query operator. To build more complex queries, you append additional query oper- ators to the expression, creating a chain. To illustrate, the following query extracts all strings containing the letter “a”, sorts them by length, and then converts the results to uppercase: using System; using System.Collections.Generic; using System.Linq; class LinqDemo { static void Main() { string[] names = { "Tom", "Dick", "Harry", "Mary", "Jay" }; IEnumerable query = names .Where (n => n.Contains ("a")) .OrderBy (n => n.Length) .Select (n => n.ToUpper()); foreach (string name in query) Console.WriteLine (name); } } JAY MARY HARRY The variable, n, in our example, is privately scoped to each of the lambda expressions. We can reuse n for the same reason we can reuse c in the following method: void Test() { foreach (char c in "string1") Console.Write (c); foreach (char c in "string2") Console.Write (c); foreach (char c in "string3") Console.Write (c); } Where, OrderBy, and Select are standard query operators that resolve to extension methods in the Enumerable class (if you import the System.Linq namespace). We already introduced the Where operator, which emits a filtered version of the input sequence. The OrderBy operator emits a sorted version of its input sequence; the Select method emits a sequence where each input element is transformed or pro- jected with a given lambda expression (n.ToUpper(), in this case). Data flows from 322 | Chapter 8: LINQ Queries left to right through the chain of operators, so the data is first filtered, then sorted, then projected. A query operator never alters the input sequence; instead, it returns a new sequence. This is consistent with the functional programming paradigm, from which LINQ was inspired. Here are the signatures of each of these extension methods (with the OrderBy signa- ture simplified slightly): public static IEnumerable Where (this IEnumerable source, Func predicate) public static IEnumerable OrderBy (this IEnumerable source, Func keySelector) public static IEnumerable Select (this IEnumerable source, Func selector) When query operators are chained as in this example, the output sequence of one operator is the input sequence of the next. The end result resembles a production line of conveyor belts, as illustrated in Figure 8-1. Figure 8-1. Chaining query operators We can construct the identical query progressively, as follows: // You must import the System.Linq namespace for this to compile: IEnumerable filtered = names .Where (n => n.Contains ("a")); IEnumerable sorted = filtered.OrderBy (n => n.Length); IEnumerable finalQuery = sorted .Select (n => n.ToUpper()); finalQuery is compositionally identical to the query we had constructed previously. Further, each intermediate step also comprises a valid query that we can execute: foreach (string name in filtered) Console.Write (name + "|"); // Harry|Mary|Jay| Console.WriteLine(); foreach (string name in sorted) Console.Write (name + "|"); // Jay|Mary|Harry| Fluent Syntax | 323 LINQ Queries Console.WriteLine(); foreach (string name in finalQuery) Console.Write (name + "|"); // JAY|MARY|HARRY| Why extension methods are important Instead of using extension method syntax, you can use conventional static method syntax to call the query operators. For example: IEnumerable filtered = Enumerable.Where (names, n => n.Contains ("a")); IEnumerable sorted = Enumerable.OrderBy (filtered, n => n.Length); IEnumerable finalQuery = Enumerable.Select (sorted, n => n.ToUpper()); This is, in fact, how the compiler translates extension method calls. Shunning extension methods comes at a cost, however, if you want to write a query in a single statement as we did earlier. Let’s revisit the single-statement query—first in exten- sion method syntax: IEnumerable query = names.Where (n => n.Contains ("a")) .OrderBy (n => n.Length) .Select (n => n.ToUpper()); Its natural linear shape reflects the left-to-right flow of data, as well as keeping lambda expressions alongside their query operators (infix notation). Without extension methods, the query loses its fluency: IEnumerable query = Enumerable.Select ( Enumerable.OrderBy ( Enumerable.Where ( names, n => n.Contains ("a") ), n => n.Length ), n => n.ToUpper() ); Composing Lambda Expressions In previous examples, we fed the following lambda expression to the Where operator: n => n.Contains ("a") // Input type=string, return type=bool. A lambda expression that takes a value and returns a bool is called a predicate. The purpose of the lambda expression depends on the particular query operator. With the Where operator, it indicates whether an element should be included in the output sequence. In the case of the OrderBy operator, the lambda expression maps each element in the input sequence to its sorting key. With the Select operator, the 324 | Chapter 8: LINQ Queries lambda expression determines how each element in the input sequence is trans- formed before being fed to the output sequence. A lambda expression in a query operator always works on individual elements in the input sequence—not the sequence as a whole. The query operator evaluates your lambda expression upon demand—typically once per element in the input sequence. Lambda expressions allow you to feed your own logic into the query operators. This makes the query operators versatile—as well as being simple under the hood. Here’s a complete implementation of Enumera ble.Where, exception handling aside: public static IEnumerable Where (this IEnumerable source, Func predicate) { foreach (TSource element in source) if (predicate (element)) yield return element; } Lambda expressions and Func signatures The standard query operators utilize generic Func delegates. Func is a family of gen- eral purpose generic delegates in System.Linq, defined with the following intent: The type arguments in Func appear in the same order they do in lambda expressions. Hence, Func matches a TSource=>bool lambda expression: one that accepts a TSource argument and returns a bool value. Similarly, Func matches a TSource=>TResult lambda expression. The Func delegates are listed in the section “Lambda Expressions” on page 135 in Chapter 4. Lambda expressions and element typing The standard query operators use the following generic type names: Generic type letter Meaning TSource Element type for the input sequence TResult Element type for the output sequence—if different from TSource TKey Element type for the key used in sorting, grouping, or joining TSource is determined by the input sequence. TResult and TKey are inferred from your lambda expression. Fluent Syntax | 325 LINQ Queries For example, consider the signature of the Select query operator: public static IEnumerable Select (this IEnumerable source, Func selector) Func matches a TSource=>TResult lambda expression: one that maps an input element to an output element. TSource and TResult are different types, so the lambda expression can change the type of each element. Further, the lambda expression determines the output sequence type. The following query uses Select to transform string type elements to integer type elements: string[] names = { "Tom", "Dick", "Harry", "Mary", "Jay" }; IEnumerable query = names.Select (n => n.Length); foreach (int length in query) Console.Write (length + "|"); // 3|4|5|4|3| The compiler infers the type of TResult from the return value of the lambda expres- sion. In this case, TResult is inferred to be of type int. The Where query operator is simpler and requires no type inference for the output, since input and output elements are of the same type. This makes sense because the operator merely filters elements; it does not transform them: public static IEnumerable Where (this IEnumerable source, Func predicate) Finally, consider the signature of the OrderBy operator: // Slightly simplified: public static IEnumerable OrderBy (this IEnumerable source, Func keySelector) Func maps an input element to a sorting key. TKey is inferred from your lambda expression and is separate from the input and output element types. For instance, we could choose to sort a list of names by length (int key) or alpha- betically (string key): string[] names = { "Tom", "Dick", "Harry", "Mary", "Jay" }; IEnumerable sortedByLength, sortedAlphabetically; sortedByLength = names.OrderBy (n => n.Length); // int key sortedAlphabetically = names.OrderBy (n => n); // string key You can call the query operators in Enumerable with traditional delegates that refer to methods instead of lambda expressions. This approach is effective in simplifying certain kinds of local queries—particularly with LINQ to XML—and is demon- strated in Chapter 10. It doesn’t work with IQueryable-based sequences, however (e.g., when querying a database), because the operators in Queryable require lambda expressions in order to emit expression trees. We discuss this later in the section “Interpreted Queries” on page 347. 326 | Chapter 8: LINQ Queries Natural Ordering The original ordering of elements within an input sequence is significant in LINQ. Some query operators rely on this behavior, such as Take, Skip, and Reverse. The Take operator outputs the first x elements, discarding the rest: int[] numbers = { 10, 9, 8, 7, 6 }; IEnumerable firstThree = numbers.Take (3); // { 10, 9, 8 } The Skip operator ignores the first x elements and outputs the rest: IEnumerable lastTwo = numbers.Skip (3); // { 7, 6 } Reverse does exactly as it says: IEnumerable reversed = numbers.Reverse(); // { 6, 7, 8, 9, 10 } With local queries (LINQ-to-objects), operators such as Where and Select preserve the original ordering of the input sequence (as do all other query operators, except for those that specifically change the ordering). Other Operators Not all query operators return a sequence. The element operators extract one element from the input sequence; examples are First, Last, and ElementAt: int[] numbers = { 10, 9, 8, 7, 6 }; int firstNumber = numbers.First(); // 10 int lastNumber = numbers.Last(); // 6 int secondNumber = numbers.ElementAt(1); // 9 int secondLowest = numbers.OrderBy(n=>n).Skip(1).First(); // 7 The aggregation operators return a scalar value; usually of numeric type: int count = numbers.Count(); // 5; int min = numbers.Min(); // 6; The quantifiers return a bool value: bool hasTheNumberNine = numbers.Contains (9); // true bool hasMoreThanZeroElements = numbers.Any(); // true bool hasAnOddElement = numbers.Any (n => n % 2 != 0); // true Because these operators don’t return a collection, you can’t call further query oper- ators on their result. In other words, they must appear as the last operator in a query. Some query operators accept two input sequences. Examples are Concat, which appends one sequence to another, and Union, which does the same but with dupli- cates removed: int[] seq1 = { 1, 2, 3 }; int[] seq2 = { 3, 4, 5 }; IEnumerable concat = seq1.Concat (seq2); // { 1, 2, 3, 3, 4, 5 } IEnumerable union = seq1.Union (seq2); // { 1, 2, 3, 4, 5 } The joining operators also fall into this category. Chapter 9 covers all the query operators in detail. Fluent Syntax | 327 LINQ Queries Query Expressions C# provides a syntactic shortcut for writing LINQ queries, called query expres- sions. Contrary to popular belief, a query expression is not a means of embedding SQL into C#. In fact, the design of query expressions was inspired primarily by list comprehensions from functional programming languages such as LISP and Haskell, although SQL had a cosmetic influence. In this book we refer to query expression syntax simply as “query syntax.” In the preceding section, we wrote a fluent-syntax query to extract strings containing the letter “a”, sorted by length and converted to uppercase. Here’s the same thing in query syntax: using System; using System.Collections.Generic; using System.Linq; class LinqDemo { static void Main() { string[] names = { "Tom", "Dick", "Harry", "Mary", "Jay" }; IEnumerable query = from n in names where n.Contains ("a") // Filter elements orderby n.Length // Sort elements select n.ToUpper(); // Translate each element (project) foreach (string name in query) Console.WriteLine (name); } } JAY MARY HARRY Query expressions always start with a from clause and end with either a select or group clause. The from clause declares a range variable (in this case, n), which you can think of as traversing the input sequence—rather like foreach. Figure 8-2 illus- trates the complete syntax as a railroad diagram. 328 | Chapter 8: LINQ Queries To read this diagram, start at the left and then proceed along the track as if you were a train. For instance, after the mandatory from clause, you can optionally include an orderby, where, let or join clause. After that, you can either continue with a select or group clause, or go back and include another from, orderby, where, let or join clause. The compiler processes a query expression by translating it into fluent syntax. It does this in a fairly mechanical fashion—much like it translates foreach statements into calls to GetEnumerator and MoveNext. This means that anything you can write in query syntax you can also write in fluent syntax. The compiler (initially) translates our example query into the following: IEnumerable query = names.Where (n => n.Contains ("a")) .OrderBy (n => n.Length) .Select (n => n.ToUpper()); Figure 8-2. Query syntax Query Expressions | 329 LINQ Queries The Where, OrderBy, and Select operators then resolve using the same rules that would apply if the query were written in fluent syntax. In this case, they bind to extension methods in the Enumerable class, because the System.Linq namespace is imported and names implements IEnumerable. The compiler doesn’t specif- ically favor the Enumerable class, however, when translating query expressions. You can think of the compiler as mechanically injecting the words “Where,” “OrderBy,” and “Select” into the statement, and then compiling it as though you’d typed the method names yourself. This offers flexibility in how they resolve. The operators in the database queries that we’ll write in later sections, for instance, will bind instead to extension methods in Queryable. If we remove the using System.Linq directive from our program, the query would not compile, because the Where, OrderBy, and Select methods would have nowhere to bind. Query expres- sions cannot compile unless you import a namespace (or write an instance method for every query operator!). Range Variables The identifier immediately following the from keyword syntax is called the range variable. A range variable refers to the current element in the sequence that the operation is to be performed on. In our examples, the range variable n appears in every clause in the query. And yet, the variable actually enumerates over a different sequence with each clause: from n in names // n is our range variable where n.Contains ("a") // n = directly from the array orderby n.Length // n = subsequent to being filtered select n.ToUpper() // n = subsequent to being sorted This becomes clear when we examine the compiler’s mechanical translation to fluent syntax: names.Where (n => n.Contains ("a")) // Locally scoped n .OrderBy (n => n.Length) // Locally scoped n .Select (n => n.ToUpper()) // Locally scoped n As you can see, each instance of n is scoped privately to its own lambda expression. Query expressions also let you introduce new range variables, via the following clauses: • let • into • An additional from clause • join We cover these later in this chapter in the section “Composition Strate- gies” on page 342, and also in Chapter 9, in the sections “Projec- ting” on page 377 and “Joining” on page 378. 330 | Chapter 8: LINQ Queries Query Syntax Versus SQL Syntax Query expressions look superficially like SQL, yet the two are very different. A LINQ query boils down to a C# expression, and so follows standard C# rules. For exam- ple, with LINQ, you cannot use a variable before you declare it. In SQL, you refer- ence a table alias in the SELECT clause before defining it in a FROM clause. A subquery in LINQ is just another C# expression and so requires no special syntax. Subqueries in SQL are subject to special rules. With LINQ, data logically flows from left to right through the query. With SQL, the order is less well-structured with regard to data flow. A LINQ query comprises a conveyor belt or pipeline of operators that accept and emit ordered sequences. A SQL query comprises a network of clauses that work mostly with unordered sets. Query Syntax Versus Fluent Syntax Query and fluent syntax each have advantages. Query syntax is simpler for queries that involve any of the following: • A let clause for introducing a new variable alongside the range variable • SelectMany, Join, or GroupJoin, followed by an outer range variable reference (We describe the let clause in the later section, “Composition Strate- gies” on page 342; we describe SelectMany, Join, and GroupJoin in Chapter 9.) The middle ground is queries that involve the simple use of Where, OrderBy, and Select. Either syntax works well; the choice here is largely personal. For queries that comprise a single operator, fluent syntax is shorter and less cluttered. Finally, there are many operators that have no keyword in query syntax. These require that you use fluent syntax—at least in part. This means any operator outside of the following: Where, Select, SelectMany OrderBy, ThenBy, OrderByDescending, ThenByDescending GroupBy, Join, GroupJoin Mixed Syntax Queries If a query operator has no query-syntax support, you can mix query syntax and fluent syntax. The only restriction is that each query-syntax component must be complete (i.e., start with a from clause and end with a select or group clause). Assuming this array declaration: string[] names = { "Tom", "Dick", "Harry", "Mary", "Jay" }; Query Expressions | 331 LINQ Queries the following example counts the number of names containing the letter “a”: int matches = (from n in names where n.Contains ("a") select n).Count(); // 3 The next query obtains the first name in alphabetical order: string first = (from n in names orderby n select n).First(); // Dick The mixed syntax approach is sometimes beneficial in more complex queries. With these simple examples, however, we could stick to fluent syntax throughout without penalty: int matches = names.Where (n => n.Contains ("a")).Count(); // 3 string first = names.OrderBy (n => n).First(); // Dick There are times when mixed syntax queries offer by far the highest “bang for the buck” in terms of function and simplicity. It’s important not to unilaterally favor either query or fluent syntax; otherwise, you’ll be unable to write mixed syntax quer- ies without feeling a sense of failure! Where applicable, the remainder of this chapter will show key concepts in both fluent and query syntax. Deferred Execution An important feature of most query operators is that they execute not when con- structed, but when enumerated (in other words, when MoveNext is called on its enu- merator). Consider the following query: var numbers = new List(); numbers.Add (1); IEnumerable query = numbers.Select (n => n * 10); // Build query numbers.Add (2); // Sneak in an extra element foreach (int n in query) Console.Write (n + "|"); // 10|20| The extra number that we sneaked into the list after constructing the query is included in the result, because it’s not until the foreach statement runs that any filtering or sorting takes place. This is called deferred or lazy execution and is the same as what happens with delegates: Action a = () => Console.WriteLine ("Foo"); // We've not written anything to the Console yet. Now let's run it: a(); // Deferred execution! All standard query operators provide deferred execution, with the following exceptions: • Operators that return a single element or scalar value, such as First or Count 332 | Chapter 8: LINQ Queries • The following conversion operators: ToArray, ToList, ToDictionary, ToLookup These operators cause immediate query execution because their result types have no mechanism for providing deferred execution. The Count method, for instance, returns a simple integer, which doesn’t then get enumerated. The following query is executed immediately: int matches = numbers.Where (n => n < 2).Count(); // 1 Deferred execution is important because it decouples query construction from query execution. This allows you to construct a query in several steps, as well as making database queries possible. Subqueries provide another level of indirection. Everything in a subquery is subject to deferred execution—including aggrega- tion and conversion methods. We describe this in the section “Subqueries” on page 338 later in this chapter. Reevaluation Deferred execution has another consequence: a deferred execution query is reeval- uated when you re-enumerate: var numbers = new List() { 1, 2 }; IEnumerable query = numbers.Select (n => n * 10); foreach (int n in query) Console.Write (n + "|"); // 10|20| numbers.Clear(); foreach (int n in query) Console.Write (n + "|"); // There are a couple of reasons why reevaluation is sometimes disadvantageous: • Sometimes you want to “freeze” or cache the results at a certain point in time. • Some queries are computationally intensive (or rely on querying a remote database), so you don’t want to unnecessarily repeat them. You can defeat reevaluation by calling a conversion operator, such as ToArray or ToList. ToArray copies the output of a query to an array; ToList copies to a generic List: var numbers = new List() { 1, 2 }; List timesTen = numbers .Select (n => n * 10) .ToList(); // Executes immediately into a List numbers.Clear(); Console.WriteLine (timesTen.Count); // Still 2 Deferred Execution | 333 LINQ Queries Captured Variables If your query’s lambda expressions capture outer variables, the query will honor the value of the those variables at the time the query runs: int[] numbers = { 1, 2 }; int factor = 10; IEnumerable query = numbers.Select (n => n * factor); factor = 20; foreach (int n in query) Console.Write (n + "|"); // 20|40| This can be a trap when building up a query within a for loop. For example, suppose we wanted to remove all vowels from a string. The following, although inefficient, gives the correct result: IEnumerable query = "Not what you might expect"; query = query.Where (c => c != 'a'); query = query.Where (c => c != 'e'); query = query.Where (c => c != 'i'); query = query.Where (c => c != 'o'); query = query.Where (c => c != 'u'); foreach (char c in query) Console.Write (c); // Nt wht y mght xpct Now watch what happens when we refactor this with a for loop: IEnumerable query = "Not what you might expect"; string vowels = "aeiou"; for (int i = 0; i < vowels.Length; i++) query = query.Where (c => c != vowels[i]); foreach (char c in query) Console.Write (c); An IndexOutOfRangeException is thrown upon enumerating the query, because as we saw in Chapter 4 (see “Capturing Outer Variables” on page 136), the compiler scopes the iteration variable in the for loop as if it was declared outside the loop. Hence each closure captures the same variable (i) whose value is 5 when the query is actually enumerated. To solve this, you must assign the loop variable to another variable declared inside the statement block: for (int i = 0; i < vowels.Length; i++) { char vowel = vowels[i]; query = query.Where (c => c != vowel); } This forces a fresh variable to be captured on each loop iteration. 334 | Chapter 8: LINQ Queries In C# 5.0, another way to solve the problem is to replace the for loop with a foreach loop: foreach (char vowel in vowels) query = query.Where (c => c != vowel); This works in C# 5.0 but fails in earlier versions of C# for the reasons we described in Chapter 4. How Deferred Execution Works Query operators provide deferred execution by returning decorator sequences. Unlike a traditional collection class such as an array or linked list, a decorator sequence (in general) has no backing structure of its own to store elements. Instead, it wraps another sequence that you supply at runtime, to which it maintains a per- manent dependency. Whenever you request data from a decorator, it in turn must request data from the wrapped input sequence. The query operator’s transformation constitutes the “decora- tion.” If the output sequence performed no transformation, it would be a proxy rather than a decorator. Calling Where merely constructs the decorator wrapper sequence, holding a reference to the input sequence, the lambda expression, and any other arguments supplied. The input sequence is enumerated only when the decorator is enumerated. Figure 8-3 illustrates the composition of the following query: IEnumerable lessThanTen = new int[] { 5, 12, 3 }.Where (n => n < 10); Figure 8-3. Decorator sequence When you enumerate lessThanTen, you’re, in effect, querying the array through the Where decorator. Deferred Execution | 335 LINQ Queries The good news—if you ever want to write your own query operator—is that implementing a decorator sequence is easy with a C# iterator. Here’s how you can write your own Select method: public static IEnumerable Select (this IEnumerable source, Func selector) { foreach (TSource element in source) yield return selector (element); } This method is an iterator by virtue of the yield return statement. Functionally, it’s a shortcut for the following: public static IEnumerable Select (this IEnumerable source, Func selector) { return new SelectSequence (source, selector); } where SelectSequence is a (compiler-written) class whose enumerator encapsulates the logic in the iterator method. Hence, when you call an operator such as Select or Where, you’re doing nothing more than instantiating an enumerable class that decorates the input sequence. Chaining Decorators Chaining query operators creates a layering of decorators. Consider the following query: IEnumerable query = new int[] { 5, 12, 3 }.Where (n => n < 10) .OrderBy (n => n) .Select (n => n * 10); Each query operator instantiates a new decorator that wraps the previous sequence (rather like a Russian nesting doll). The object model of this query is illustrated in Figure 8-4. Note that this object model is fully constructed prior to any enumeration. When you enumerate query, you’re querying the original array, transformed through a layering or chain of decorators. Adding ToList onto the end of this query would cause the pre- ceding operators to execute right away, collapsing the whole object model into a single list. Figure 8-5 shows the same object composition in UML syntax. Select’s decorator references the OrderBy decorator, which references Where’s decorator, which refer- ences the array. A feature of deferred execution is that you build the identical object model if you compose the query progressively: 336 | Chapter 8: LINQ Queries Figure 8-5. UML decorator composition IEnumerable source = new int[] { 5, 12, 3 }, filtered = source .Where (n => n < 10), sorted = filtered .OrderBy (n => n), query = sorted .Select (n => n * 10); How Queries Are Executed Here are the results of enumerating the preceding query: foreach (int n in query) Console.WriteLine (n); Figure 8-4. Layered decorator sequences Deferred Execution | 337 LINQ Queries 30 50 Behind the scenes, the foreach calls GetEnumerator on Select’s decorator (the last or outermost operator), which kicks everything off. The result is a chain of enumerators that structurally mirrors the chain of decorator sequences. Figure 8-6 illustrates the flow of execution as enumeration proceeds. Figure 8-6. Execution of a local query In the first section of this chapter, we depicted a query as a production line of con- veyor belts. Extending this analogy, we can say a LINQ query is a lazy production line, where the conveyor belts roll elements only upon demand. Constructing a query constructs a production line—with everything in place—but with nothing rolling. Then when the consumer requests an element (enumerates over the query), the rightmost conveyor belt activates; this in turn triggers the others to roll—as and when input sequence elements are needed. LINQ follows a demand-driven pull model, rather than a supply-driven push model. This is important—as we’ll see later —in allowing LINQ to scale to querying SQL databases. Subqueries A subquery is a query contained within another query’s lambda expression. The following example uses a subquery to sort musicians by their last name: string[] musos = { "David Gilmour", "Roger Waters", "Rick Wright", "Nick Mason" }; IEnumerable query = musos.OrderBy (m => m.Split().Last()); 338 | Chapter 8: LINQ Queries m.Split converts each string into a collection of words, upon which we then call the Last query operator. m.Split().Last is the subquery; query references the outer query. Subqueries are permitted because you can put any valid C# expression on the right- hand side of a lambda. A subquery is simply another C# expression. This means that the rules for subqueries are a consequence of the rules for lambda expressions (and the behavior of query operators in general). The term subquery, in the general sense, has a broader meaning. For the purpose of describing LINQ, we use the term only for a query referenced from within the lambda expression of another query. In a query expression, a subquery amounts to a query referenced from an expression in any clause except the from clause. A subquery is privately scoped to the enclosing expression and is able to reference parameters in the outer lambda expression (or range variables in a query expression). m.Split().Last is a very simple subquery. The next query retrieves all strings in an array whose length matches that of the shortest string: string[] names = { "Tom", "Dick", "Harry", "Mary", "Jay" }; IEnumerable outerQuery = names .Where (n => n.Length == names.OrderBy (n2 => n2.Length) .Select (n2 => n2.Length).First()); Tom, Jay Here’s the same thing as a query expression: IEnumerable outerQuery = from n in names where n.Length == (from n2 in names orderby n2.Length select n2.Length).First() select n; Because the outer range variable (n) is in scope for a subquery, we cannot reuse n as the subquery’s range variable. A subquery is executed whenever the enclosing lambda expression is evaluated. This means a subquery is executed upon demand, at the discretion of the outer query. You could say that execution proceeds from the outside in. Local queries follow this model literally; interpreted queries (e.g., database queries) follow this model conceptually. The subquery executes as and when required, to feed the outer query. In our exam- ple, the subquery (the top conveyor belt in Figure 8-7) executes once for every outer loop iteration. This is illustrated in Figure 8-7 and Figure 8-8. Subqueries | 339 LINQ Queries We can express our preceding subquery more succinctly as follows: IEnumerable query = from n in names where n.Length == names.OrderBy (n2 => n2.Length).First().Length select n; With the Min aggregation function, we can simplify the query further: IEnumerable query = from n in names where n.Length == names.Min (n2 => n2.Length) select n; In the later section “Interpreted Queries” on page 347, we’ll describe how remote sources such as SQL tables can be queried. Our example makes an ideal database query, because it would be processed as a unit, requiring only one round trip to the database server. This query, however, is inefficient for a local collection because the subquery is recalculated on each outer loop iteration. We can avoid this inefficiency by running the subquery separately (so that it’s no longer a subquery): int shortest = names.Min (n => n.Length); IEnumerable query = from n in names where n.Length == shortest select n; Factoring out subqueries in this manner is nearly always desir- able when querying local collections. An exception is when the subquery is correlated, meaning that it references the outer range variable. We explore correlated subqueries in “Projec- ting” on page 377 in Chapter 9. Figure 8-7. Subquery composition 340 | Chapter 8: LINQ Queries Subqueries and Deferred Execution An element or aggregation operator such as First or Count in a subquery doesn’t force the outer query into immediate execution—deferred execution still holds for the outer query. This is because subqueries are called indirectly—through a delegate in the case of a local query, or through an expression tree in the case of an interpreted query. An interesting case arises when you include a subquery within a Select expression. In the case of a local query, you’re actually projecting a sequence of queries—each itself subject to deferred execution. The effect is generally transparent, and it serves to further improve efficiency. We revisit Select subqueries in some detail in Chapter 9. Figure 8-8. UML subquery composition Subqueries | 341 LINQ Queries Composition Strategies In this section, we describe three strategies for building more complex queries: • Progressive query construction • Using the into keyword • Wrapping queries All are chaining strategies and produce identical runtime queries. Progressive Query Building At the start of the chapter, we demonstrated how you could build a fluent query progressively: var filtered = names .Where (n => n.Contains ("a")); var sorted = filtered .OrderBy (n => n); var query = sorted .Select (n => n.ToUpper()); Because each of the participating query operators returns a decorator sequence, the resultant query is the same chain or layering of decorators that you would get from a single-expression query. There are a couple of potential benefits, however, to building queries progressively: • It can make queries easier to write. • You can add query operators conditionally. For example: if (includeFilter) query = query.Where (...) This is more efficient than: query = query.Where (n => !includeFilter || ) because it avoids adding an extra query operator if includeFilter is false. A progressive approach is often useful in query comprehensions. To illustrate, imag- ine we want to remove all vowels from a list of names, and then present in alpha- betical order those whose length is still more than two characters. In fluent syntax, we could write this query as a single expression—by projecting before we filter: IEnumerable query = names .Select (n => n.Replace ("a", "").Replace ("e", "").Replace ("i", "") .Replace ("o", "").Replace ("u", "")) .Where (n => n.Length > 2) .OrderBy (n => n); RESULT: { "Dck", "Hrry", "Mry" } 342 | Chapter 8: LINQ Queries Rather than calling string’s Replace method five times, we could remove vowels from a string more efficiently with a reg- ular expression: n => Regex.Replace (n, "[aeiou]", "") string’s Replace method has the advantage, though, of also working in database queries. Translating this directly into a query expression is troublesome because the select clause must come after the where and orderby clauses. And if we rearrange the query so as to project last, the result would be different: IEnumerable query = from n in names where n.Length > 2 orderby n select n.Replace ("a", "").Replace ("e", "").Replace ("i", "") .Replace ("o", "").Replace ("u", ""); RESULT: { "Dck", "Hrry", "Jy", "Mry", "Tm" } Fortunately, there are a number of ways to get the original result in query syntax. The first is by querying progressively: IEnumerable query = from n in names select n.Replace ("a", "").Replace ("e", "").Replace ("i", "") .Replace ("o", "").Replace ("u", ""); query = from n in query where n.Length > 2 orderby n select n; RESULT: { "Dck", "Hrry", "Mry" } The into Keyword The into keyword is interpreted in two very different ways by query expressions, depending on context. The meaning we’re describing now is for signaling query continuation (the other is for signaling a GroupJoin). The into keyword lets you “continue” a query after a projection and is a shortcut for progressively querying. With into, we can rewrite the preceding query as: IEnumerable query = from n in names select n.Replace ("a", "").Replace ("e", "").Replace ("i", "") .Replace ("o", "").Replace ("u", "") into noVowel where noVowel.Length > 2 orderby noVowel select noVowel; Composition Strategies | 343 LINQ Queries The only place you can use into is after a select or group clause. into “restarts” a query, allowing you to introduce fresh where, orderby, and select clauses. Although it’s easiest to think of into as restarting a query from the perspective of a query expression, it’s all one query when translated to its final fluent form. Hence, there’s no intrinsic performance hit with into. Nor do you lose any points for its use! The equivalent of into in fluent syntax is simply a longer chain of operators. Scoping rules All range variables are out of scope following an into keyword. The following will not compile: var query = from n1 in names select n1.ToUpper() into n2 // Only n2 is visible from here on. where n1.Contains ("x") // Illegal: n1 is not in scope. select n2; To see why, consider how this maps to fluent syntax: var query = names .Select (n1 => n1.ToUpper()) .Where (n2 => n1.Contains ("x")); // Error: n1 no longer in scope The original name (n1) is lost by the time the Where filter runs. Where’s input sequence contains only uppercase names, so it cannot filter based on n1. Wrapping Queries A query built progressively can be formulated into a single statement by wrapping one query around another. In general terms: var tempQuery = tempQueryExpr var finalQuery = from ... in tempQuery ... can be reformulated as: var finalQuery = from ... in (tempQueryExpr) Wrapping is semantically identical to progressive query building or using the into keyword (without the intermediate variable). The end result in all cases is a linear chain of query operators. For example, consider the following query: IEnumerable query = from n in names select n.Replace ("a", "").Replace ("e", "").Replace ("i", "") .Replace ("o", "").Replace ("u", ""); query = from n in query where n.Length > 2 orderby n select n; 344 | Chapter 8: LINQ Queries Reformulated in wrapped form, it’s the following: IEnumerable query = from n1 in ( from n2 in names select n2.Replace ("a", "").Replace ("e", "").Replace ("i", "") .Replace ("o", "").Replace ("u", "") ) where n1.Length > 2 orderby n1 select n1; When converted to fluent syntax, the result is the same linear chain of operators as in previous examples. IEnumerable query = names .Select (n => n.Replace ("a", "").Replace ("e", "").Replace ("i", "") .Replace ("o", "").Replace ("u", "")) .Where (n => n.Length > 2) .OrderBy (n => n); (The compiler does not emit the final .Select (n => n) because it’s redundant.) Wrapped queries can be confusing because they resemble the subqueries we wrote earlier. Both have the concept of an inner and outer query. When converted to fluent syntax, however, you can see that wrapping is simply a strategy for sequentially chaining operators. The end result bears no resemblance to a subquery, which embeds an inner query within the lambda expression of another. Returning to a previous analogy: when wrapping, the “inner” query amounts to the preceding conveyor belts. In contrast, a subquery rides above a conveyor belt and is activated upon demand through the conveyor belt’s lambda worker (as illustrated in Figure 8-7). Projection Strategies Object Initializers So far, all our select clauses have projected scalar element types. With C# object initializers, you can project into more complex types. For example, suppose, as a first step in a query, we want to strip vowels from a list of names while still retaining the original versions alongside, for the benefit of subsequent queries. We can write the following class to assist: class TempProjectionItem { public string Original; // Original name public string Vowelless; // Vowel-stripped name } and then project into it with object initializers: string[] names = { "Tom", "Dick", "Harry", "Mary", "Jay" }; IEnumerable temp = from n in names Projection Strategies | 345 LINQ Queries select new TempProjectionItem { Original = n, Vowelless = n.Replace ("a", "").Replace ("e", "").Replace ("i", "") .Replace ("o", "").Replace ("u", "") }; The result is of type IEnumerable, which we can subsequently query: IEnumerable query = from item in temp where item.Vowelless.Length > 2 select item.Original; Dick Harry Mary Anonymous Types Anonymous types allow you to structure your intermediate results without writing special classes. We can eliminate the TempProjectionItem class in our previous example with anonymous types: var intermediate = from n in names select new { Original = n, Vowelless = n.Replace ("a", "").Replace ("e", "").Replace ("i", "") .Replace ("o", "").Replace ("u", "") }; IEnumerable query = from item in intermediate where item.Vowelless.Length > 2 select item.Original; This gives the same result as the previous example, but without needing to write a one-off class. The compiler does the job instead, writing a temporary class with fields that match the structure of our projection. This means, however, that the intermedi ate query has the following type: IEnumerable The only way we can declare a variable of this type is with the var keyword. In this case, var is more than just a clutter reduction device; it’s a necessity. We can write the whole query more succinctly with the into keyword: var query = from n in names select new { Original = n, Vowelless = n.Replace ("a", "").Replace ("e", "").Replace ("i", "") .Replace ("o", "").Replace ("u", "") } into temp 346 | Chapter 8: LINQ Queries where temp.Vowelless.Length > 2 select temp.Original; Query expressions provide a shortcut for writing this kind of query: the let keyword. The let Keyword The let keyword introduces a new variable alongside the range variable. With let, we can write a query extracting strings whose length, excluding vowels, exceeds two characters, as follows: string[] names = { "Tom", "Dick", "Harry", "Mary", "Jay" }; IEnumerable query = from n in names let vowelless = n.Replace ("a", "").Replace ("e", "").Replace ("i", "") .Replace ("o", "").Replace ("u", "") where vowelless.Length > 2 orderby vowelless select n; // Thanks to let, n is still in scope. The compiler resolves a let clause by projecting into a temporary anonymous type that contains both the range variable and the new expression variable. In other words, the compiler translates this query into the preceding example. let accomplishes two things: • It projects new elements alongside existing elements. • It allows an expression to be used repeatedly in a query without being rewritten. The let approach is particularly advantageous in this example, because it allows the select clause to project either the original name (n) or its vowel-removed version (vowelless). You can have any number of let statements, before or after a where statement (see Figure 8-2). A let statement can reference variables introduced in earlier let state- ments (subject to the boundaries imposed by an into clause). let reprojects all existing variables transparently. A let expression need not evaluate to a scalar type: sometimes it’s useful to have it evaluate to a subsequence, for instance. Interpreted Queries LINQ provides two parallel architectures: local queries for local object collections, and interpreted queries for remote data sources. So far, we’ve examined the archi- tecture of local queries, which operate over collections implementing IEnumera ble. Local queries resolve to query operators in the Enumerable class, which in turn resolve to chains of decorator sequences. The delegates that they accept— whether expressed in query syntax, fluent syntax, or traditional delegates—are fully local to Intermediate Language (IL) code, just like any other C# method. Interpreted Queries | 347 LINQ Queries By contrast, interpreted queries are descriptive. They operate over sequences that implement IQueryable, and they resolve to the query operators in the Queryable class, which emit expression trees that are interpreted at runtime. The query operators in Enumerable can actually work with IQueryable sequences. The difficulty is that the resultant queries always execute locally on the client—this is why a sec- ond set of query operators is provided in the Queryable class. There are two IQueryable implementations in the .NET Framework: • LINQ to SQL • Entity Framework (EF) These LINQ-to-db technologies are very similar in their LINQ support: the LINQ- to-db queries in this book will work with both LINQ to SQL and EF unless otherwise specified. It’s also possible to generate an IQueryable wrapper around an ordinary enu- merable collection by calling the AsQueryable method. We describe AsQueryable in the section “Building Query Expressions” on page 368 later in this chapter. In this section, we’ll use LINQ to SQL to illustrate interpreted query architecture because LINQ to SQL lets us query without having to first write an Entity Data Model. The queries that we write, however, work equally well with Entity Frame- work (and also many third-party products). IQueryable is an extension of IEnumerable with additional methods for constructing expression trees. Most of the time you can ignore the details of these methods; they’re called indirectly by the Framework. The “Building Query Expres- sions” on page 368 section covers IQueryable in more detail. Suppose we create a simple customer table in SQL Server and populate it with a few names using the following SQL script: create table Customer ( ID int not null primary key, Name varchar(30) ) insert Customer values (1, 'Tom') insert Customer values (2, 'Dick') insert Customer values (3, 'Harry') insert Customer values (4, 'Mary') insert Customer values (5, 'Jay') 348 | Chapter 8: LINQ Queries With this table in place, we can write an interpreted LINQ query in C# to retrieve customers whose name contains the letter “a” as follows: using System; using System.Linq; using System.Data.Linq; // in System.Data.Linq.dll using System.Data.Linq.Mapping; [Table] public class Customer { [Column(IsPrimaryKey=true)] public int ID; [Column] public string Name; } class Test { static void Main() { DataContext dataContext = new DataContext ("connection string"); Table customers = dataContext.GetTable (); IQueryable query = from c in customers where c.Name.Contains ("a") orderby c.Name.Length select c.Name.ToUpper(); foreach (string name in query) Console.WriteLine (name); } } LINQ to SQL translates this query into the following SQL: SELECT UPPER([t0].[Name]) AS [value] FROM [Customer] AS [t0] WHERE [t0].[Name] LIKE @p0 ORDER BY LEN([t0].[Name]) with the following end result: JAY MARY HARRY How Interpreted Queries Work Let’s examine how the preceding query is processed. First, the compiler converts query syntax to fluent syntax. This is done exactly as with local queries: IQueryable query = customers.Where (n => n.Name.Contains ("a")) .OrderBy (n => n.Name.Length) .Select (n => n.Name.ToUpper()); Next, the compiler resolves the query operator methods. Here’s where local and interpreted queries differ—interpreted queries resolve to query operators in the Queryable class instead of the Enumerable class. Interpreted Queries | 349 LINQ Queries To see why, we need to look at the customers variable, the source upon which the whole query builds. customers is of type Table, which implements IQueryable (a subtype of IEnumerable). This means the compiler has a choice in resolving Where: it could call the extension method in Enumerable or the following extension method in Queryable: public static IQueryable Where (this IQueryable source, Expression > predicate) The compiler chooses Queryable.Where because its signature is a more specific match. Queryable.Where accepts a predicate wrapped in an Expression type. This instructs the compiler to translate the supplied lambda expression—in other words, n=>n.Name.Contains("a")—to an expression tree rather than a compiled del- egate. An expression tree is an object model based on the types in Sys tem.Linq.Expressions that can be inspected at runtime (so that LINQ to SQL or EF can later translate it to a SQL statement). Because Queryable.Where also returns IQueryable, the same process follows with the OrderBy and Select operators. The end result is illustrated in Figure 8-9. In the shaded box, there is an expression tree describing the entire query, which can be traversed at runtime. Figure 8-9. Interpreted query composition Execution Interpreted queries follow a deferred execution model—just like local queries. This means that the SQL statement is not generated until you start enumerating the query. 350 | Chapter 8: LINQ Queries Further, enumerating the same query twice results in the database being queried twice. Under the covers, interpreted queries differ from local queries in how they execute. When you enumerate over an interpreted query, the outermost sequence runs a program that traverses the entire expression tree, processing it as a unit. In our example, LINQ to SQL translates the expression tree to a SQL statement, which it then executes, yielding the results as a sequence. To work, LINQ to SQL needs some clues as to the schema of the database. The Table and Column attributes that we applied to the Customer class serve just this function. The section “LINQ to SQL and Entity Framework” on page 354, later in this chap- ter, describes these attributes in more detail. Entity Framework is similar except that it also requires an Entity Data Model (EDM)—an XML file describing the mapping between database and entities. We said previously that a LINQ query is like a production line. When you enumerate an IQueryable conveyor belt, though, it doesn’t start up the whole production line, like with a local query. Instead, just the IQueryable belt starts up, with a special enumerator that calls upon a production manager. The manager reviews the entire production line—which consists not of compiled code, but of dummies (method call expressions) with instructions pasted to their foreheads (expression trees). The man- ager then traverses all the expressions, in this case transcribing them to a single piece of paper (a SQL statement), which it then executes, feeding the results back to the consumer. Only one belt turns; the rest of the production line is a network of empty shells, existing just to describe what has to be done. This has some practical implications. For instance, with local queries, you can write your own query methods (fairly easily, with iterators) and then use them to supple- ment the predefined set. With remote queries, this is difficult, and even undesirable. If you wrote a MyWhere extension method accepting IQueryable, it would be like putting your own dummy into the production line. The production manager wouldn’t know what to do with your dummy. Even if you intervened at this stage, your solution would be hard-wired to a particular provider, such as LINQ to SQL, and would not work with other IQueryable implementations. Part of the benefit of having a standard set of methods in Queryable is that they define a standard vocabu- lary for querying any remote collection. As soon as you try to extend the vocabulary, you’re no longer interoperable. Another consequence of this model is that an IQueryable provider may be unable to cope with some queries—even if you stick to the standard methods. LINQ to SQL and EF are both limited by the capabilities of the database server; some LINQ queries have no SQL translation. If you’re familiar with SQL, you’ll have a good intuition for what these are, although at times you have to experiment to see what causes a runtime error; it can be surprising what does work! Interpreted Queries | 351 LINQ Queries Combining Interpreted and Local Queries A query can include both interpreted and local operators. A typical pattern is to have the local operators on the outside and the interpreted components on the inside; in other words, the interpreted queries feed the local queries. This pattern works well with LINQ-to-database queries. For instance, suppose we write a custom extension method to pair up strings in a collection: public static IEnumerable Pair (this IEnumerable source) { string firstHalf = null; foreach (string element in source) if (firstHalf == null) firstHalf = element; else { yield return firstHalf + ", " + element; firstHalf = null; } } We can use this extension method in a query that mixes LINQ to SQL and local operators: DataContext dataContext = new DataContext ("connection string"); Table customers = dataContext.GetTable (); IEnumerable q = customers .Select (c => c.Name.ToUpper()) .OrderBy (n => n) .Pair() // Local from this point on. .Select ((n, i) => "Pair " + i.ToString() + " = " + n); foreach (string element in q) Console.WriteLine (element); Pair 0 = HARRY, MARY Pair 1 = TOM, DICK Because customers is of a type implementing IQueryable, the Select operator resolves to Queryable.Select. This returns an output sequence also of type IQuerya ble. But the next query operator, Pair, has no overload accepting IQueryable —only the less specific IEnumerable. So, it resolves to our local Pair method— wrapping the interpreted query in a local query. Pair also emits IEnumerable, so OrderBy wraps another local operator. On the LINQ to SQL side, the resulting SQL statement is equivalent to: SELECT UPPER (Name) FROM Customer ORDER BY UPPER (Name) The remaining work is done locally. In effect, we end up with a local query (on the outside), whose source is an interpreted query (the inside). 352 | Chapter 8: LINQ Queries AsEnumerable Enumerable.AsEnumerable is the simplest of all query operators. Here’s its complete definition: public static IEnumerable AsEnumerable (this IEnumerable source) { return source; } Its purpose is to cast an IQueryable sequence to IEnumerable, forcing subse- quent query operators to bind to Enumerable operators instead of Queryable opera- tors. This causes the remainder of the query to execute locally. To illustrate, suppose we had a MedicalArticles table in SQL Server and wanted to use LINQ to SQL or EF to retrieve all articles on influenza whose abstract contained less than 100 words. For the latter predicate, we need a regular expression: Regex wordCounter = new Regex (@"\b(\w|[-'])+\b"); var query = dataContext.MedicalArticles .Where (article => article.Topic == "influenza" && wordCounter.Matches (article.Abstract).Count < 100); The problem is that SQL Server doesn’t support regular expressions, so the LINQ- to-db providers will throw an exception, complaining that the query cannot be translated to SQL. We can solve this by querying in two steps: first retrieving all articles on influenza through a LINQ to SQL query, and then filtering locally for abstracts of less than 100 words: Regex wordCounter = new Regex (@"\b(\w|[-'])+\b"); IEnumerable sqlQuery = dataContext.MedicalArticles .Where (article => article.Topic == "influenza"); IEnumerable localQuery = sqlQuery .Where (article => wordCounter.Matches (article.Abstract).Count < 100); Because sqlQuery is of type IEnumerable, the second query binds to the local query operators, forcing that part of the filtering to run on the client. With AsEnumerable, we can do the same in a single query: Regex wordCounter = new Regex (@"\b(\w|[-'])+\b"); var query = dataContext.MedicalArticles .Where (article => article.Topic == "influenza") .AsEnumerable() .Where (article => wordCounter.Matches (article.Abstract).Count < 100); An alternative to calling AsEnumerable is to call ToArray or ToList. The advantage of AsEnumerable is that it doesn’t force immediate query execution, nor does it create any storage structure. Interpreted Queries | 353 LINQ Queries Moving query processing from the database server to the client can hurt performance, especially if it means retrieving more rows. A more efficient (though more complex) way to solve our example would be to use SQL CLR integration to expose a function on the database that implemented the regular expression. We demonstrate combined interpreted and local queries further in Chapter 10. LINQ to SQL and Entity Framework Throughout this and the following chapter, we use LINQ to SQL (L2S) and Entity Framework (EF) to demonstrate interpreted queries. We’ll now examine the key features of these technologies. If you’re already familiar with L2S, take an advance look at Table 8-1 (at end of this section) for a summary of the API dif- ferences with respect to querying. LINQ to SQL Versus Entity Framework Both LINQ to SQL and Entity Framework are LINQ-enabled object-relational mappers. The essential difference is that EF allows for stronger decoupling between the database schema and the classes that you query. Instead of querying classes that closely represent the database schema, you query a higher-level abstraction described by an Entity Data Model. This offers extra flexibility, but incurs a cost in both performance and simplicity. L2S was written by the C# team and was released with Framework 3.5; EF was written by the ADO.NET team and was released later as part of Service Pack 1. L2S has since been taken over by the ADO.NET team. This has resulted in the product receiving only minor subsequent improvements, with the team concen- trating more on EF. EF has improved considerably in later versions, although each technology still has unique strengths. L2S’s strengths are ease of use, simplicity, performance, and the quality of its SQL translations. EF’s strength is its flexibility in creating sophisti- cated mappings between the database and entity classes. EF also allows for data- bases other than SQL Server via a provider model (L2S also features a provider model, but this was made internal to encourage third parties to focus on EF instead). A welcome enhancement in EF 4 is that it supports (almost) the same querying functionality as L2S. This means that the LINQ-to-db queries that we demonstrate in this book work with either technology. Further, it makes L2S excellent for learning how to query databases in LINQ—because it keeps the object-relational side of things simple while you learn querying principles that also work with EF. 354 | Chapter 8: LINQ Queries LINQ to SQL Entity Classes L2S allows you to use any class to represent data, as long as you decorate it with appropriate attributes. Here’s a simple example: [Table] public class Customer { [Column(IsPrimaryKey=true)] public int ID; [Column] public string Name; } The [Table] attribute, in the System.Data.Linq.Mapping namespace, tells L2S that an object of this type represents a row in a database table. By default, it assumes the table name matches the class name; if this is not the case, you can specify the table name as follows: [Table (Name="Customers")] A class decorated with the [Table] attribute is called an entity in L2S. To be useful, its structure must closely—or exactly—match that of a database table, making it a low-level construct. The [Column] attribute flags a field or property that maps to a column in a table. If the column name differs from the field or property name, you can specify the column name as follows: [Column (Name="FullName")] public string Name; The IsPrimaryKey property in the [Column] attribute indicates that the column par- takes in the table’s primary key and is required for maintaining object identity, as well as allowing updates to be written back to the database. Instead of defining public fields, you can define public properties in conjunction with private fields. This allows you to write validation logic into the property ac- cessors. If you take this route, you can optionally instruct L2S to bypass your prop- erty accessors and write to the field directly when populating from the database: string _name; [Column (Storage="_name")] public string Name { get { return _name; } set { _name = value; } } Column(Storage="_name") tells L2S to write directly to the _name field (rather than the Name property) when populating the entity. L2S’s use of reflection allows the field to be private—as in this example. You can generate entity classes automatically from a database using either Visual Studio (add a new “LINQ to SQL Classes” project item) or with the SqlMetal command-line tool. LINQ to SQL and Entity Framework | 355 LINQ Queries Entity Framework Entity Classes As with L2S, EF lets you use any class to represent data (although you have to im- plement special interfaces if you want functionality such as navigation properties). The following entity class, for instance, represents a customer that ultimately maps to a customer table in the database: // You'll need to reference System.Data.Entity.dll [EdmEntityType (NamespaceName = "NutshellModel", Name = "Customer")] public partial class Customer { [EdmScalarPropertyAttribute (EntityKeyProperty = true, IsNullable = false)] public int ID { get; set; } [EdmScalarProperty (EntityKeyProperty = false, IsNullable = false)] public string Name { get; set; } } Unlike with L2S, however, a class such as this is not enough on its own. Remember that with EF, you’re not querying the database directly—you’re querying a higher- level model called the Entity Data Model (EDM). There needs to be some way to describe the EDM, and this is most commonly done via an XML file with an .edmx extension, which contains three parts: • The conceptual model, which describes the EDM in isolation of the database • The store model, which describes the database schema • The mapping, which describes how the conceptual model maps to the store The easiest way to create an .edmx file is to add an “ADO.NET Entity Data Model” project item in Visual Studio and then follow the wizard for generating entities from a database. This creates not only the .edmx file, but the entity classes as well. The entity classes in EF map to the conceptual model. The types that support querying and updating the conceptual model are collectively called Object Services. The designer assumes that you initially want a simple 1:1 mapping between tables and entities. You can enrich this, however, by tweaking the EDM either with the designer or by editing the underlying .edmx file that it creates for you. Here are some of the things you can do: • Map several tables into one entity. • Map one table into several entities. • Map inherited types to tables using the three standard kinds of strategies pop- ular in the ORM world. 356 | Chapter 8: LINQ Queries The three kinds of inheritance strategies are: Table per hierarchy A single table maps to a whole class hierarchy. The table contains a discrimi- nator column to indicate which type each row should map to. Table per type A single table maps to one type, meaning that an inherited type maps to several tables. EF generates a SQL JOIN when you query an entity, to merge all its base types together. Table per concrete type A separate table maps to each concrete type. This means that a base type maps to several tables and EF generates a SQL UNION when you query for entities of a base type. (In contrast, L2S supports only table per hierarchy.) The EDM is complex: a thorough discussion can fill hundreds of pages! A good book that describes this in detail is Julia Ler- man’s Programming Entity Framework. EF also lets you query through the EDM without LINQ—using a textual language called Entity SQL (ESQL). This can be useful for dynamically constructed queries. DataContext and ObjectContext Once you’ve defined entity classes (and an EDM in the case of EF) you can start querying. The first step is to instantiate a DataContext (L2S) or ObjectContext (EF), specifying a connection string: var l2sContext = new DataContext ("database connection string"); var efContext = new ObjectContext ("entity connection string"); Instantiating a DataContext/ObjectContext directly is a low-level approach and is good for demonstrating how the classes work. More typically, though, you instantiate a typed context (a sub- classed version of these classes), a process we’ll describe shortly. With L2S, you pass in the database connection string; with EF, you must pass an entity connection string, which incorporates the database connection string plus information on how to find the EDM. (If you’ve created an EDM in Visual Studio, you can find the entity connection string for your EDM in the app.config file.) You can then obtain a queryable object calling GetTable (L2S) or CreateObjectSet (EF). The following example uses the Customer class that we defined earlier: var context = new DataContext ("database connection string"); Table customers = context.GetTable (); LINQ to SQL and Entity Framework | 357 LINQ Queries Console.WriteLine (customers.Count()); // # of rows in table. Customer cust = customers.Single (c => c.ID == 2); // Retrieves Customer // with ID of 2. Here’s the same thing with EF: var context = new ObjectContext ("entity connection string"); context.DefaultContainerName = "NutshellEntities"; ObjectSet customers = context.CreateObjectSet(); Console.WriteLine (customers.Count()); // # of rows in table. Customer cust = customers.Single (c => c.ID == 2); // Retrieves Customer // with ID of 2. The Single operator is ideal for retrieving a row by primary key. Unlike First, it throws an exception if more than one element is returned. A DataContext/ObjectContext object does two things. First, it acts as a factory for generating objects that you can query. Second, it keeps track of any changes that you make to your entities so that you can write them back. We can continue our previous example to update a customer with L2S as follows: Customer cust = customers.OrderBy (c => c.Name).First(); cust.Name = "Updated Name"; context.SubmitChanges(); With EF, the only difference is that you call SaveChanges instead: Customer cust = customers.OrderBy (c => c.Name).First(); cust.Name = "Updated Name"; context.SaveChanges(); Typed contexts Having to call GetTable() or CreateObjectSet() all the time is awkward. A better approach is to subclass DataContext/ObjectContext for a partic- ular database, adding properties that do this for each entity. This is called a typed context: class NutshellContext : DataContext // For LINQ to SQL { public Table Customers { get { return GetTable(); } } // ... and so on, for each table in the database } Here’s the same thing for EF: class NutshellContext : ObjectContext // For Entity Framework { 358 | Chapter 8: LINQ Queries public ObjectSet Customers { get { return CreateObjectSet(); } } // ... and so on, for each entity in the conceptual model } You can then simply do this: var context = new NutshellContext ("connection string"); Console.WriteLine (context.Customers.Count()); If you use Visual Studio to create a “LINQ to SQL Classes” or “ADO.NET Entity Data Model” project item, it builds a typed context for you automatically. The designers can also do additional work such as pluralizing identifiers—in this exam- ple, it’s context.Customers and not context.Customer, even though the SQL table and entity class are both called Customer. Disposing DataContext/ObjectContext Although DataContext/ObjectContext implement IDisposable, you can (in general) get away without disposing instances. Disposing forces the context’s connection to dispose—but this is usually unnecessary because L2S and EF close connections automatically whenever you finish retrieving results from a query. Disposing a context can actually be problematic because of lazy evaluation. Con- sider the following: IQueryable GetCustomers (string prefix) { using (var dc = new NutshellContext ("connection string")) return dc.GetTable() .Where (c => c.Name.StartsWith (prefix)); } ... foreach (Customer c in GetCustomers ("a")) Console.WriteLine (c.Name); This will fail because the query is evaluated when we enumerate it—which is after disposing its DataContext. There are some caveats, though, on not disposing contexts: • It relies on the connection object releasing all unmanaged resources on the Close method. While this holds true with SqlConnection, it’s theoretically possible for a third-party connection to keep resources open if you call Close but not Dispose (though this would arguably violate the contract defined by IDbConnection.Close). • If you manually call GetEnumerator on a query (instead of using foreach) and then fail to either dispose the enumerator or consume the sequence, the con- nection will remain open. Disposing the DataContext/ObjectContext provides a backup in such scenarios. • Some people feel that it’s tidier to dispose contexts (and all objects that implement IDisposable). LINQ to SQL and Entity Framework | 359 LINQ Queries If you want to explicitly dispose contexts, you must pass a DataContext/Object Context instance into methods such as GetCustomers to avoid the problem described. Object tracking A DataContext/ObjectContext instance keeps track of all the entities it instantiates, so it can feed the same ones back to you whenever you request the same rows in a table. In other words, a context in its lifetime will never emit two separate entities that refer to the same row in a table (where a row is identified by primary key). You can disable this behavior in L2S by setting ObjectTrackingEnabled to false on the DataContext object. In EF, you can disable change tracking on a per-type basis: context.Customers.MergeOption = MergeOption.NoTracking; Disabling object tracking also prevents you from submitting updates to the data. To illustrate object tracking, suppose the customer whose name is alphabetically first also has the lowest ID. In the following example, a and b will reference the same object: var context = new NutshellContext ("connection string"); Customer a = context.Customers.OrderBy (c => c.Name).First(); Customer b = context.Customers.OrderBy (c => c.ID).First(); This has a couple of interesting consequences. First, consider what happens when L2S or EF encounters the second query. It starts by querying the database—and obtaining a single row. It then reads the primary key of this row and performs a lookup in the context’s entity cache. Seeing a match, it returns the existing object without updating any values. So, if another user had just updated that customer’s Name in the database, the new value would be ignored. This is essential for avoiding unexpected side effects (the Customer object could be in use elsewhere) and also for managing concurrency. If you had altered properties on the Customer object and not yet called SubmitChanges/SaveChanges, you wouldn’t want your properties automat- ically overwritten. To get fresh information from the database, you must either instantiate a new context or call its Refresh method, passing in the entity or entities that you want refreshed. The second consequence is that you cannot explicitly project into an entity type— to select a subset of the row’s columns—without causing trouble. For example, if 360 | Chapter 8: LINQ Queries you want to retrieve only a customer’s name, any of the following approaches is valid: customers.Select (c => c.Name); customers.Select (c => new { Name = c.Name } ); customers.Select (c => new MyCustomType { Name = c.Name } ); The following, however, is not: customers.Select (c => new Customer { Name = c.Name } ); This is because the Customer entities will end up partially populated. So, the next time you perform a query that requests all customer columns, you get the same cached Customer objects with only the Name property populated. In a multitier application, you cannot use a single static instance of a DataContext or ObjectContext in the middle tier to handle all requests, because contexts are not thread-safe. Instead, mid- dle-tier methods must create a fresh context per client request. This is actually beneficial because it shifts the burden in han- dling simultaneous updates to the database server, which is properly equipped for the job. A database server, for instance, will apply transaction isolation-level semantics. Associations The entity generation tools perform another useful job. For each relationship defined in your database, they generate properties on each side that allow you to query that relationship. For example, suppose we define a customer and purchase table in a one-to-many relationship: create table Customer ( ID int not null primary key, Name varchar(30) not null ) create table Purchase ( ID int not null primary key, CustomerID int references Customer (ID), Description varchar(30) not null, Price decimal not null ) With automatically generated entity classes, we can write queries such as this: var context = new NutshellContext ("connection string"); // Retrieve all purchases made by the first customer (alphabetically): Customer cust1 = context.Customers.OrderBy (c => c.Name).First(); foreach (Purchase p in cust1.Purchases) Console.WriteLine (p.Price); LINQ to SQL and Entity Framework | 361 LINQ Queries // Retrieve the customer who made the lowest value purchase: Purchase cheapest = context.Purchases.OrderBy (p => p.Price).First(); Customer cust2 = cheapest.Customer; Further, if cust1 and cust2 happened to refer to the same customer, c1 and c2 would refer to the same object: cust1==cust2 would return true. Let’s examine the signature of the automatically generated Purchases property on the Customer entity. With L2S: [Association (Storage="_Purchases", OtherKey="CustomerID")] public EntitySet Purchases { get {...} set {...} } With EF: [EdmRelationshipNavigationProperty ("NutshellModel", "FK...", "Purchase")] public EntityCollection Purchases { get {...} set {...} } An EntitySet or EntityCollection is like a predefined query, with a built-in Where clause that extracts related entities. The [Association] attribute gives L2S the information it needs to formulate the SQL query; the [EdmRelationshipNavigation Property] attribute tells EF where to look in the EDM for information about that relationship. As with any other type of query, you get deferred execution. With L2S, an Entity Set is populated when you enumerate over it; with EF, an EntityCollection is popu- lated when you explicitly call its Load method. Here’s the Purchases.Customer property, on the other side of the relationship, with L2S: [Association (Storage="_Customer",ThisKey="CustomerID",IsForeignKey=true)] public Customer Customer { get {...} set {...} } Although the property is of type Customer, its underlying field (_Customer) is of type EntityRef. The EntityRef type implements deferred loading, so the related Cus tomer is not retrieved from the database until you actually ask for it. EF works in the same way, except that it doesn’t populate the property simply by you accessing it: you must call Load on its EntityReference object. This means EF contexts must expose properties for both the actual parent object and its EntityRe ference wrapper: [EdmRelationshipNavigationProperty ("NutshellModel", "FK..., "Customer")] public Customer Customer { get {...} set {...} } public EntityReference CustomerReference { get; set; } You can make EF behave like L2S and have it populate Entity Collections and EntityReferences simply by virtue of their properties being accessed as follows: context.ContextOptions.DeferredLoadingEnabled = true; 362 | Chapter 8: LINQ Queries Deferred Execution with L2S and EF L2S and EF queries are subject to deferred execution, just like local queries. This allows you to build queries progressively. There is one aspect, however, in which L2S/EF have special deferred execution semantics, and that is when a subquery appears inside a Select expression: • With local queries, you get double deferred execution, because from a func- tional perspective, you’re selecting a sequence of queries. So, if you enumerate the outer result sequence, but never enumerate the inner sequences, the sub- query will never execute. • With L2S/EF, the subquery is executed at the same time as the main outer query. This avoids excessive round-tripping. For example, the following query executes in a single round trip upon reaching the first foreach statement: var context = new NutshellContext ("connection string"); var query = from c in context.Customers select from p in c.Purchases select new { c.Name, p.Price }; foreach (var customerPurchaseResults in query) foreach (var namePrice in customerPurchaseResults) Console.WriteLine (namePrice.Name + " spent " + namePrice.Price); Any EntitySets/EntityCollections that you explicitly project are fully populated in a single round trip: var query = from c in context.Customers select new { c.Name, c.Purchases }; foreach (var row in query) foreach (Purchase p in row.Purchases) // No extra round-tripping Console.WriteLine (row.Name + " spent " + p.Price); But if we enumerate EntitySet/EntityCollection properties without first having projected, deferred execution rules apply. In the following example, L2S and EF execute another Purchases query on each loop iteration: context.ContextOptions.DeferredLoadingEnabled = true; // For EF only. foreach (Customer c in context.Customers) foreach (Purchase p in c.Purchases) // Another SQL round-trip Console.WriteLine (c.Name + " spent " + p.Price); This model is advantageous when you want to selectively execute the inner loop, based on a test that can be performed only on the client: foreach (Customer c in context.Customers) if (myWebService.HasBadCreditHistory (c.ID)) foreach (Purchase p in c.Purchases) // Another SQL round trip Console.WriteLine (...); LINQ to SQL and Entity Framework | 363 LINQ Queries (In Chapter 9, we explore Select subqueries in more detail, in “Projec- ting” on page 377.) We’ve seen that you can avoid round-tripping by explicitly projecting associations. L2S and EF offer other mechanisms for this too, which we cover in the following two sections. DataLoadOptions The DataLoadOptions class is specific to L2S. It has two distinct uses: • It lets you specify, in advance, a filter for EntitySet associations (AssociateWith). • It lets you request that certain EntitySets be eagerly loaded, to lessen round- tripping (LoadWith). Specifying a filter in advance Let’s refactor our previous example as follows: foreach (Customer c in context.Customers) if (myWebService.HasBadCreditHistory (c.ID)) ProcessCustomer (c); We’ll define ProcessCustomer like this: void ProcessCustomer (Customer c) { Console.WriteLine (c.ID + " " + c.Name); foreach (Purchase p in c.Purchases) Console.WriteLine (" - purchased a " + p.Description); } Now suppose we want to feed ProcessCustomer only a subset of each customer’s purchases; say, the high-value ones. Here’s one solution: foreach (Customer c in context.Customers) if (myWebService.HasBadCreditHistory (c.ID)) ProcessCustomer (c.ID, c.Name, c.Purchases.Where (p => p.Price > 1000)); ... void ProcessCustomer (int custID, string custName, IEnumerable purchases) { Console.WriteLine (custID + " " + custName); foreach (Purchase p in purchases) Console.WriteLine (" - purchased a " + p.Description); } This is messy. It would get messier still if ProcessCustomer required more Customer fields. A better solution is to use DataLoadOptions’s AssociateWith method: DataLoadOptions options = new DataLoadOptions(); options.AssociateWith (c => c.Purchases.Where (p => p.Price > 1000)); context.LoadOptions = options; 364 | Chapter 8: LINQ Queries This instructs our DataContext instance always to filter a Customer’s Purchases using the given predicate. We can now use the original version of ProcessCustomer. AssociateWith doesn’t change deferred execution semantics. When a particular relationship is used, it simply instructs to implicitly add a particular filter to the equation. Eager loading The second use for a DataLoadOptions is to request that certain EntitySets be eagerly loaded with their parent. For instance, suppose you want to load all customers and their purchases in a single SQL round trip. The following does exactly this: DataLoadOptions options = new DataLoadOptions(); options.LoadWith (c => c.Purchases); context.LoadOptions = options; foreach (Customer c in context.Customers) // One round trip: foreach (Purchase p in c.Purchases) Console.WriteLine (c.Name + " bought a " + p.Description); This instructs that whenever a Customer is retrieved, its Purchases should also be retrieved at the same time. You can combine LoadWith with AssociateWith. The fol- lowing instructs that whenever a customer is retrieved, its high-value purchases should be retrieved in the same round trip: options.LoadWith (c => c.Purchases); options.AssociateWith (c => c.Purchases.Where (p => p.Price > 1000)); Eager Loading in Entity Framework You can request in EF that associations be eagerly loaded with the Include method. The following enumerates over each customer’s purchases—while generating just one SQL query: foreach (var c in context.Customers.Include ("Purchases")) foreach (var p in c.Purchases) Console.WriteLine (p.Description); Include can be used with arbitrary breadth and depth. For example, if each Pur chase also had PurchaseDetails and SalesPersons navigation properties, the entire nested structure could be eagerly loaded as follows: context.Customers.Include ("Purchases.PurchaseDetails") .Include ("Purchases.SalesPersons") Updates L2S and EF also keep track of changes that you make to your entities and allow you to write them back to the database by calling SubmitChanges on the DataContext object, or SaveChanges on the ObjectContext object. LINQ to SQL and Entity Framework | 365 LINQ Queries L2S’s Table class provides InsertOnSubmit and DeleteOnSubmit methods for inserting and deleting rows in a table; EF’s ObjectSet class provides AddObject and DeleteObject methods to do the same thing. Here’s how to insert a row: var context = new NutshellContext ("connection string"); Customer cust = new Customer { ID=1000, Name="Bloggs" }; context.Customers.InsertOnSubmit (cust); // AddObject with EF context.SubmitChanges(); // SaveChanges with EF We can later retrieve that row, update it, and then delete it: var context = new NutshellContext ("connection string"); Customer cust = context.Customers.Single (c => c.ID == 1000); cust.Name = "Bloggs2"; context.SubmitChanges(); // Updates the customer context.Customers.DeleteOnSubmit (cust); // DeleteObject with EF context.SubmitChanges(); // Deletes the customer SubmitChanges/SaveChanges gathers all the changes that were made to its entities since the context’s creation (or the last save), and then executes a SQL statement to write them to the database. Any TransactionScope is honored; if none is present it wraps all statements in a new transaction. You can also add new or existing rows to an EntitySet/EntityCollection by calling Add. L2S and EF automatically populate the foreign keys when you do this (after calling SubmitChanges or SaveChanges): Purchase p1 = new Purchase { ID=100, Description="Bike", Price=500 }; Purchase p2 = new Purchase { ID=101, Description="Tools", Price=100 }; Customer cust = context.Customers.Single (c => c.ID == 1); cust.Purchases.Add (p1); cust.Purchases.Add (p2); context.SubmitChanges(); // (or SaveChanges with EF) If you don’t want the burden of allocating unique keys, you can use either an auto-incrementing field (IDENTITY in SQL Server) or a Guid for the primary key. In this example, L2S/EF automatically writes 1 into the CustomerID column of each of the new purchases (L2S knows to do this because of the association attribute that we defined on the Purchases property; EF knows to do this because of information in the EDM). [Association (Storage="_Purchases", OtherKey="CustomerID")] public EntitySet Purchases { get {...} set {...} } 366 | Chapter 8: LINQ Queries If the Customer and Purchase entities were generated by the Visual Studio designer or the SqlMetal command-line tool, the generated classes would include further code to keep the two sides of each relationship in sync. In other words, assigning the Purchase.Customer property would automatically add the new customer to the Cus tomer.Purchases entity set—and vice versa. We can illustrate this by rewriting the preceding example as follows: var context = new NutshellContext ("connection string"); Customer cust = context.Customers.Single (c => c.ID == 1); new Purchase { ID=100, Description="Bike", Price=500, Customer=cust }; new Purchase { ID=101, Description="Tools", Price=100, Customer=cust }; context.SubmitChanges(); // (SaveChanges with EF) When you remove a row from an EntitySet/EntityCollection, its foreign key field is automatically set to null. The following disassociates our two recently added purchases from their customer: var context = new NutshellContext ("connection string"); Customer cust = context.Customers.Single (c => c.ID == 1); cust.Purchases.Remove (cust.Purchases.Single (p => p.ID == 100)); cust.Purchases.Remove (cust.Purchases.Single (p => p.ID == 101)); context.SubmitChanges(); // Submit SQL to database (SaveChanges in EF) Because this tries to set each purchase’s CustomerID field to null, Purchase.Custom erID must be nullable in the database; otherwise, an exception is thrown. (Further, the CustomerID field or property in the entity class must be a nullable type.) To delete child entities entirely, remove them from the Table or ObjectSet instead. With L2S: var c = context; c.Purchases.DeleteOnSubmit (c.Purchases.Single (p => p.ID == 100)); c.Purchases.DeleteOnSubmit (c.Purchases.Single (p => p.ID == 101)); c.SubmitChanges(); // Submit SQL to database With EF: var c = context; c.Purchases.DeleteObject (c.Purchases.Single (p => p.ID == 100)); c.Purchases.DeleteObject (c.Purchases.Single (p => p.ID == 101)); c.SaveChanges(); // Submit SQL to database API Differences Between L2S and EF As we’ve seen, L2S and EF are similar in the aspect of querying with LINQ and performing updates. Table 8-1 summarizes the API differences. LINQ to SQL and Entity Framework | 367 LINQ Queries Table 8-1. API differences between L2S and EF Purpose LINQ to SQL Entity Framework Gatekeeper class for all CRUD operations DataContext ObjectContext Method to (lazily) retrieve all entities of a given type from the store GetTable CreateObjectSet Type returned by the above method Table ObjectSet Method to update the store with any additions, modifications, or deletions to entity objects SubmitChanges SaveChanges Method to add a new entity to the store when the context is updated InsertOnSubmit AddObject Method to delete an entity from the store when the context is updated DeleteOnSubmit DeleteObject Type to represent one side of a relationship property, when that side has a multiplicity of many EntitySet EntityCollection Type to represent one side of a relationship property, when that side has a multiplicity of one EntityRef EntityReference Default strategy for loading relationship prop- erties Lazy Explicit Construct that enables eager loading DataLoadOptions .Include() Building Query Expressions So far in this chapter, when we’ve needed to dynamically compose queries, we’ve done so by conditionally chaining query operators. Although this is adequate in many scenarios, sometimes you need to work at a more granular level and dynam- ically compose the lambda expressions that feed the operators. In this section, we’ll assume the following Product class: [Table] public partial class Product { [Column(IsPrimaryKey=true)] public int ID; [Column] public string Description; [Column] public bool Discontinued; [Column] public DateTime LastSale; } Delegates Versus Expression Trees Recall that: • Local queries, which use Enumerable operators, take delegates. • Interpreted queries, which use Queryable operators, take expression trees. 368 | Chapter 8: LINQ Queries We can see this by comparing the signature of the Where operator in Enumerable and Queryable: public static IEnumerable Where (this IEnumerable source, Func predicate) public static IQueryable Where (this IQueryable source, Expression> predicate) When embedded within a query, a lambda expression looks identical whether it binds to Enumerable’s operators or Queryable’s operators: IEnumerable q1 = localProducts.Where (p => !p.Discontinued); IQueryable q2 = sqlProducts.Where (p => !p.Discontinued); When you assign a lambda expression to an intermediate variable, however, you must be explicit on whether to resolve to a delegate (i.e., Func<>) or an expression tree (i.e., Expression>). In the following example, predicate1 and predi cate2 are not interchangeable: Func predicate1 = p => !p.Discontinued; IEnumerable q1 = localProducts.Where (predicate1); Expression > predicate2 = p => !p.Discontinued; IQueryable q2 = sqlProducts.Where (predicate2); Compiling expression trees You can convert an expression tree to a delegate by calling Compile. This is of par- ticular value when writing methods that return reusable expressions. To illustrate, we’ll add a static method to the Product class that returns a predicate evaluating to true if a product is not discontinued and has sold in the past 30 days: public partial class Product { public static Expression> IsSelling() { return p => !p.Discontinued && p.LastSale > DateTime.Now.AddDays (-30); } } (We’ve defined this in a separate partial class to avoid being overwritten by an automatic DataContext generator such as Visual Studio’s code generator.) The method just written can be used both in interpreted and in local queries as follows: void Test() { var dataContext = new NutshellContext ("connection string"); Product[] localProducts = dataContext.Products.ToArray(); IQueryable sqlQuery = dataContext.Products.Where (Product.IsSelling()); IEnumerable localQuery = Building Query Expressions | 369 LINQ Queries localProducts.Where (Product.IsSelling.Compile()); } You cannot convert in the reverse direction, from a delegate to an expression tree. This makes expression trees more versatile. AsQueryable The AsQueryable operator lets you write whole queries that can run over either local or remote sequences: IQueryable FilterSortProducts (IQueryable input) { return from p in input where ... order by ... select p; } void Test() { var dataContext = new NutshellContext ("connection string"); Product[] localProducts = dataContext.Products.ToArray(); var sqlQuery = FilterSortProducts (dataContext.Products); var localQuery = FilterSortProducts (localProducts.AsQueryable()); ... } AsQueryable wraps IQueryable clothing around a local sequence so that subse- quent query operators resolve to expression trees. When you later enumerate over the result, the expression trees are implicitly compiled (at a small performance cost), and the local sequence enumerates as it would ordinarily. Expression Trees We said previously that an implicit conversion from a lambda expression to Expres sion causes the C# compiler to emit code that builds an expression tree. With some programming effort, you can do the same thing manually at runtime— in other words, dynamically build an expression tree from scratch. The result can be cast to an Expression and used in LINQ-to-db queries, or compiled into an ordinary delegate by calling Compile. The Expression DOM An expression tree is a miniature code DOM. Each node in the tree is represented by a type in the System.Linq.Expressions namespace; these types are illustrated in Figure 8-10. 370 | Chapter 8: LINQ Queries From Framework 4.0, this namespace features additional ex- pression types and methods to support language constructs that can appear in code blocks. These are for the benefit of the DLR and not lambda expressions. In other words, code-block-style lambdas still cannot be converted to expression trees: Expression> invalid = { return true; } // Code blocks not permitted The base class for all nodes is the (nongeneric) Expression class. The generic Expres sion class actually means “typed lambda expression” and might have been named LambdaExpression if it wasn’t for the clumsiness of this: LambdaExpression> f = ... Expression’s base type is the (nongeneric) LambdaExpression class. LamdbaExpres sion provides type unification for lambda expression trees: any typed Expres sion can be cast to a LambdaExpression. The thing that distinguishes LambdaExpressions from ordinary Expressions is that lambda expressions have parameters. Figure 8-10. Expression types To create an expression tree, don’t instantiate node types directly; rather, call static methods provided on the Expression class. Here are all the methods: Add AddChecked And AndAlso ArrayIndex ArrayLength ElementInit Equal ExclusiveOr Field GreaterThan GreaterThanOrEqual MakeMemberAccess MakeUnary MemberBind MemberInit Modulo Multiply Or OrElse Parameter Power Property PropertyOrField Building Query Expressions | 371 LINQ Queries Bind Call Coalesce Condition Constant Convert ConvertChecked Divide Invoke Lambda LeftShift LessThan LessThanOrEqual ListBind ListInit MakeBinary MultiplyChecked Negate NegateChecked New NewArrayBounds NewArrayInit Not NotEqual Quote RightShift Subtract SubtractChecked TypeAs TypeIs UnaryPlus Figure 8-11 shows the expression tree that the following assignment creates: Expression> f = s => s.Length < 5; Figure 8-11. Expression tree We can demonstrate this as follows: Console.WriteLine (f.Body.NodeType); // LessThan Console.WriteLine (((BinaryExpression) f.Body).Right); // 5 Let’s now build this expression from scratch. The principle is that you start from the bottom of the tree and work your way up. The bottommost thing in our tree is a ParameterExpression, the lambda expression parameter called “s” of type string: ParameterExpression p = Expression.Parameter (typeof (string), "s"); 372 | Chapter 8: LINQ Queries The next step is to build the MemberExpression and ConstantExpression. In the former case, we need to access the Length property of our parameter, “s”: MemberExpression stringLength = Expression.Property (p, "Length"); ConstantExpression five = Expression.Constant (5); Next is the LessThan comparison: BinaryExpression comparison = Expression.LessThan (stringLength, five); The final step is to construct the lambda expression, which links an expression Body to a collection of parameters: Expression> lambda = Expression.Lambda> (comparison, p); A convenient way to test our lambda is by compiling it to a delegate: Func runnable = lambda.Compile(); Console.WriteLine (runnable ("kangaroo")); // False Console.WriteLine (runnable ("dog")); // True The easiest way to figure out which expression type to use is to examine an existing lambda expression in the Visual Studio debugger. We continue this discussion online, at http://www.albahari.com/expressions/. Building Query Expressions | 373 LINQ Queries 9LINQ Operators This chapter describes each of the LINQ query operators. As well as serving as a reference, two of the sections, “Projection” and “Joining” on page 378, cover a number of conceptual areas: • Projecting object hierarchies • Joining with Select, SelectMany, Join, and GroupJoin • Query expressions with multiple range variables All of the examples in this chapter assume that a names array is defined as follows: string[] names = { "Tom", "Dick", "Harry", "Mary", "Jay" }; Examples that query a database assume that a typed DataContext variable called dataContext is instantiated as follows: var dataContext = new NutshellContext ("connection string..."); ... public class NutshellContext : DataContext { public NutshellContext (string cxString) : base (cxString) {} public Table Customers { get { return GetTable(); } } public Table Purchases { get { return GetTable(); } } } [Table] public class Customer { [Column(IsPrimaryKey=true)] public int ID; [Column] public string Name; [Association (OtherKey="CustomerID")] public EntitySet Purchases = new EntitySet(); } [Table] public class Purchase 375 { [Column(IsPrimaryKey=true)] public int ID; [Column] public int? CustomerID; [Column] public string Description; [Column] public decimal Price; [Column] public DateTime Date; EntityRef custRef; [Association (Storage="custRef",ThisKey="CustomerID",IsForeignKey=true)] public Customer Customer { get { return custRef.Entity; } set { custRef.Entity = value; } } } All the examples in this chapter are preloaded into LINQPad, along with a sample database with a matching schema. You can download LINQPad from http://www.linqpad.net. The entity classes shown are a simplified version of what LINQ to SQL tools typically produce, and do not include code to update the opposing side in a relationship when their entities have been reassigned. Here are the corresponding SQL table definitions: create table Customer ( ID int not null primary key, Name varchar(30) not null ) create table Purchase ( ID int not null primary key, CustomerID int references Customer (ID), Description varchar(30) not null, Price decimal not null ) All examples will also work with Entity Framework, except where otherwise indicated. You can build an Entity Framework ObjectContext from these tables by creating a new Entity Data Model in Visual Studio, and then dragging the tables on to the designer surface. 376 | Chapter 9: LINQ Operators Overview In this section, we provide an overview of the standard query operators. The standard query operators fall into three categories: • Sequence in, sequence out (sequence-to-sequence) • Sequence in, single element or scalar value out • Nothing in, sequence out (generation methods) We first present each of the three categories and the query operators they include, and then we take up each individual query operator in detail. Sequence→Sequence Most query operators fall into this category—accepting one or more sequences as input and emitting a single output sequence. Figure 9-1 illustrates those operators that restructure the shape of the sequences. Figure 9-1. Shape-changing operators Filtering IEnumerable →IEnumerable Returns a subset of the original elements. Where, Take, TakeWhile, Skip, SkipWhile, Distinct Projecting IEnumerable →IEnumerable Transforms each element with a lambda function. SelectMany flattens nested sequences; Select and SelectMany perform inner joins, left outer joins, cross joins, and non-equi joins with LINQ to SQL and EF. Select, SelectMany Overview | 377 LINQ Operators Joining IEnumerable, IEnumerable→ IEnumerable Meshes elements of one sequence with another. Join and GroupJoin operators are designed to be efficient with local queries and support inner and left outer joins. The Zip operator enumerates two sequences in step, applying a function over each ele- ment pair. Rather than naming the type arguments TOuter and TInner, the Zip operator names them TFirst and TSecond: IEnumerable, IEnumerable→ IEnumerable Join, GroupJoin, Zip Ordering IEnumerable →IOrderedEnumerable Returns a reordering of a sequence. OrderBy, ThenBy, Reverse Grouping IEnumerable →IEnumerable> Groups a sequence into subsequences. GroupBy Set operators IEnumerable, IEnumerable→ IEnumerable Takes two same-typed sequences and returns their commonality, sum, or difference. Concat, Union, Intersect, Except Conversion methods: Import IEnumerable→IEnumerable OfType, Cast Conversion methods: Export IEnumerable →An array, list, dictionary, lookup, or sequence ToArray, ToList, ToDictionary, ToLookup, AsEnumerable, AsQueryable Sequence→Element or Value The following query operators accept an input sequence and emit a single element or value. Element operators IEnumerable →TSource 378 | Chapter 9: LINQ Operators Picks a single element from a sequence. First, FirstOrDefault, Last, LastOrDefault, Single, SingleOrDefault, ElementAt, ElementAtOrDefault, DefaultIfEmpty Aggregation methods IEnumerable →scalar Performs a computation across a sequence, returning a scalar value (typically a number). Aggregate, Average, Count, LongCount, Sum, Max, Min Quantifiers IEnumerable →bool An aggregation returning true or false. All, Any, Contains, SequenceEqual Void→Sequence In the third and final category are query operators that produce an output sequence from scratch. Generation methods void→IEnumerable Manufactures a simple sequence. Empty, Range, Repeat Filtering IEnumerable→ IEnumerable Method Description SQL equivalents Where Returns a subset of elements that satisfy a given condition WHERE Take Returns the first count elements and discards the rest WHERE ROW_NUMBER()... or TOP n subquery Skip Ignores the first count elements and returns the rest WHERE ROW_NUMBER()... or NOT IN (SELECT TOP n...) TakeWhile Emits elements from the input sequence until the predicate is false Exception thrown SkipWhile Ignores elements from the input sequence until the predicate is false, and then emits the rest Exception thrown Distinct Returns a sequence that excludes duplicates SELECT DISTINCT... Filtering | 379 LINQ Operators The “SQL equivalents” column in the reference tables in this chapter do not necessarily correspond to what an IQueryable implementation such as LINQ to SQL will produce. Rather, it indicates what you’d typically use to do the same job if you were writing the SQL query yourself. Where there is no simple trans- lation, the column is left blank. Where there is no translation at all, the column reads “Exception thrown”. Enumerable implementation code, when shown, excludes check- ing for null arguments and indexing predicates. With each of the filtering methods, you always end up with either the same number or fewer elements than you started with. You can never get more! The elements are also identical when they come out; they are not transformed in any way. Where Argument Type Source sequence IEnumerable Predicate TSource => bool or (TSource,int) => boola a. Prohibited with LINQ to SQL and Entity Framework Query syntax where bool-expression Enumerable.Where implementation The internal implementation of Enumerable.Where, null checking aside, is function- ally equivalent to the following: public static IEnumerable Where (this IEnumerable source, Func predicate) { foreach (TSource element in source) if (predicate (element)) yield return element; } Overview Where returns the elements from the input sequence that satisfy the given predicate. For instance: string[] names = { "Tom", "Dick", "Harry", "Mary", "Jay" }; IEnumerable query = names.Where (name => name.EndsWith ("y")); // Result: { "Harry", "Mary", "Jay" } 380 | Chapter 9: LINQ Operators In query syntax: IEnumerable query = from n in names where n.EndsWith ("y") select n; A where clause can appear more than once in a query and be interspersed with let, orderby and join clauses: from n in names where n.Length > 3 let u = n.ToUpper() where u.EndsWith ("Y") select u; // Result: { "HARRY", "MARY" } Standard C# scoping rules apply to such queries. In other words, you cannot refer to a variable prior to declaring it with a range variable or a let clause. Indexed filtering Where’s predicate optionally accepts a second argument, of type int. This is fed with the position of each element within the input sequence, allowing the predicate to use this information in its filtering decision. For example, the following skips every second element: IEnumerable query = names.Where ((n, i) => i % 2 == 0); // Result: { "Tom", "Harry", "Jay" } An exception is thrown if you use indexed filtering in LINQ to SQL or EF. SQL LIKE comparisons in LINQ to SQL and EF The following methods on string translate to SQL’s LIKE operator: Contains, StartsWith, EndsWith For instance, c.Name.Contains ("abc") translates to customer.Name LIKE '%abc%' (or more accurately, a parameterized version of this). Contains lets you compare only against a locally evaluated expression; to compare against another column, you must use the SqlMethods.Like method: ... where SqlMethods.Like (c.Description, "%" + c.Name + "%") SqlMethods.Like also lets you perform more complex comparisons (e.g., LIKE 'abc %def%'). < and > string comparisons in LINQ to SQL and EF You can perform order comparison on strings with string’s CompareTo method; this maps to SQL’s < and > operators: dataContext.Purchases.Where (p => p.Description.CompareTo ("C") < 0) Filtering | 381 LINQ Operators WHERE x IN (..., ..., ...) in LINQ to SQL and EF With LINQ to SQL and EF, you can apply the Contains operator to a local collection within a filter predicate. For instance: string[] chosenOnes = { "Tom", "Jay" }; from c in dataContext.Customers where chosenOnes.Contains (c.Name) ... This maps to SQL’s IN operator—in other words: WHERE customer.Name IN ("Tom", "Jay") If the local collection is an array of entities or nonscalar types, LINQ to SQL or EF may instead emit an EXISTS clause. Take and Skip Argument Type Source sequence IEnumerable Number of elements to take or skip int Take emits the first n elements and discards the rest; Skip discards the first n elements and emits the rest. The two methods are useful together when implementing a web page allowing a user to navigate through a large set of matching records. For instance, suppose a user searches a book database for the term “mercury”, and there are 100 matches. The following returns the first 20: IQueryable query = dataContext.Books .Where (b => b.Title.Contains ("mercury")) .OrderBy (b => b.Title) .Take (20); The next query returns books 21 to 40: IQueryable query = dataContext.Books .Where (b => b.Title.Contains ("mercury")) .OrderBy (b => b.Title) .Skip (20).Take (20); LINQ to SQL and EF translate Take and Skip to the ROW_NUMBER function in SQL Server 2005, or a TOP n subquery in earlier versions of SQL Server. TakeWhile and SkipWhile Argument Type Source sequence IEnumerable Predicate TSource => bool or (TSource,int) => bool 382 | Chapter 9: LINQ Operators TakeWhile enumerates the input sequence, emitting each item, until the given pred- icate is false. It then ignores the remaining elements: int[] numbers = { 3, 5, 2, 234, 4, 1 }; var takeWhileSmall = numbers.TakeWhile (n => n < 100); // { 3, 5, 2 } SkipWhile enumerates the input sequence, ignoring each item until the given pred- icate is false. It then emits the remaining elements: int[] numbers = { 3, 5, 2, 234, 4, 1 }; var skipWhileSmall = numbers.SkipWhile (n => n < 100); // { 234, 4, 1 } TakeWhile and SkipWhile have no translation to SQL and cause a runtime error if used in a LINQ-to-db query. Distinct Distinct returns the input sequence, stripped of duplicates. You can optionally pass in a custom equality comparer. The following returns distinct letters in a string: char[] distinctLetters = "HelloWorld".Distinct().ToArray(); string s = new string (distinctLetters); // HeloWrd We can call LINQ methods directly on a string, because string implements IEnumerable. Projecting IEnumerable→ IEnumerable Method Description SQL equivalents Select Transforms each input element with the given lambda expression SELECT SelectMany Transforms each input element, and then flattens and concatenates the resultant subsequences INNER JOIN, LEFT OUTER JOIN, CROSS JOIN When querying a database, Select and SelectMany are the most versatile joining constructs; for local queries, Join and Group Join are the most efficient joining constructs. Select Argument Type Source sequence IEnumerable Result selector TSource => TResult or (TSource,int) => TResulta a. Prohibited with LINQ to SQL and Entity Framework Projecting | 383 LINQ Operators Query syntax select projection-expression Enumerable implementation public static IEnumerable Select (this IEnumerable source, Func selector) { foreach (TSource element in source) yield return selector (element); } Overview With Select, you always get the same number of elements that you started with. Each element, however, can be transformed in any manner by the lambda function. The following selects the names of all fonts installed on the computer (from System.Drawing): IEnumerable query = from f in FontFamily.Families select f.Name; foreach (string name in query) Console.WriteLine (name); In this example, the select clause converts a FontFamily object to its name. Here’s the lambda equivalent: IEnumerable query = FontFamily.Families.Select (f => f.Name); Select statements are often used to project into anonymous types: var query = from f in FontFamily.Families select new { f.Name, LineSpacing = f.GetLineSpacing (FontStyle.Bold) }; A projection with no transformation is sometimes used with query syntax, in order to satisfy the requirement that the query end in a select or group clause. The fol- lowing selects fonts supporting strikeout: IEnumerable query = from f in FontFamily.Families where f.IsStyleAvailable (FontStyle.Strikeout) select f; foreach (FontFamily ff in query) Console.WriteLine (ff.Name); In such cases, the compiler omits the projection when translating to fluent syntax. Indexed projection The selector expression can optionally accept an integer argument, which acts as an indexer, providing the expression with the position of each input in the input sequence. This works only with local queries: string[] names = { "Tom", "Dick", "Harry", "Mary", "Jay" }; 384 | Chapter 9: LINQ Operators IEnumerable query = names .Select ((s,i) => i + "=" + s); // { "0=Tom", "1=Dick", ... } Select subqueries and object hierarchies You can nest a subquery in a select clause to build an object hierarchy. The fol- lowing example returns a collection describing each directory under D:\source, with a subcollection of files under each directory: DirectoryInfo[] dirs = new DirectoryInfo (@"d:\source").GetDirectories(); var query = from d in dirs where (d.Attributes & FileAttributes.System) == 0 select new { DirectoryName = d.FullName, Created = d.CreationTime, Files = from f in d.GetFiles() where (f.Attributes & FileAttributes.Hidden) == 0 select new { FileName = f.Name, f.Length, } }; foreach (var dirFiles in query) { Console.WriteLine ("Directory: " + dirFiles.DirectoryName); foreach (var file in dirFiles.Files) Console.WriteLine (" " + file.FileName + " Len: " + file.Length); } The inner portion of this query can be called a correlated subquery. A subquery is correlated if it references an object in the outer query—in this case, it references d, the directory being enumerated. A subquery inside a Select allows you to map one object hierarchy to another, or map a relational object model to a hi- erarchical object model. With local queries, a subquery within a Select causes double-deferred execution. In our example, the files don’t get filtered or projected until the inner foreach state- ment enumerates. Subqueries and joins in LINQ to SQL and EF Subquery projections work well in LINQ to SQL and EF and can be used to do the work of SQL-style joins. Here’s how we retrieve each customer’s name along with their high-value purchases: var query = from c in dataContext.Customers select new { c.Name, Projecting | 385 LINQ Operators Purchases = from p in dataContext.Purchases where p.CustomerID == c.ID && p.Price > 1000 select new { p.Description, p.Price } }; foreach (var namePurchases in query) { Console.WriteLine ("Customer: " + namePurchases.Name); foreach (var purchaseDetail in namePurchases.Purchases) Console.WriteLine (" - $$$: " + purchaseDetail.Price); } This style of query is ideally suited to interpreted queries. The outer query and subquery are processed as a unit, avoiding unnecessary round-tripping. With local queries, however, it’s inefficient because every combination of outer and inner ele- ments must be enumerated to get the few matching combina- tions. A better choice for local queries is Join or GroupJoin, described in the following sections. This query matches up objects from two disparate collections, and it can be thought of as a “Join.” The difference between this and a conventional database join (or subquery) is that we’re not flattening the output into a single two-dimensional result set. We’re mapping the relational data to hierarchical data, rather than to flat data. Here’s the same query simplified by using the Purchases association property on the Customer entity: from c in dataContext.Customers select new { c.Name, Purchases = from p in c.Purchases // Purchases is EntitySet where p.Price > 1000 select new { p.Description, p.Price } }; Both queries are analogous to a left outer join in SQL in the sense that we get all customers in the outer enumeration, regardless of whether they have any purchases. To emulate an inner join—where customers without high-value purchases are excluded—we would need to add a filter condition on the purchases collection: from c in dataContext.Customers where c.Purchases.Any (p => p.Price > 1000) select new { c.Name, Purchases = from p in c.Purchases where p.Price > 1000 select new { p.Description, p.Price } }; 386 | Chapter 9: LINQ Operators This is slightly untidy, however, in that we’ve written the same predicate (Price > 1000) twice. We can avoid this duplication with a let clause: from c in dataContext.Customers let highValueP = from p in c.Purchases where p.Price > 1000 select new { p.Description, p.Price } where highValueP.Any() select new { c.Name, Purchases = highValueP }; This style of query is flexible. By changing Any to Count, for instance, we can modify the query to retrieve only customers with at least two high-value purchases: ... where highValueP.Count() >= 2 select new { c.Name, Purchases = highValueP }; Projecting into concrete types Projecting into anonymous types is useful in obtaining intermediate results, but not so useful if you want to send a result set back to a client, for instance, because anonymous types can exist only as local variables within a method. An alternative is to use concrete types for projections, such as DataSets or custom business entity classes. A custom business entity is simply a class that you write with some proper- ties, similar to a LINQ to SQL [Table] annotated class or an EF Entity, but designed to hide lower-level (database-related) details. You might exclude foreign key fields from business entity classes, for instance. Assuming we wrote custom entity classes called CustomerEntity and PurchaseEntity, here’s how we could project into them: IQueryable query = from c in dataContext.Customers select new CustomerEntity { Name = c.Name, Purchases = (from p in c.Purchases where p.Price > 1000 select new PurchaseEntity { Description = p.Description, Value = p.Price } ).ToList() }; // Force query execution, converting output to a more convenient List: List result = query.ToList(); Notice that so far, we’ve not had to use a Join or SelectMany statement. This is because we’re maintaining the hierarchical shape of the data, as illustrated in Fig- ure 9-2. With LINQ, you can often avoid the traditional SQL approach of flattening tables into a two-dimensional result set. Projecting | 387 LINQ Operators SelectMany Argument Type Source sequence IEnumerable Result selector TSource => IEnumerable or (TSource,int) => IEnumerablea a. Prohibited with LINQ to SQL Query syntax from identifier1 in enumerable-expression1 from identifier2 in enumerable-expression2 ... Enumerable implementation public static IEnumerable SelectMany (IEnumerable source, Func > selector) { foreach (TSource element in source) foreach (TResult subElement in selector (element)) yield return subElement; } Overview SelectMany concatenates subsequences into a single flat output sequence. Recall that for each input element, Select yields exactly one output element. In contrast, SelectMany yields 0..n output elements. The 0..n elements come from a subsequence or child sequence that the lambda expression must emit. SelectMany can be used to expand child sequences, flatten nested collections, and join two collections into a flat output sequence. Using the conveyor belt analogy, SelectMany funnels fresh material onto a conveyor belt. With SelectMany, each input Figure 9-2. Projecting an object hierarchy 388 | Chapter 9: LINQ Operators element is the trigger for the introduction of fresh material. The fresh material is emitted by the selector lambda expression and must be a sequence. In other words, the lambda expression must emit a child sequence per input element. The final result is a concatenation of the child sequences emitted for each input element. Starting with a simple example, suppose we have an array of names as follows: string[] fullNames = { "Anne Williams", "John Fred Smith", "Sue Green" }; which we wish to convert to a single flat collection of words—in other words: "Anne", "Williams", "John", "Fred", "Smith", "Sue", Green" SelectMany is ideal for this task, because we’re mapping each input element to a variable number of output elements. All we must do is come up with a selector expression that converts each input element to a child sequence. string.Split does the job nicely: it takes a string and splits it into words, emitting the result as an array: string testInputElement = "Anne Williams"; string[] childSequence = testInputElement.Split(); // childSequence is { "Anne", "Williams" }; So, here’s our SelectMany query and the result: IEnumerable query = fullNames.SelectMany (name => name.Split()); foreach (string name in query) Console.Write (name + "|"); // Anne|Williams|John|Fred|Smith|Sue|Green| If you replace SelectMany with Select, you get the same results in hierarchical form. The following emits a sequence of string arrays, requiring nested foreach statements to enumerate: IEnumerable query = fullNames.Select (name => name.Split()); foreach (string[] stringArray in query) foreach (string name in stringArray) Console.Write (name + "|"); The benefit of SelectMany is that it yields a single flat result sequence. SelectMany is supported in query syntax and is invoked by having an additional gen- erator—in other words, an extra from clause in the query. The from keyword has two meanings in query syntax. At the start of a query, it introduces the original range variable and input sequence. Anywhere else in the query, it translates to Select Many. Here’s our query in query syntax: IEnumerable query = from fullName in fullNames from name in fullName.Split() // Translates to SelectMany select name; Projecting | 389 LINQ Operators Note that the additional generator introduces a new range variable—in this case, name. The old range variable stays in scope, however, and we can subsequently access both. Multiple range variables In the preceding example, both name and fullName remain in scope until the query either ends or reaches an into clause. The extended scope of these variables is the killer scenario for query syntax over fluent syntax. To illustrate, we can take the preceding query and include fullName in the final projection: IEnumerable query = from fullName in fullNames from name in fullName.Split() select name + " came from " + fullName; Anne came from Anne Williams Williams came from Anne Williams John came from John Fred Smith ... Behind the scenes, the compiler must pull some tricks to let you access both vari- ables. A good way to appreciate this is to try writing the same query in fluent syntax. It’s tricky! It gets harder still if you insert a where or orderby clause before projecting: from fullName in fullNames from name in fullName.Split() orderby fullName, name select name + " came from " + fullName; The problem is that SelectMany emits a flat sequence of child elements—in our case, a flat collection of words. The original “outer” element from which it came (full Name) is lost. The solution is to “carry” the outer element with each child, in a tem- porary anonymous type: from fullName in fullNames from x in fullName.Split().Select (name => new { name, fullName } ) orderby x.fullName, x.name select x.name + " came from " + x.fullName; The only change here is that we’re wrapping each child element (name) in an anony- mous type that also contains its fullName. This is similar to how a let clause is resolved. Here’s the final conversion to fluent syntax: IEnumerable query = fullNames .SelectMany (fName => fName.Split() .Select (name => new { name, fName } )) .OrderBy (x => x.fName) .ThenBy (x => x.name) .Select (x => x.name + " came from " + x.fName); 390 | Chapter 9: LINQ Operators Thinking in query syntax As we just demonstrated, there are good reasons to use query syntax if you need multiple range variables. In such cases, it helps not only to use query syntax, but also to think directly in its terms. There are two basic patterns when writing additional generators. The first is expanding and flattening subsequences. To do this, you call a property or method on an existing range variable in your additional generator. We did this in the previous example: from fullName in fullNames from name in fullName.Split() Here, we’ve expanded from enumerating full names to enumerating words. An analogous LINQ-to-db query is when you expand child association properties. The following query lists all customers along with their purchases: IEnumerable query = from c in dataContext.Customers from p in c.Purchases select c.Name + " bought a " + p.Description; Tom bought a Bike Tom bought a Holiday Dick bought a Phone Harry bought a Car ... Here, we’ve expanded each customer into a subsequence of purchases. The second pattern is performing a cartesian product or cross join—where every element of one sequence is matched with every element of another. To do this, introduce a generator whose selector expression returns a sequence unrelated to a range variable: int[] numbers = { 1, 2, 3 }; string[] letters = { "a", "b" }; IEnumerable query = from n in numbers from l in letters select n.ToString() + l; RESULT: { "1a", "1b", "2a", "2b", "3a", "3b" } This style of query is the basis of SelectMany-style joins. Joining with SelectMany You can use SelectMany to join two sequences, simply by filtering the results of a cross product. For instance, suppose we wanted to match players for a game. We could start as follows: string[] players = { "Tom", "Jay", "Mary" }; IEnumerable query = from name1 in players from name2 in players select name1 + " vs " + name2; Projecting | 391 LINQ Operators RESULT: { "Tom vs Tom", "Tom vs Jay", "Tom vs Mary", "Jay vs Tom", "Jay vs Jay", "Jay vs Mary", "Mary vs Tom", "Mary vs "Jay", "Mary vs Mary" } The query reads: “For every player, reiterate every player, selecting player 1 vs player 2.” Although we got what we asked for (a cross join), the results are not useful until we add a filter: IEnumerable query = from name1 in players from name2 in players where name1.CompareTo (name2) < 0 orderby name1, name2 select name1 + " vs " + name2; RESULT: { "Jay vs Mary", "Jay vs Tom", "Mary vs Tom" } The filter predicate constitutes the join condition. Our query can be called a non-equi join, because the join condition doesn’t use an equality operator. We’ll demonstrate the remaining types of joins with LINQ to SQL (they’ll also work with EF except where we explicitly use a foreign key field). SelectMany in LINQ to SQL and EF SelectMany in LINQ to SQL and EF can perform cross joins, non-equi joins, inner joins, and left outer joins. You can use SelectMany with both predefined associations and ad hoc relationships—just as with Select. The difference is that SelectMany returns a flat rather than a hierarchical result set. A LINQ-to-db cross join is written just as in the preceding section. The following query matches every customer to every purchase (a cross join): var query = from c in dataContext.Customers from p in dataContext.Purchases select c.Name + " might have bought a " + p.Description; More typically, though, you’d want to match customers to their own purchases only. You achieve this by adding a where clause with a joining predicate. This results in a standard SQL-style equi-join: var query = from c in dataContext.Customers from p in dataContext.Purchases where c.ID == p.CustomerID select c.Name + " bought a " + p.Description; This translates well to SQL. In the next section, we’ll see how it extends to support outer joins. Reformulating such queries with LINQ’s Join operator actually makes them less extensible —LINQ is opposite to SQL in this sense. 392 | Chapter 9: LINQ Operators If you have association properties for relationships in your entities, you can express the same query by expanding the subcollection instead of filtering the cross product: from c in dataContext.Customers from p in c.Purchases select new { c.Name, p.Description }; Entity Framework doesn’t expose foreign keys in the entities, so for recognized relationships you must use its association properties rather than joining manually as we did previously. The advantage is that we’ve eliminated the joining predicate. We’ve gone from fil- tering a cross product to expanding and flattening. Both queries, however, will result in the same SQL. You can add where clauses to such a query for additional filtering. For instance, if we wanted only customers whose names started with “T”, we could filter as follows: from c in dataContext.Customers where c.Name.StartsWith ("T") from p in c.Purchases select new { c.Name, p.Description }; This LINQ-to-db query would work equally well if the where clause is moved one line down. If it is a local query, however, moving the where clause down would make it less efficient. With local queries, you should filter before joining. You can introduce new tables into the mix with additional from clauses. For instance, if each purchase had purchase item child rows, you could produce a flat result set of customers with their purchases, each with their purchase detail lines as follows: from c in dataContext.Customers from p in c.Purchases from pi in p.PurchaseItems select new { c.Name, p.Description, pi.DetailLine }; Each from clause introduces a new child table. To include data from a parent table (via an association property), you don’t add a from clause—you simply navigate to the property. For example, if each customer has a salesperson whose name you want to query, just do this: from c in dataContext.Customers select new { Name = c.Name, SalesPerson = c.SalesPerson.Name }; You don’t use SelectMany in this case because there’s no subcollection to flatten. Parent association properties return a single item. Outer joins with SelectMany We saw previously that a Select subquery yields a result analogous to a left outer join. from c in dataContext.Customers select new { Projecting | 393 LINQ Operators c.Name, Purchases = from p in c.Purchases where p.Price > 1000 select new { p.Description, p.Price } }; In this example, every outer element (customer) is included, regardless of whether the customer has any purchases. But suppose we rewrite this query with Select Many, so we can obtain a single flat collection rather than a hierarchical result set: from c in dataContext.Customers from p in c.Purchases where p.Price > 1000 select new { c.Name, p.Description, p.Price }; In the process of flattening the query, we’ve switched to an inner join: customers are now included only for whom one or more high-value purchases exist. To get a left outer join with a flat result set, we must apply the DefaultIfEmpty query operator on the inner sequence. This method returns a sequence with a single null element if its input sequence has no elements. Here’s such a query, price predicate aside: from c in dataContext.Customers from p in c.Purchases.DefaultIfEmpty() select new { c.Name, p.Description, Price = (decimal?) p.Price }; This works perfectly with LINQ to SQL and EF, returning all customers, even if they have no purchases. But if we were to run this as a local query, it would crash, because when p is null, p.Description and p.Price throw a NullReferenceException. We can make our query robust in either scenario as follows: from c in dataContext.Customers from p in c.Purchases.DefaultIfEmpty() select new { c.Name, Descript = p == null ? null : p.Description, Price = p == null ? (decimal?) null : p.Price }; Let’s now reintroduce the price filter. We cannot use a where clause as we did before, because it would execute after DefaultIfEmpty: from c in dataContext.Customers from p in c.Purchases.DefaultIfEmpty() where p.Price > 1000... The correct solution is to splice the Where clause before DefaultIfEmpty with a subquery: from c in dataContext.Customers from p in c.Purchases.Where (p => p.Price > 1000).DefaultIfEmpty() select new { c.Name, Descript = p == null ? null : p.Description, Price = p == null ? (decimal?) null : p.Price }; 394 | Chapter 9: LINQ Operators LINQ to SQL and EF translate this to a left outer join. This is an effective pattern for writing such queries. If you’re used to writing outer joins in SQL, you might be tempted to overlook the simpler option of a Select subquery for this style of query, in favor of the awkward but familiar SQL- centric flat approach. The hierarchical result set from a Select subquery is often better suited to outer join-style queries because there are no additional nulls to deal with. Joining Method Description SQL equivalents Join Applies a lookup strategy to match elements from two collections, emitting a flat result set INNER JOIN GroupJoin As above, but emits a hierarchical result set INNER JOIN, LEFT OUTER JOIN Zip Enumerates two sequences in step (like a zipper), applying a function over each element pair Join and GroupJoin IEnumerable, IEnumerable→IEnumerable Join arguments Argument Type Outer sequence IEnumerable Inner sequence IEnumerable Outer key selector TOuter => TKey Inner key selector TInner => TKey Result selector (TOuter,TInner) => TResult GroupJoin arguments Argument Type Outer sequence IEnumerable Inner sequence IEnumerable Outer key selector TOuter => TKey Inner key selector TInner => TKey Result selector (TOuter,IEnumerable) => TResult Joining | 395 LINQ Operators Query syntax from outer-var in outer-enumerable join inner-var in inner-enumerable on outer-key-expr equals inner-key-expr [ into identifier ] Overview Join and GroupJoin mesh two input sequences into a single output sequence. Join emits flat output; GroupJoin emits hierarchical output. Join and GroupJoin provide an alternative strategy to Select and SelectMany. The advantage of Join and GroupJoin is that they execute efficiently over local in-memory collections, since they first load the inner sequence into a keyed lookup, avoiding the need to repeatedly enumerate over every inner element. The disadvantage is that they offer the equivalent of inner and left outer joins only; cross joins and non-equi joins must still be done with Select/SelectMany. With LINQ to SQL and Entity Framework queries, Join and GroupJoin offer no real benefits over Select and SelectMany. Table 9-1 summarizes the differences between each of the joining strategies. Table 9-1. Joining strategies Strategy Result shape Local query efficiency Inner joins Left outer joins Cross joins Non-equi joins Select + SelectMany Flat Bad Yes Yes Yes Yes Select + Select Nested Bad Yes Yes Yes Yes Join Flat Good Yes - - - GroupJoin Nested Good Yes Yes - - GroupJoin + SelectMany Flat Good Yes Yes - - Join The Join operator performs an inner join, emitting a flat output sequence. Entity Framework hides foreign key fields, so you can’t man- ually join across natural relationships (instead, you can query across association properties, as we described in the previous two sections). The simplest way to demonstrate Join is with LINQ to SQL. The following query lists all customers alongside their purchases, without using an association property: IQueryable query = from c in dataContext.Customers 396 | Chapter 9: LINQ Operators join p in dataContext.Purchases on c.ID equals p.CustomerID select c.Name + " bought a " + p.Description; The results match what we would get from a SelectMany-style query: Tom bought a Bike Tom bought a Holiday Dick bought a Phone Harry bought a Car To see the benefit of Join over SelectMany, we must convert this to a local query. We can demonstrate this by first copying all customers and purchases to arrays, and then querying the arrays: Customer[] customers = dataContext.Customers.ToArray(); Purchase[] purchases = dataContext.Purchases.ToArray(); var slowQuery = from c in customers from p in purchases where c.ID == p.CustomerID select c.Name + " bought a " + p.Description; var fastQuery = from c in customers join p in purchases on c.ID equals p.CustomerID select c.Name + " bought a " + p.Description; Although both queries yield the same results, the Join query is considerably faster because its implementation in Enumerable preloads the inner collection (purchases) into a keyed lookup. The query syntax for join can be written in general terms as follows: join inner-var in inner-sequence on outer-key-expr equals inner-key-expr Join operators in LINQ differentiate between the outer sequence and inner sequence. Syntactically: • The outer sequence is the input sequence (in this case, customers). • The inner sequence is the new collection you introduce (in this case, purchases). Join performs inner joins, meaning customers without purchases are excluded from the output. With inner joins, you can swap the inner and outer sequences in the query and still get the same results: from p in purchases // p is now outer join c in customers on p.CustomerID equals c.ID // c is now inner ... You can add further join clauses to the same query. If each purchase, for instance, has one or more purchase items, you could join the purchase items as follows: from c in customers join p in purchases on c.ID equals p.CustomerID // first join join pi in purchaseItems on p.ID equals pi.PurchaseID // second join ... Joining | 397 LINQ Operators purchases acts as the inner sequence in the first join and as the outer sequence in the second join. You could obtain the same results (inefficiently) using nested foreach statements as follows: foreach (Customer c in customers) foreach (Purchase p in purchases) if (c.ID == p.CustomerID) foreach (PurchaseItem pi in purchaseItems) if (p.ID == pi.PurchaseID) Console.WriteLine (c.Name + "," + p.Price + "," + pi.Detail); In query syntax, variables from earlier joins remain in scope—just as they do with SelectMany-style queries. You’re also permitted to insert where and let clauses in between join clauses. Joining on multiple keys You can join on multiple keys with anonymous types as follows: from x in sequenceX join y in sequenceY on new { K1 = x.Prop1, K2 = x.Prop2 } equals new { K1 = y.Prop3, K2 = y.Prop4 } ... For this to work, the two anonymous types must be structured identically. The compiler then implements each with the same internal type, making the joining keys compatible. Joining in fluent syntax The following query syntax join: from c in customers join p in purchases on c.ID equals p.CustomerID select new { c.Name, p.Description, p.Price }; in fluent syntax is as follows: customers.Join ( // outer collection purchases, // inner collection c => c.ID, // outer key selector p => p.CustomerID, // inner key selector (c, p) => new { c.Name, p.Description, p.Price } // result selector ); The result selector expression at the end creates each element in the output sequence. If you have additional clauses prior to projecting, such as orderby in this example: from c in customers join p in purchases on c.ID equals p.CustomerID orderby p.Price select c.Name + " bought a " + p.Description; 398 | Chapter 9: LINQ Operators you must manufacture a temporary anonymous type in the result selector in fluent syntax. This keeps both c and p in scope following the join: customers.Join ( // outer collection purchases, // inner collection c => c.ID, // outer key selector p => p.CustomerID, // inner key selector (c, p) => new { c, p } ) // result selector .OrderBy (x => x.p.Price) .Select (x => x.c.Name + " bought a " + x.p.Description); Query syntax is usually preferable when joining; it’s less fiddly. GroupJoin GroupJoin does the same work as Join, but instead of yielding a flat result, it yields a hierarchical result, grouped by each outer element. It also allows left outer joins. The query syntax for GroupJoin is the same as for Join, but is followed by the into keyword. Here’s the most basic example: IEnumerable> query = from c in customers join p in purchases on c.ID equals p.CustomerID into custPurchases select custPurchases; // custPurchases is a sequence An into clause translates to GroupJoin only when it appears directly after a join clause. After a select or group clause, it means query continuation. The two uses of the into keyword are quite different, although they have one feature in common: they both introduce a new range variable. The result is a sequence of sequences, which we could enumerate as follows: foreach (IEnumerable purchaseSequence in query) foreach (Purchase p in purchaseSequence) Console.WriteLine (p.Description); This isn’t very useful, however, because purchaseSequence has no reference to the customer. More commonly, you’d do this: from c in customers join p in purchases on c.ID equals p.CustomerID into custPurchases select new { CustName = c.Name, custPurchases }; This gives the same results as the following (inefficient) Select subquery: from c in customers select new { CustName = c.Name, Joining | 399 LINQ Operators custPurchases = purchases.Where (p => c.ID == p.CustomerID) }; By default, GroupJoin does the equivalent of a left outer join. To get an inner join— where customers without purchases are excluded—you need to filter on custPurchases: from c in customers join p in purchases on c.ID equals p.CustomerID into custPurchases where custPurchases.Any() select ... Clauses after a group-join into operate on subsequences of inner child elements, not individual child elements. This means that to filter individual purchases, you’d have to call Where before joining: from c in customers join p in purchases.Where (p2 => p2.Price > 1000) on c.ID equals p.CustomerID into custPurchases ... You can construct lambda queries with GroupJoin as you would with Join. Flat outer joins You run into a dilemma if you want both an outer join and a flat result set. Group Join gives you the outer join; Join gives you the flat result set. The solution is to first call GroupJoin, and then DefaultIfEmpty on each child sequence, and then finally SelectMany on the result: from c in customers join p in purchases on c.ID equals p.CustomerID into custPurchases from cp in custPurchases.DefaultIfEmpty() select new { CustName = c.Name, Price = cp == null ? (decimal?) null : cp.Price }; DefaultIfEmpty emits a null value if a subsequence of purchases is empty. The second from clause translates to SelectMany. In this role, it expands and flattens all the pur- chase subsequences, concatenating them into a single sequence of purchase elements. Joining with lookups The Join and GroupJoin methods in Enumerable work in two steps. First, they load the inner sequence into a lookup. Second, they query the outer sequence in combi- nation with the lookup. A lookup is a sequence of groupings that can be accessed directly by key. Another way to think of it is as a dictionary of sequences—a dictionary that can accept many elements under each key (sometimes called a multi-dictionary). 400 | Chapter 9: LINQ Operators Lookups are read-only and defined by the following interface: public interface ILookup : IEnumerable>, IEnumerable { int Count { get; } bool Contains (TKey key); IEnumerable this [TKey key] { get; } } The joining operators—like other sequence-emitting operators —honor deferred or lazy execution semantics. This means the lookup is not built until you begin enumerating the output sequence (and then the entire lookup is built right then). You can create and query lookups manually as an alternative strategy to using the joining operators, when dealing with local collections. There are a couple of benefits in doing so: • You can reuse the same lookup over multiple queries—as well as in ordinary imperative code. • Querying a lookup is an excellent way of understanding how Join and Group Join work. The ToLookup extension method creates a lookup. The following loads all purchases into a lookup—keyed by their CustomerID: ILookup purchLookup = purchases.ToLookup (p => p.CustomerID, p => p); The first argument selects the key; the second argument selects the objects that are to be loaded as values into the lookup. Reading a lookup is rather like reading a dictionary, except that the indexer returns a sequence of matching items, rather than a single matching item. The following enumerates all purchases made by the customer whose ID is 1: foreach (Purchase p in purchLookup [1]) Console.WriteLine (p.Description); With a lookup in place, you can write SelectMany/Select queries that execute as efficiently as Join/GroupJoin queries. Join is equivalent to using SelectMany on a lookup: from c in customers from p in purchLookup [c.ID] select new { c.Name, p.Description, p.Price }; Tom Bike 500 Tom Holiday 2000 Dick Bike 600 Dick Phone 300 ... Joining | 401 LINQ Operators Adding a call to DefaultIfEmpty makes this into an outer join: from c in customers from p in purchLookup [c.ID].DefaultIfEmpty() select new { c.Name, Descript = p == null ? null : p.Description, Price = p == null ? (decimal?) null : p.Price }; GroupJoin is equivalent to reading the lookup inside a projection: from c in customers select new { CustName = c.Name, CustPurchases = purchLookup [c.ID] }; Enumerable implementations Here’s the simplest valid implementation of Enumerable.Join, null checking aside: public static IEnumerable Join ( this IEnumerable outer, IEnumerable inner, Func outerKeySelector, Func innerKeySelector, Func resultSelector) { ILookup lookup = inner.ToLookup (innerKeySelector); return from outerItem in outer from innerItem in lookup [outerKeySelector (outerItem)] select resultSelector (outerItem, innerItem); } GroupJoin’s implementation is like that of Join, but simpler: public static IEnumerable GroupJoin ( this IEnumerable outer, IEnumerable inner, Func outerKeySelector, Func innerKeySelector, Func ,TResult> resultSelector) { ILookup lookup = inner.ToLookup (innerKeySelector); return from outerItem in outer select resultSelector (outerItem, lookup [outerKeySelector (outerItem)]); } 402 | Chapter 9: LINQ Operators The Zip Operator IEnumerable, IEnumerable→ IEnumerable The Zip operator was added in Framework 4.0. It enumerates two sequences in step (like a zipper), returning a sequence based on applying a function over each element pair. For instance, the following: int[] numbers = { 3, 5, 7 }; string[] words = { "three", "five", "seven", "ignored" }; IEnumerable zip = numbers.Zip (words, (n, w) => n + "=" + w); produces a sequence with the following elements: 3=three 5=five 7=seven Extra elements in either input sequence are ignored. Zip is not supported when querying a database. Ordering IEnumerable→ IOrderedEnumerable Method Description SQL equivalents OrderBy, ThenBy Sorts a sequence in ascending order ORDER BY ... OrderByDescending, ThenByDescending Sorts a sequence in descending order ORDER BY ... DESC Reverse Returns a sequence in reverse order Exception thrown Ordering operators return the same elements in a different order. OrderBy, OrderByDescending, ThenBy, and ThenByDescending OrderBy and OrderByDescending arguments Argument Type Input sequence IEnumerable Key selector TSource => TKey Return type = IOrderedEnumerable ThenBy and ThenByDescending arguments Argument Type Input sequence IOrderedEnumerable Key selector TSource => TKey Ordering | 403 LINQ Operators Query syntax orderby expression1 [descending] [, expression2 [descending] ... ] Overview OrderBy returns a sorted version of the input sequence, using the keySelector expression to make comparisons. The following query emits a sequence of names in alphabetical order: IEnumerable query = names.OrderBy (s => s); The following sorts names by length: IEnumerable query = names.OrderBy (s => s.Length); // Result: { "Jay", "Tom", "Mary", "Dick", "Harry" }; The relative order of elements with the same sorting key (in this case, Jay/Tom and Mary/Dick) is indeterminate—unless you append a ThenBy operator: IEnumerable query = names.OrderBy (s => s.Length).ThenBy (s => s); // Result: { "Jay", "Tom", "Dick", "Mary", "Harry" }; ThenBy reorders only elements that had the same sorting key in the preceding sort. You can chain any number of ThenBy operators. The following sorts first by length, then by the second character, and finally by the first character: names.OrderBy (s => s.Length).ThenBy (s => s[1]).ThenBy (s => s[0]); The equivalent in query syntax is this: from s in names orderby s.Length, s[1], s[0] select s; The following variation is incorrect—it will actually order first by s[1], and then by s.Length (or in the case of a database query, it will order only by s[1] and discard the former ordering): from s in names orderby s.Length orderby s[1] ... LINQ also provides OrderByDescending and ThenByDescending operators, which do the same things, emitting the results in reverse order. The following LINQ-to-db query retrieves purchases in descending order of price, with those of the same price listed alphabetically: dataContext.Purchases.OrderByDescending (p => p.Price) .ThenBy (p => p.Description); 404 | Chapter 9: LINQ Operators In query syntax: from p in dataContext.Purchases orderby p.Price descending, p.Description select p; Comparers and collations In a local query, the key selector objects themselves determine the ordering algorithm via their default IComparable implementation (see Chapter 7). You can override the sorting algorithm by passing in an IComparer object. The following performs a case- insensitive sort: names.OrderBy (n => n, StringComparer.CurrentCultureIgnoreCase); Passing in a comparer is not supported in query syntax, nor in any way by LINQ to SQL or EF. When querying a database, the comparison algorithm is determined by the participating column’s collation. If the collation is case-sensitive, you can request a case-insensitive sort by calling ToUpper in the key selector: from p in dataContext.Purchases orderby p.Description.ToUpper() select p; IOrderedEnumerable and IOrderedQueryable The ordering operators return special subtypes of IEnumerable. Those in Enumera ble return IOrderedEnumerable; those in Queryable return IOrderedQueryable. These subtypes allow a subsequent ThenBy operator to refine rather than replace the existing ordering. The additional members that these subtypes define are not publicly exposed, so they present like ordinary sequences. The fact that they are different types comes into play when building queries progressively: IOrderedEnumerable query1 = names.OrderBy (s => s.Length); IOrderedEnumerable query2 = query1.ThenBy (s => s); If we instead declare query1 of type IEnumerable, the second line would not compile—ThenBy requires an input of type IOrderedEnumerable. You can avoid worrying about this by implicitly typing range variables: var query1 = names.OrderBy (s => s.Length); var query2 = query1.ThenBy (s => s); Implicit typing can create problems of its own, though. The following will not compile: var query = names.OrderBy (s => s.Length); query = query.Where (n => n.Length > 3); // Compile-time error The compiler infers query to be of type IOrderedEnumerable, based on OrderBy’s output sequence type. However, the Where on the next line returns an ordinary IEnumerable, which cannot be assigned back to query. You can work around this either with explicit typing or by calling AsEnumerable() after OrderBy: Ordering | 405 LINQ Operators var query = names.OrderBy (s => s.Length).AsEnumerable(); query = query.Where (n => n.Length > 3); // OK The equivalent in interpreted queries is to call AsQueryable. Grouping IEnumerable→ IEnumerable> Method Description SQL equivalents GroupBy Groups a sequence into subsequences GROUP BY GroupBy Argument Type Input sequence IEnumerable Key selector TSource => TKey Element selector (optional) TSource => TElement Comparer (optional) IEqualityComparer Query syntax group element-expression by key-expression Overview GroupBy organizes a flat input sequence into sequences of groups. For example, the following organizes all the files in c:\temp by extension: string[] files = Directory.GetFiles ("c:\\temp"); IEnumerable> query = files.GroupBy (file => Path.GetExtension (file)); Or if you’re comfortable with implicit typing: var query = files.GroupBy (file => Path.GetExtension (file)); Here’s how to enumerate the result: foreach (IGrouping grouping in query) { Console.WriteLine ("Extension: " + grouping.Key); foreach (string filename in grouping) Console.WriteLine (" - " + filename); } Extension: .pdf -- chapter03.pdf -- chapter04.pdf Extension: .doc -- todo.doc 406 | Chapter 9: LINQ Operators -- menu.doc -- Copy of menu.doc ... Enumerable.GroupBy works by reading the input elements into a temporary dictionary of lists so that all elements with the same key end up in the same sublist. It then emits a sequence of groupings. A grouping is a sequence with a Key property: public interface IGrouping : IEnumerable, IEnumerable { TKey Key { get; } // Key applies to the subsequence as a whole } By default, the elements in each grouping are untransformed input elements, unless you specify an elementSelector argument. The following projects each input element to uppercase: files.GroupBy (file => Path.GetExtension (file), file => file.ToUpper()); An elementSelector is independent of the keySelector. In our case, this means that the Key on each grouping is still in its original case: Extension: .pdf -- CHAPTER03.PDF -- CHAPTER04.PDF Extension: .doc -- TODO.DOC Note that the subcollections are not emitted in alphabetical order of key. GroupBy groups only; it does not sort; in fact, it preserves the original ordering. To sort, you must add an OrderBy operator: files.GroupBy (file => Path.GetExtension (file), file => file.ToUpper()) .OrderBy (grouping => grouping.Key); GroupBy has a simple and direct translation in query syntax: group element-expr by key-expr Here’s our example in query syntax: from file in files group file.ToUpper() by Path.GetExtension (file); As with select, group “ends” a query—unless you add a query continuation clause: from file in files group file.ToUpper() by Path.GetExtension (file) into grouping orderby grouping.Key select grouping; Query continuations are often useful in a group by query. The next query filters out groups that have fewer than five files in them: from file in files group file.ToUpper() by Path.GetExtension (file) into grouping where grouping.Count() >= 5 select grouping; Grouping | 407 LINQ Operators A where after a group by is equivalent to HAVING in SQL. It applies to each subsequence or grouping as a whole, rather than the individual elements. Sometimes you’re interested purely in the result of an aggregation on a grouping and so can abandon the subsequences: string[] votes = { "Bush", "Gore", "Gore", "Bush", "Bush" }; IEnumerable query = from vote in votes group vote by vote into g orderby g.Count() descending select g.Key; string winner = query.First(); // Bush GroupBy in LINQ to SQL and EF Grouping works in the same way when querying a database. If you have association properties set up, you’ll find, however, that the need to group arises less frequently than with standard SQL. For instance, to select customers with at least two pur- chases, you don’t need to group; the following query does the job nicely: from c in dataContext.Customers where c.Purchases.Count >= 2 select c.Name + " has made " + c.Purchases.Count + " purchases"; An example of when you might use grouping is to list total sales by year: from p in dataContext.Purchases group p.Price by p.Date.Year into salesByYear select new { Year = salesByYear.Key, TotalValue = salesByYear.Sum() }; LINQ’s grouping operators expose a superset of SQL’s “GROUP BY” functionality. Another departure from traditional SQL comes in there being no obligation to project the variables or expressions used in grouping or sorting. Grouping by multiple keys You can group by a composite key, using an anonymous type: from n in names group n by new { FirstLetter = n[0], Length = n.Length }; Custom equality comparers You can pass a custom equality comparer into GroupBy, in a local query, to change the algorithm for key comparison. Rarely is this required, though, because changing the key selector expression is usually sufficient. 408 | Chapter 9: LINQ Operators For instance, the following creates a case-insensitive grouping: group name by name.ToUpper() Set Operators IEnumerable, IEnumerable→IEnumerable Method Description SQL equivalents Concat Returns a concatenation of elements in each of the two sequences UNION ALL Union Returns a concatenation of elements in each of the two sequences, excluding duplicates UNION Intersect Returns elements present in both sequences WHERE ... IN (...) Except Returns elements present in the first, but not the second sequence EXCEPT or WHERE ... NOT IN (...) Concat and Union Concat returns all the elements of the first sequence, followed by all the elements of the second. Union does the same, but removes any duplicates: int[] seq1 = { 1, 2, 3 }, seq2 = { 3, 4, 5 }; IEnumerable concat = seq1.Concat (seq2), // { 1, 2, 3, 3, 4, 5 } union = seq1.Union (seq2); // { 1, 2, 3, 4, 5 } Specifying the type argument explicitly is useful when the sequences are differently typed, but the elements have a common base type. For instance, with the reflection API (Chapter 19), methods and properties are represented with MethodInfo and PropertyInfo classes, which have a common base class called MemberInfo. We can concatenate methods and properties by stating that base class explicitly when calling Concat: MethodInfo[] methods = typeof (string).GetMethods(); PropertyInfo[] props = typeof (string).GetProperties(); IEnumerable both = methods.Concat (props); In the next example, we filter the methods before concatenating: var methods = typeof (string).GetMethods().Where (m => !m.IsSpecialName); var props = typeof (string).GetProperties(); var both = methods.Concat (props); This example relies on interface type parameter variance: methods is of type IEnumer able, which requires a covariant conversion to IEnumerable. It’s a good illustration of how variance makes things work more as you’d expect. Set Operators | 409 LINQ Operators Intersect and Except Intersect returns the elements that two sequences have in common. Except returns the elements in the first input sequence that are not present in the second: int[] seq1 = { 1, 2, 3 }, seq2 = { 3, 4, 5 }; IEnumerable commonality = seq1.Intersect (seq2), // { 3 } difference1 = seq1.Except (seq2), // { 1, 2 } difference2 = seq2.Except (seq1); // { 4, 5 } Enumerable.Except works internally by loading all of the elements in the first collec- tion into a dictionary, then removing from the dictionary all elements present in the second sequence. The equivalent in SQL is a NOT EXISTS or NOT IN subquery: SELECT number FROM numbers1Table WHERE number NOT IN (SELECT number FROM numbers2Table) Conversion Methods LINQ deals primarily in sequences—in other words, collections of type IEnumera ble. The conversion methods convert to and from other types of collections: Method Description OfType Converts IEnumerable to IEnumerable, discarding wrongly typed elements Cast Converts IEnumerable to IEnumerable, throwing an exception if there are any wrongly typed elements ToArray Converts IEnumerable to T[] ToList Converts IEnumerable to List ToDictionary Converts IEnumerable to Dictionary ToLookup Converts IEnumerable to ILookup AsEnumerable Downcasts to IEnumerable AsQueryable Casts or converts to IQueryable OfType and Cast OfType and Cast accept a nongeneric IEnumerable collection and emit a generic IEnumerable sequence that you can subsequently query: ArrayList classicList = new ArrayList(); // in System.Collections classicList.AddRange ( new int[] { 3, 4, 5 } ); IEnumerable sequence1 = classicList.Cast(); Cast and OfType differ in their behavior when encountering an input element that’s of an incompatible type. Cast throws an exception; OfType ignores the incompatible element. Continuing the preceding example: DateTime offender = DateTime.Now; classicList.Add (offender); 410 | Chapter 9: LINQ Operators IEnumerable sequence2 = classicList.OfType(), // OK - ignores offending DateTime sequence3 = classicList.Cast(); // Throws exception The rules for element compatibility exactly follow those of C#’s is operator, and therefore consider only reference conversions and unboxing conversions. We can see this by examining the internal implementation of OfType: public static IEnumerable OfType (IEnumerable source) { foreach (object element in source) if (element is TSource) yield return (TSource)element; } Cast has an identical implementation, except that it omits the type compatibility test: public static IEnumerable Cast (IEnumerable source) { foreach (object element in source) yield return (TSource)element; } A consequence of these implementations is that you cannot use Cast to perform numeric or custom conversions (for these, you must perform a Select operation instead). In other words, Cast is not as flexible as C#’s cast operator: int i = 3; long l = i; // Implicit numeric conversion int->long int i2 = (int) l; // Explicit numeric conversion long->int We can demonstrate this by attempting to use OfType or Cast to convert a sequence of ints to a sequence of longs: int[] integers = { 1, 2, 3 }; IEnumerable test1 = integers.OfType(); IEnumerable test2 = integers.Cast(); When enumerated, test1 emits zero elements and test2 throws an exception. Examining OfType’s implementation, it’s fairly clear why. After substituting TSource, we get the following expression: (element is long) which returns false for an int element, due to the lack of an inheritance relationship. Conversion Methods | 411 LINQ Operators The reason for test2 throwing an exception, when enumerated, is subtler. Notice in Cast’s implementation that element is of type object. When TSource is a value type, the CLR assumes this is an unboxing conversion, and synthesizes a method that reproduces the scenario described in the section “Boxing and Unboxing” on page 90 in Chapter 3: int value = 123; object element = value; long result = (long) element; // exception Because the element variable is declared of type object, an object-to-long cast is performed (an unboxing) rather than an int-to-long numeric conversion. Unboxing operations require an exact type match, so the object-to-long unbox fails when given an int. As we suggested previously, the solution is to use an ordinary Select: IEnumerable castLong = integers.Select (s => (long) s); OfType and Cast are also useful in downcasting elements in a generic input sequence. For instance, if you have an input sequence of type IEnumerable, OfType would return just the apples. This is particularly useful in LINQ to XML (see Chapter 10). Cast has query syntax support: simply precede the range variable with a type: from TreeNode node in myTreeView.Nodes ... ToArray, ToList, ToDictionary, and ToLookup ToArray and ToList emit the results into an array or generic list. These operators force the immediate enumeration of the input sequence (unless indirected via a sub- query or expression tree). For examples, refer to the section “Deferred Execu- tion” on page 332 in Chapter 8. ToDictionary and ToLookup accept the following arguments: Argument Type Input sequence IEnumerable Key selector TSource => TKey Element selector (optional) TSource => TElement Comparer (optional) IEqualityComparer ToDictionary also forces immediate execution of a sequence, writing the results to a generic Dictionary. The keySelector expression you provide must evaluate to a unique value for each element in the input sequence; otherwise, an exception is 412 | Chapter 9: LINQ Operators thrown. In contrast, ToLookup allows many elements of the same key. We describe lookups in the earlier section “Joining with lookups” on page 400. AsEnumerable and AsQueryable AsEnumerable upcasts a sequence to IEnumerable, forcing the compiler to bind subsequent query operators to methods in Enumerable, instead of Queryable. For an example, see the section “Combining Interpreted and Local Queries” on page 352 in Chapter 8. AsQueryable downcasts a sequence to IQueryable if it implements that interface. Otherwise, it instantiates an IQueryable wrapper over the local query. Element Operators IEnumerable→ TSource Method Description SQL equivalents First, FirstOrDefault Returns the first element in the sequence, optionally satisfying a predicate SELECT TOP 1 ... ORDER BY ... Last, LastOrDefault Returns the last element in the sequence, optionally satisfying a predicate SELECT TOP 1 ... ORDER BY ... DESC Single, SingleOrDefault Equivalent to First/First OrDefault, but throws an ex- ception if there is more than one match ElementAt, ElementAtOrDefault Returns the element at the specified position Exception thrown DefaultIfEmpty Returns a single-element se- quence whose value is default(TSource) if the sequence has no elements OUTER JOIN Methods ending in “OrDefault” return default(TSource) rather than throwing an exception if the input sequence is empty or if no elements match the supplied predicate. default(TSource) is null for reference type elements, false for the bool type and zero for numeric types. First, Last, and Single Argument Type Source sequence IEnumerable Predicate (optional) TSource => bool Element Operators | 413 LINQ Operators The following example demonstrates First and Last: int[] numbers = { 1, 2, 3, 4, 5 }; int first = numbers.First(); // 1 int last = numbers.Last(); // 5 int firstEven = numbers.First (n => n % 2 == 0); // 2 int lastEven = numbers.Last (n => n % 2 == 0); // 4 The following demonstrates First versus FirstOrDefault: int firstBigError = numbers.First (n => n > 10); // Exception int firstBigNumber = numbers.FirstOrDefault (n => n > 10); // 0 To avoid an exception, Single requires exactly one matching element; SingleOrDe fault requires one or zero matching elements: int onlyDivBy3 = numbers.Single (n => n % 3 == 0); // 3 int divBy2Err = numbers.Single (n => n % 2 == 0); // Error: 2 & 4 match int singleError = numbers.Single (n => n > 10); // Error int noMatches = numbers.SingleOrDefault (n => n > 10); // 0 int divBy2Error = numbers.SingleOrDefault (n => n % 2 == 0); // Error Single is the “fussiest” in this family of element operators. FirstOrDefault and LastOrDefault are the most tolerant. In LINQ to SQL and EF, Single is often used to retrieve a row from a table by primary key: Customer cust = dataContext.Customers.Single (c => c.ID == 3); ElementAt Argument Type Source sequence IEnumerable Index of element to return int ElementAt picks the nth element from the sequence: int[] numbers = { 1, 2, 3, 4, 5 }; int third = numbers.ElementAt (2); // 3 int tenthError = numbers.ElementAt (9); // Exception int tenth = numbers.ElementAtOrDefault (9); // 0 Enumerable.ElementAt is written such that if the input sequence happens to imple- ment IList, it calls IList’s indexer. Otherwise, it enumerates n times, and then returns the next element. ElementAt is not supported in LINQ to SQL or EF. DefaultIfEmpty DefaultIfEmpty returns a sequence containing a single element whose value is default(TSource) if the input sequence has no elements. Otherwise it returns the input sequence unchanged. This is used in writing flat outer joins: see the earlier 414 | Chapter 9: LINQ Operators sections “Outer joins with SelectMany” on page 393 and “Flat outer joins” on page 400. Aggregation Methods IEnumerable→ scalar Method Description SQL equivalents Count, LongCount Returns the number of elements in the input sequence, optionally satisfying a predicate COUNT (...) Min, Max Returns the smallest or largest element in the sequence MIN (...), MAX (...) Sum, Average Calculates a numeric sum or average over elements in the sequence SUM (...), AVG (...) Aggregate Performs a custom aggregation Exception thrown Count and LongCount Argument Type Source sequence IEnumerable Predicate (optional) TSource => bool Count simply enumerates over a sequence, returning the number of items: int fullCount = new int[] { 5, 6, 7 }.Count(); // 3 The internal implementation of Enumerable.Count tests the input sequence to see whether it happens to implement ICollection. If it does, it simply calls ICollec tion.Count. Otherwise, it enumerates over every item, incrementing a counter. You can optionally supply a predicate: int digitCount = "pa55w0rd".Count (c => char.IsDigit (c)); // 3 LongCount does the same job as Count, but returns a 64-bit integer, allowing for sequences of greater than 2 billion elements. Min and Max Argument Type Source sequence IEnumerable Result selector (optional) TSource => TResult Min and Max return the smallest or largest element from a sequence: int[] numbers = { 28, 32, 14 }; int smallest = numbers.Min(); // 14; int largest = numbers.Max(); // 32; Aggregation Methods | 415 LINQ Operators If you include a selector expression, each element is first projected: int smallest = numbers.Max (n => n % 10); // 8; A selector expression is mandatory if the items themselves are not intrinsically comparable—in other words, if they do not implement IComparable: Purchase runtimeError = dataContext.Purchases.Min (); // Error decimal? lowestPrice = dataContext.Purchases.Min (p => p.Price); // OK A selector expression determines not only how elements are compared, but also the final result. In the preceding example, the final result is a decimal value, not a purchase object. To get the cheapest purchase, you need a subquery: Purchase cheapest = dataContext.Purchases .Where (p => p.Price == dataContext.Purchases.Min (p2 => p2.Price)) .FirstOrDefault(); In this case, you could also formulate the query without an aggregation—using an OrderBy followed by FirstOrDefault. Sum and Average Argument Type Source sequence IEnumerable Result selector (optional) TSource => TResult Sum and Average are aggregation operators that are used in a similar manner to Min and Max: decimal[] numbers = { 3, 4, 8 }; decimal sumTotal = numbers.Sum(); // 15 decimal average = numbers.Average(); // 5 (mean value) The following returns the total length of each of the strings in the names array: int combinedLength = names.Sum (s => s.Length); // 19 Sum and Average are fairly restrictive in their typing. Their definitions are hard-wired to each of the numeric types (int, long, float, double, decimal, and their nullable versions). In contrast, Min and Max can operate directly on anything that implements IComparable—such as a string, for instance. Further, Average always returns either decimal or double, according to the following table: Selector type Result type decimal decimal int, long, float, double double This means the following does not compile (“cannot convert double to int”): int avg = new int[] { 3, 4 }.Average(); 416 | Chapter 9: LINQ Operators But this will compile: double avg = new int[] { 3, 4 }.Average(); // 3.5 Average implicitly upscales the input values to avoid loss of precision. In this exam- ple, we averaged integers and got 3.5, without needing to resort to an input element cast: double avg = numbers.Average (n => (double) n); When querying a database, Sum and Average translate to the standard SQL aggrega- tions. The following query returns customers whose average purchase was more than $500: from c in dataContext.Customers where c.Purchases.Average (p => p.Price) > 500 select c.Name; Aggregate Aggregate allows you to specify a custom accumulation algorithm for implementing unusual aggregations. Aggregate is not supported in LINQ to SQL or Entity Frame- work, and is somewhat specialized in its use cases. The following demonstrates how Aggregate can do the work of Sum: int[] numbers = { 2, 3, 4 }; int sum = numbers.Aggregate (0, (total, n) => total + n); // 9 The first argument to Aggregate is the seed, from which accumulation starts. The second argument is an expression to update the accumulated value, given a fresh element. You can optionally supply a third argument to project the final result value from the accumulated value. Most problems for which Aggregate has been designed can be solved as easily with a foreach loop—and with more familiar syntax. The advantage of using Aggregate is that with large or complex aggregations, you can automatically parallelize the operation with PLINQ (see Chapter 23). Unseeded aggregations You can omit the seed value when calling Aggregate, in which case the first element becomes the implicit seed, and aggregation proceeds from the second element. Here’s the preceding example, unseeded: int[] numbers = { 1, 2, 3 }; int sum = numbers.Aggregate ((total, n) => total + n); // 6 This gives the same result as before, but we’re actually doing a different calcula- tion. Before, we were calculating 0+1+2+3; now we’re calculating 1+2+3. We can better illustrate the difference by multiplying instead of adding: Aggregation Methods | 417 LINQ Operators int[] numbers = { 1, 2, 3 }; int x = numbers.Aggregate (0, (prod, n) => prod * n); // 0*1*2*3 = 0 int y = numbers.Aggregate ( (prod, n) => prod * n); // 1*2*3 = 6 As we’ll see in Chapter 23, unseeded aggregations have the advantage of being par- allelizable without requiring the use of special overloads. However, there are some traps with unseeded aggregations. Traps with unseeded aggregations The unseeded aggregation methods are intended for use with delegates that are commutative and associative. If used otherwise, the result is either unintuitive (with ordinary queries) or nondeterministic (in the case that you parallelize the query with PLINQ). For example, consider the following function: (total, n) => total + n * n This is neither commutative nor associative. (For example, 1+2*2 != 2+1*1). Let’s see what happens when we use it to sum the square of the numbers 2, 3, and 4: int[] numbers = { 2, 3, 4 }; int sum = numbers.Aggregate ((total, n) => total + n * n); // 27 Instead of calculating: 2*2 + 3*3 + 4*4 // 29 it calculates: 2 + 3*3 + 4*4 // 27 We can fix this in a number of ways. First, we could include 0 as the first element: int[] numbers = { 0, 2, 3, 4 }; Not only is this inelegant, but it will still give incorrect results if parallelized— because PLINQ leverages the function’s assumed associativity by selecting multi- ple elements as seeds. To illustrate, if we denote our aggregation function as follows: f(total, n) => total + n * n then LINQ to Objects would calculate this: f(f(f(0, 2),3),4) whereas PLINQ may do this: f(f(0,2),f(3,4)) with the following result: First partition: a = 0 + 2*2 (= 4) Second partition: b = 3 + 4*4 (= 19) Final result: a + b*b (= 365) OR EVEN: b + a*a (= 35) There are two good solutions. The first is to turn this into a seeded aggregation— with zero as the seed. The only complication is that with PLINQ, we’d need to use a special overload in order for the query not to execute sequentially (see “Optimizing PLINQ” on page 926 in Chapter 23). 418 | Chapter 9: LINQ Operators The second solution is to restructure the query such that the aggregation function is commutative and associative: int sum = numbers.Select (n => n * n).Aggregate ((total, n) => total + n); Of course, in such simple scenarios you can (and should) use the Sum operator instead of Aggregate: int sum = numbers.Sum (n => n * n); You can actually go quite far just with Sum and Average. For instance, you can use Average to calculate a root-mean-square: Math.Sqrt (numbers.Average (n => n * n)) and even standard deviation: double mean = numbers.Average(); double sdev = Math.Sqrt (numbers.Average (n => { double dif = n - mean; return dif * dif; })); Both are safe, efficient and fully parallelizable. In Chapter 23, we’ll give a practical example of a custom aggregation that can’t be reduced to Sum or Average. Quantifiers IEnumerable→ bool Method Description SQL equivalents Contains Returns true if the input sequence contains the given element WHERE ... IN (...) Any Returns true if any elements satisfy the given predicate WHERE ... IN (...) All Returns true if all elements satisfy the given predicate WHERE (...) SequenceEqual Returns true if the second sequence has identical elements to the input sequence Contains and Any The Contains method accepts an argument of type TSource; Any accepts an optional predicate. Contains returns true if the given element is present: bool hasAThree = new int[] { 2, 3, 4 }.Contains (3); // true; Any returns true if the given expression is true for at least one element. We can rewrite the preceding query with Any as follows: bool hasAThree = new int[] { 2, 3, 4 }.Any (n => n == 3); // true; Quantifiers | 419 LINQ Operators Any can do everything that Contains can do, and more: bool hasABigNumber = new int[] { 2, 3, 4 }.Any (n => n > 10); // false; Calling Any without a predicate returns true if the sequence has one or more ele- ments. Here’s another way to write the preceding query: bool hasABigNumber = new int[] { 2, 3, 4 }.Where (n => n > 10).Any(); Any is particularly useful in subqueries and is used often when querying databases, for example: from c in dataContext.Customers where c.Purchases.Any (p => p.Price > 1000) select c All and SequenceEqual All returns true if all elements satisfy a predicate. The following returns customers whose purchases are less than $100: dataContext.Customers.Where (c => c.Purchases.All (p => p.Price < 100)); SequenceEqual compares two sequences. To return true, each sequence must have identical elements, in the identical order. Generation Methods void→IEnumerable Method Description Empty Creates an empty sequence Repeat Creates a sequence of repeating elements Range Creates a sequence of integers Empty, Repeat, and Range are static (nonextension) methods that manufacture simple local sequences. Empty Empty manufactures an empty sequence and requires just a type argument: foreach (string s in Enumerable.Empty()) Console.Write (s); // In conjunction with the ?? operator, Empty does the reverse of DefaultIfEmpty. For example, suppose we have a jagged array of integers, and we want to get all the integers into a single flat list. The following SelectMany query fails if any of the inner arrays is null: int[][] numbers = { new int[] { 1, 2, 3 }, new int[] { 4, 5, 6 }, 420 | Chapter 9: LINQ Operators null // this null makes the query below fail. }; IEnumerable flat = numbers.SelectMany (innerArray => innerArray); Empty in conjunction with ?? fixes the problem: IEnumerable flat = numbers .SelectMany (innerArray => innerArray ?? Enumerable.Empty ()); foreach (int i in flat) Console.Write (i + " "); // 1 2 3 4 5 6 Range and Repeat Range and Repeat work only with integers. Range accepts a starting index and count: foreach (int i in Enumerable.Range (5, 3)) Console.Write (i + " "); // 5 6 7 Repeat accepts the number to repeat, and the number of iterations: foreach (int i in Enumerable.Repeat (5, 3)) Console.Write (i + " "); // 5 5 5 Generation Methods | 421 LINQ Operators 10LINQ to XML The .NET Framework provides a number of APIs for working with XML data. From Framework 3.5, the primary choice for general-purpose XML document processing is LINQ to XML. LINQ to XML comprises a lightweight LINQ-friendly XML docu- ment object model, plus a set of supplementary query operators. LINQ to XML is supported fully in the Metro profile. In this chapter, we concentrate entirely on LINQ to XML. In Chapter 11, we cover the more specialized XML types and APIs, including the forward-only reader/writer, the types for working with schemas, stylesheets and XPaths, and the legacy XmlDocument-based DOM. The LINQ to XML DOM is extremely well designed and highly performant. Even without LINQ, the LINQ to XML DOM is valuable as a lightweight façade over the low-level XmlReader and XmlWriter classes. All LINQ to XML types are defined in the System.Xml.Linq namespace. Architectural Overview This section starts with a very brief introduction to the concept of a DOM, and then explains the rationale behind LINQ to XML’s DOM. What Is a DOM? Consider the following XML file: Joe Bloggs 423 As with all XML files, we start with a declaration, and then a root element, whose name is customer. The customer element has two attributes, each with a name (id and status) and value ("123" and "archived"). Within customer, there are two child elements, firstname and lastname, each having simple text content ("Joe" and "Bloggs"). Each of these constructs—declaration, element, attribute, value, and text content —can be represented with a class. And if such classes have collection properties for storing child content, we can assemble a tree of objects to fully describe a document. This is called a document object model, or DOM. The LINQ to XML DOM LINQ to XML comprises two things: • An XML DOM, which we call the X-DOM • A set of about 10 supplementary query operators As you might expect, the X-DOM consists of types such as XDocument, XElement, and XAttribute. Interestingly, the X-DOM types are not tied to LINQ—you can load, instantiate, update, and save an X-DOM without ever writing a LINQ query. Conversely, you could use LINQ to query a DOM created of the older W3C-compliant types. However, this would be frustrating and limiting. The dis- tinguishing feature of the X-DOM is that it’s LINQ-friendly. This means: • It has methods that emit useful IEnumerable sequences, upon which you can query. • Its constructors are designed such that you can build an X-DOM tree through a LINQ projection. X-DOM Overview Figure 10-1 shows the core X-DOM types. The most frequently used of these types is XElement. XObject is the root of the inheritance hierarchy; XElement and XDocu ment are roots of the containership hierarchy. Figure 10-2 shows the X-DOM tree created from the following code: string xml = @" Joe Bloggs "; XElement customer = XElement.Parse (xml); XObject is the abstract base class for all XML content. It defines a link to the Par ent element in the containership tree as well as an optional XDocument. 424 | Chapter 10: LINQ to XML XNode is the base class for most XML content excluding attributes. The distinguishing feature of XNode is that it can sit in an ordered collection of mixed-type XNodes. For instance, consider the following XML: Hello world Figure 10-1. Core X-DOM types Figure 10-2. A simple X-DOM tree X-DOM Overview | 425 LINQ to XML Within the parent element , there’s first an XText node (Hello world), then an XElement node, then an XComment node, and then a second XElement node. In contrast, an XAttribute will tolerate only other XAttributes as peers. Although an XNode can access its parent XElement, it has no concept of child nodes: this is the job of its subclass XContainer. XContainer defines members for dealing with children and is the abstract base class for XElement and XDocument. XElement introduces members for managing attributes—as well as a Name and Value. In the (fairly common) case of an element having a single XText child node, the Value property on XElement encapsulates this child’s content for both get and set operations, cutting unnecessary navigation. Thanks to Value, you can mostly avoid working directly with XText nodes. XDocument represents the root of an XML tree. More precisely, it wraps the root XElement, adding an XDeclaration, processing instructions, and other root-level “fluff.” Unlike with the W3C DOM, its use is optional: you can load, manipulate, and save an X-DOM without ever creating an XDocument! The nonreliance on XDocu ment also means you can efficiently and easily move a node subtree to another X-DOM hierarchy. Loading and Parsing Both XElement and XDocument provide static Load and Parse methods to build an X-DOM tree from an existing source: • Load builds an X-DOM from a file, URI, Stream, TextReader, or XmlReader. • Parse builds an X-DOM from a string. For example: XDocument fromWeb = XDocument.Load ("http://albahari.com/sample.xml"); XElement fromFile = XElement.Load (@"e:\media\somefile.xml"); XElement config = XElement.Parse ( @" 30 "); In later sections, we describe how to traverse and update an X-DOM. As a quick preview, here’s how to manipulate the config element we just populated: foreach (XElement child in config.Elements()) Console.WriteLine (child.Name); // client XElement client = config.Element ("client"); bool enabled = (bool) client.Attribute ("enabled"); // Read attribute 426 | Chapter 10: LINQ to XML Console.WriteLine (enabled); // True client.Attribute ("enabled").SetValue (!enabled); // Update attribute int timeout = (int) client.Element ("timeout"); // Read element Console.WriteLine (timeout); // 30 client.Element ("timeout").SetValue (timeout * 2); // Update element client.Add (new XElement ("retries", 3)); // Add new elememt Console.WriteLine (config); // Implicitly call config.ToString() Here’s the result of that last Console.WriteLine: 60 3 Joe Bloggs A value is optional when constructing an XElement—you can provide just the element name and add content later. Notice that when we did provide a value, a simple string sufficed—we didn’t need to explicitly create and add an XText child node. The X-DOM does this work automatically, so you can deal simply with “values.” Functional Construction In our preceding example, it’s hard to glean the XML structure from the code. X-DOM supports another mode of instantiation, called functional construction (from functional programming). With functional construction, you build an entire tree in a single expression: XElement customer = new XElement ("customer", new XAttribute ("id", 123), new XElement ("firstname", "joe"), new XElement ("lastname", "bloggs", new XComment ("nice name") ) ); This has two benefits. First, the code resembles the shape of the XML. Second, it can be incorporated into the select clause of a LINQ query. For example, the fol- lowing LINQ to SQL query projects directly into an X-DOM: XElement query = new XElement ("customers", from c in dataContext.Customers select new XElement ("customer", new XAttribute ("id", c.ID), new XElement ("firstname", c.FirstName), new XElement ("lastname", c.LastName, new XComment ("nice name") ) ) ); 428 | Chapter 10: LINQ to XML More on this later in this chapter, in the section “Projecting into an X- DOM” on page 450. Specifying Content Functional construction is possible because the constructors for XElement (and XDocument) are overloaded to accept a params object array: public XElement (XName name, params object[] content) The same holds true for the Add method in XContainer: public void Add (params object[] content) Hence, you can specify any number of child objects of any type when building or appending an X-DOM. This works because anything counts as legal content. To see how, we need to examine how each content object is processed internally. Here are the decisions made by XContainer, in order: 1. If the object is null, it’s ignored. 2. If the object is based on XNode or XStreamingElement, it’s added as is to the Nodes collection. 3. If the object is an XAttribute, it’s added to the Attributes collection. 4. If the object is a string, it gets wrapped in an XText node and added to Nodes.1 5. If the object implements IEnumerable, it’s enumerated, and the same rules are applied to each element. 6. Otherwise, the object is converted to a string, wrapped in an XText node, and then added to Nodes.2 Everything ends up in one of two buckets: Nodes or Attributes. Furthermore, any object is valid content because it can always ultimately call ToString on it and treat it as an XText node. Before calling ToString on an arbitrary type, XContainer first tests whether it is one of the following types: float, double, decimal, bool, DateTime, DateTimeOffset, TimeSpan If so, it calls an appropriate typed ToString method on the XmlConvert helper class instead of calling ToString on the object itself. This ensures that the data is round-trippable and com- pliant with standard XML formatting rules. 1. The X-DOM actually optimizes this step internally by storing simple text content in a string. The XTEXT node is not actually created until you call Nodes( ) on the XContainer. 2. See footnote 1. Instantiating an X-DOM | 429 LINQ to XML Automatic Deep Cloning When a node or attribute is added to an element (whether via functional construc- tion or an Add method) the node or attribute’s Parent property is set to that element. A node can have only one parent element: if you add an already parented node to a second parent, the node is automatically deep-cloned. In the following example, each customer has a separate copy of address: var address = new XElement ("address", new XElement ("street", "Lawley St"), new XElement ("town", "North Beach") ); var customer1 = new XElement ("customer1", address); var customer2 = new XElement ("customer2", address); customer1.Element ("address").Element ("street").Value = "Another St"; Console.WriteLine ( customer2.Element ("address").Element ("street").Value); // Lawley St This automatic duplication keeps X-DOM object instantiation free of side effects— another hallmark of functional programming. Navigating and Querying As you might expect, the XNode and XContainer classes define methods and properties for traversing the X-DOM tree. Unlike a conventional DOM, however, these func- tions don’t return a collection that implements IList. Instead, they return either a single value or a sequence that implements IEnumerable—upon which you are then expected to execute a LINQ query (or enumerate with a foreach). This allows for advanced queries as well as simple navigation tasks—using familiar LINQ query syntax. Element and attribute names are case-sensitive in the X-DOM —just as they are in XML. Child Node Navigation Return type Members Works on XNode FirstNode { get; } XContainer LastNode { get; } XContainer IEnumerable Nodes() XContainer* DescendantNodes() XContainer* DescendantNodesAndSelf() XElement* XElement Element (XName) XContainer IEnumerable Elements() XContainer* 430 | Chapter 10: LINQ to XML Return type Members Works on Elements (XName) XContainer* Descendants() XContainer* Descendants (XName) XContainer* DescendantsAndSelf() XElement* DescendantsAndSelf (XName) XElement* bool HasElements { get; } XElement Functions marked with an asterisk in the third column of this and other tables also operate on sequences of the same type. For instance, you can call Nodes on either an XContainer or a sequence of XContainer objects. This is possible because of extension methods defined in System.Xml.Linq—the supple- mentary query operators we talked about in the overview. FirstNode, LastNode, and Nodes FirstNode and LastNode give you direct access to the first or last child node; Nodes returns all children as a sequence. All three functions consider only direct descendants. For example: var bench = new XElement ("bench", new XElement ("toolbox", new XElement ("handtool", "Hammer"), new XElement ("handtool", "Rasp") ), new XElement ("toolbox", new XElement ("handtool", "Saw"), new XElement ("powertool", "Nailgun") ), new XComment ("Be careful with the nailgun") ); foreach (XNode node in bench.Nodes()) Console.WriteLine (node.ToString (SaveOptions.DisableFormatting) + "."); This is the output: HammerRasp. SawNailgun. . Retrieving elements The Elements method returns just the child nodes of type XElement: foreach (XElement e in bench.Elements()) Console.WriteLine (e.Name + "=" + e.Value); // toolbox=HammerRasp // toolbox=SawNailgun Navigating and Querying | 431 LINQ to XML The following LINQ query finds the toolbox with the nail gun: IEnumerable query = from toolbox in bench.Elements() where toolbox.Elements().Any (tool => tool.Value == "Nailgun") select toolbox.Value; RESULT: { "SawNailgun" } The next example uses a SelectMany query to retrieve the hand tools in all toolboxes: IEnumerable query = from toolbox in bench.Elements() from tool in toolbox.Elements() where tool.Name == "handtool" select tool.Value; RESULT: { "Hammer", "Rasp", "Saw" } Elements itself is equivalent to a LINQ query on Nodes. Our pre- ceding query could be started as follows: from toolbox in bench.Nodes().OfType() where ... Elements can also return just the elements of a given name. For example: int x = bench.Elements ("toolbox").Count(); // 2 This is equivalent to: int x = bench.Elements().Where (e => e.Name == "toolbox").Count(); // 2 Elements is also defined as an extension method accepting IEnumerable or, more precisely, it accepts an argument of this type: IEnumerable where T : XContainer This allows it to work with sequences of elements, too. Using this method, we can rewrite the query that finds the hand tools in all toolboxes as follows: from tool in bench.Elements ("toolbox").Elements ("handtool") select tool.Value.ToUpper(); The first call to Elements binds to XContainer’s instance method; the second call to Elements binds to the extension method. Retrieving a single element The method Element (singular) returns the first matching element of the given name. Element is useful for simple navigation, as follows: XElement settings = XElement.Load ("databaseSettings.xml"); string cx = settings.Element ("database").Element ("connectString").Value; 432 | Chapter 10: LINQ to XML Element is equivalent to calling Elements() and then applying LINQ’s FirstOrDe fault query operator with a name-matching predicate. Element returns null if the requested element doesn’t exist. Element("xyz").Value will throw a NullReferenceException if element xyz does not exist. If you’d prefer a null rather than an exception, cast the XElement to a string instead of querying its Value property. In other words: string xyz = (string) settings.Element ("xyz"); This works because XElement defines an explicit string conver- sion—just for this purpose! Retrieving descendants XContainer also provides Descendants and DescendantNodes methods that return child elements or nodes plus all of their children, and so on (the entire tree). Descendants accepts an optional element name. Returning to our earlier example, we can use Descendants to find all the hand tools as follows: Console.WriteLine (bench.Descendants ("handtool").Count()); // 3 Both parent and leaf nodes are included, as the following example demonstrates: foreach (XNode node in bench.DescendantNodes()) Console.WriteLine (node.ToString (SaveOptions.DisableFormatting)); HammerRasp Hammer Hammer Rasp Rasp SawNailgun Saw Saw Nailgun Nailgun The next query extracts all comments anywhere within the X-DOM that contain the word “careful”: IEnumerable query = from c in bench.DescendantNodes().OfType() where c.Value.Contains ("careful") orderby c.Value select c.Value; Navigating and Querying | 433 LINQ to XML Parent Navigation All XNodes have a Parent property and AncestorXXX methods for parent navigation. A parent is always an XElement: Return type Members Works on XElement Parent { get; } XNode* Enumerable Ancestors() XNode* Ancestors (XName) XNode* AncestorsAndSelf() XElement* AncestorsAndSelf (XName) XElement* If x is an XElement, the following always prints true: foreach (XNode child in x.Nodes()) Console.WriteLine (child.Parent == x); The same is not the case, however, if x is an XDocument. XDocument is peculiar: it can have children, but can never be anyone’s parent! To access the XDocument, you instead use the Document property—this works on any object in the X-DOM tree. Ancestors returns a sequence whose first element is Parent, and whose next element is Parent.Parent, and so on, until the root element. You can navigate to the root element with the LINQ query AncestorsAndSelf().Last(). Another way to achieve the same thing is to call Document.Root —although this works only if an XDocument is present. Peer Node Navigation Return type Members Defined in bool IsBefore (XNode node) XNode IsAfter (XNode node) XNode XNode PreviousNode { get; } XNode NextNode { get; } XNode IEnumerable NodesBeforeSelf() XNode NodesAfterSelf() XNode IEnumerable ElementsBeforeSelf() XNode ElementsBeforeSelf (XName name) XNode ElementsAfterSelf() XNode ElementsAfterSelf (XName name) XNode 434 | Chapter 10: LINQ to XML With PreviousNode and NextNode (and FirstNode/LastNode), you can traverse nodes with the feel of a linked list. This is noncoincidental: internally, nodes are stored in a linked list. XNode internally uses a singly linked list, so PreviousNode is not performant. Attribute Navigation Return type Members Defined in bool HasAttributes { get; } XElement XAttribute Attribute (XName name) XElement FirstAttribute { get; } XElement LastAttribute { get; } XElement IEnumerable Attributes() XElement Attributes (XName name) XElement In addition, XAttribute defines PreviousAttribute and NextAttribute properties, as well as Parent. The Attributes method that accepts a name returns a sequence with either zero or one element; an element cannot have duplicate attribute names in XML. Updating an X-DOM You can update elements and attributes in the following ways: • Call SetValue or reassign the Value property. • Call SetElementValue or SetAttributeValue. • Call one of the RemoveXXX methods. • Call one of the AddXXX or ReplaceXXX methods, specifying fresh content. You can also reassign the Name property on XElement objects. Simple Value Updates Members Works on SetValue (object value) XElement, XAttribute Value { get; set } XElement, XAttribute The SetValue method replaces an element or attribute’s content with a simple value. Setting the Value property does the same, but accepts string data only. We describe Updating an X-DOM | 435 LINQ to XML both of these functions in detail later in this chapter (see the section “Working with Values” on page 438). An effect of calling SetValue (or reassigning Value) is that it replaces all child nodes: XElement settings = new XElement ("settings", new XElement ("timeout", 30) ); settings.SetValue ("blah"); Console.WriteLine (settings.ToString()); // blah Updating Child Nodes and Attributes Category Members Works on Add Add (params object[] content) XContainer AddFirst (params object[] content) XContainer Remove RemoveNodes() XContainer RemoveAttributes() XElement RemoveAll() XElement Update ReplaceNodes (params object[] content) XContainer ReplaceAttributes (params object[] content) XElement ReplaceAll (params object[] content XElement SetElementValue (XName name, object value) XElement SetAttributeValue (XName name, object value) XElement The most convenient methods in this group are the last two: SetElementValue and SetAttributeValue. They serve as shortcuts for instantiating an XElement or XAttri bute and then Adding it to a parent, replacing any existing element or attribute of that name: XElement settings = new XElement ("settings"); settings.SetElementValue ("timeout", 30); // Adds child node settings.SetElementValue ("timeout", 60); // Update it to 60 Add appends a child node to an element or document. AddFirst does the same thing, but inserts at the beginning of the collection rather than the end. You can remove all child nodes or attributes in one hit with RemoveNodes or Remove Attributes. RemoveAll is equivalent to calling both of these methods. The ReplaceXXX methods are equivalent to Removing and then Adding. They take a snapshot of the input, so e.ReplaceNodes(e.Nodes()) works as expected. Updating through the Parent Members Works on AddBeforeSelf (params object[] content) XNode AddAfterSelf (params object[] content) XNode 436 | Chapter 10: LINQ to XML Members Works on Remove() XNode*, XAttribute* ReplaceWith (params object[] content) XNode The methods AddBeforeSelf, AddAfterSelf, Remove, and ReplaceWith don’t operate on the node’s children. Instead, they operate on the collection in which the node itself is in. This requires that the node have a parent element—otherwise, an excep- tion is thrown. AddBeforeSelf and AddAfterSelf are useful for inserting a node into an arbitrary position: XElement items = new XElement ("items", new XElement ("one"), new XElement ("three") ); items.FirstNode.AddAfterSelf (new XElement ("two")); Here’s the result: Inserting into an arbitrary position within a long sequence of elements is actually quite efficient, because nodes are stored internally in a linked list. The Remove method removes the current node from its parent. ReplaceWith does the same—and then inserts some other content at the same position. For instance: XElement items = XElement.Parse (""); items.FirstNode.ReplaceWith (new XComment ("One was here")); Here’s the result: Removing a sequence of nodes or attributes Thanks to extension methods in System.Xml.Linq, you can also call Remove on a sequence of nodes or attributes. Consider this X-DOM: XElement contacts = XElement.Parse ( @" 012345678 "); The following removes all customers: contacts.Elements ("customer").Remove(); The next statement removes all archived contacts (so Chris disappears): contacts.Elements().Where (e => (bool?) e.Attribute ("archived") == true) .Remove(); Updating an X-DOM | 437 LINQ to XML If we replaced Elements() with Descendants(), all archived elements throughout the DOM would disappear, with this result: The next example removes all contacts that feature the comment “confidential” anywhere in their tree: contacts.Elements().Where (e => e.DescendantNodes() .OfType() .Any (c => c.Value == "confidential") ).Remove(); This is the result: Contrast this with the following simpler query, which strips all comment nodes from the tree: contacts.DescendantNodes().OfType().Remove(); Internally, the Remove methods first read all matching elements into a temporary list, and then enumerate over the temporary list to perform the deletions. This avoids errors that could other- wise result from deleting and querying at the same time. Working with Values XElement and XAttribute both have a Value property of type string. If an element has a single XText child node, XElement’s Value property acts as a convenient shortcut to the content of that node. With XAttribute, the Value property is simply the attribute’s value. Despite the storage differences, the X-DOM provides a consistent set of operations for working with element and attribute values. Setting Values There are two ways to assign a value: call SetValue or assign the Value property. SetValue is more flexible because it accepts not just strings, but other simple data types too: var e = new XElement ("date", DateTime.Now); e.SetValue (DateTime.Now.AddDays(1)); Console.Write (e.Value); // 2007-03-02T16:39:10.734375+09:00 438 | Chapter 10: LINQ to XML We could have instead just set the element’s Value property, but this would mean manually converting the DateTime to a string. This is more complicated than calling ToString—it requires the use of XmlConvert for an XML-compliant result. When you pass a value into XElement or XAttribute’s constructor, the same auto- matic conversion takes place for nonstring types. This ensures that DateTimes are correctly formatted; true is written in lowercase, and double.NegativeInfinity is written as “-INF”. Getting Values To go the other way around and parse a Value back to a base type, you simply cast the XElement or XAttribute to the desired type. It sounds like it shouldn’t work— but it does! For instance: XElement e = new XElement ("now", DateTime.Now); DateTime dt = (DateTime) e; XAttribute a = new XAttribute ("resolution", 1.234); double res = (double) a; An element or attribute doesn’t store DateTimes or numbers natively—they’re always stored as text, and then parsed as needed. It also doesn’t “remember” the original type, so you must cast it correctly to avoid a runtime error. To make your code robust, you can put the cast in a try/catch block, catching a FormatException. Explicit casts on XElement and XAttribute can parse to the following types: • All standard numeric types • string, bool, DateTime, DateTimeOffset, TimeSpan, and Guid • Nullable<> versions of the aforementioned value types Casting to a nullable type is useful in conjunction with the Element and Attribute methods, because if the requested name doesn’t exist, the cast still works. For instance, if x has no timeout element, the first line generates a runtime error and the second line does not: int timeout = (int) x.Element ("timeout"); // Error int? timeout = (int?) x.Element ("timeout"); // OK; timeout is null. You can factor away the nullable type in the final result with the ?? operator. The following evaluates to 1.0 if the resolution attribute doesn’t exist: double resolution = (double?) x.Attribute ("resolution") ?? 1.0; Casting to a nullable type won’t get you out of trouble, though, if the element or attribute exists and has an empty (or improperly formatted) value. For this, you must catch a FormatException. You can also use casts in LINQ queries. The following returns “John”: var data = XElement.Parse ( @" Working with Values | 439 LINQ to XML "); IEnumerable query = from cust in data.Elements() where (int?) cust.Attribute ("credit") > 100 select cust.Attribute ("name").Value; Casting to a nullable int avoids a NullReferenceException in the case of Anne, who has no credit attribute. Another solution would be to add a predicate to the where clause: where cust.Attributes ("credit").Any() && (int) cust.Attribute... The same principles apply in querying element values. Values and Mixed Content Nodes Given the value of Value, you might wonder when you’d ever need to deal directly with XText nodes. The answer is when you have mixed content. For example: An XAttribute is not an XNode A simple Value property is not enough to capture summary’s content. The summary element contains three children: an XText node followed by an XElement, followed by another XText node. Here’s how to construct it: XElement summary = new XElement ("summary", new XText ("An XAttribute is "), new XElement ("bold", "not"), new XText (" an XNode") ); Interestingly, we can still query summary’s Value—without getting an exception. Instead, we get a concatenation of each child’s value: An XAttribute is not an XNode It’s also legal to reassign summary’s Value, at the cost of replacing all previous children with a single new XText node. Automatic XText Concatenation When you add simple content to an XElement, the X-DOM appends to the existing XText child rather than creating a new one. In the following examples, e1 and e2 end up with just one child XText element whose value is HelloWorld: var e1 = new XElement ("test", "Hello"); e1.Add ("World"); var e2 = new XElement ("test", "Hello", "World"); If you specifically create XText nodes, however, you end up with multiple children: var e = new XElement ("test", new XText ("Hello"), new XText ("World")); Console.WriteLine (e.Value); // HelloWorld Console.WriteLine (e.Nodes().Count()); // 2 XElement doesn’t concatenate the two XText nodes, so the nodes’ object identities are preserved. 440 | Chapter 10: LINQ to XML Documents and Declarations XDocument As we said previously, an XDocument wraps a root XElement and allows you to add an XDeclaration, processing instructions, a document type, and root-level comments. An XDocument is optional and can be ignored or omitted: unlike with the W3C DOM, it does not serve as glue to keep everything together. An XDocument provides the same functional constructors as XElement. And because it’s based on XContainer, it also supports the AddXXX, RemoveXXX, and ReplaceXXX methods. Unlike XElement, however, an XDocument can accept only limited content: • A single XElement object (the “root”) • A single XDeclaration object • A single XDocumentType object (to reference a DTD) • Any number of XProcessingInstruction objects • Any number of XComment objects Of these, only the root XElement is mandatory in order to have a valid XDocument. The XDeclaration is optional—if omitted, default settings are applied during serialization. The simplest valid XDocument has just a root element: var doc = new XDocument ( new XElement ("test", "data") ); Notice that we didn’t include an XDeclaration object. The file generated by calling doc.Save would still contain an XML declaration, however, because one is generated by default. The next example produces a simple but correct XHTML file, illustrating all the constructs that an XDocument can accept: var styleInstruction = new XProcessingInstruction ( "xml-stylesheet", "href='styles.css' type='text/css'"); var docType = new XDocumentType ("html", "-//W3C//DTD XHTML 1.0 Strict//EN", "http://www.w3.org/TR/xhtml1/DTD/xhtml1-strict.dtd", null); XNamespace ns = "http://www.w3.org/1999/xhtml"; var root = new XElement (ns + "html", new XElement (ns + "head", new XElement (ns + "title", "An XHTML page")), new XElement (ns + "body", new XElement (ns + "p", "This is the content")) Documents and Declarations | 441 LINQ to XML ); var doc = new XDocument ( new XDeclaration ("1.0", "utf-8", "no"), new XComment ("Reference a stylesheet"), styleInstruction, docType, root); doc.Save ("test.html"); The resultant test.html reads as follows: An XHTML page

This is the content

XDocument has a Root property that serves as a shortcut for accessing a document’s single XElement. The reverse link is provided by XObject’s Document property, which works for all objects in the tree: Console.WriteLine (doc.Root.Name.LocalName); // html XElement bodyNode = doc.Root.Element (ns + "body"); Console.WriteLine (bodyNode.Document == doc); // True Recall that a document’s children have no Parent: Console.WriteLine (doc.Root.Parent == null); // True foreach (XNode node in doc.Nodes()) Console.Write (node.Parent == null); // TrueTrueTrueTrue An XDeclaration is not an XNode and does not appear in the document’s Nodes collection—unlike comments, processing instructions, and the root element. Instead, it gets assigned to a dedicated property called Declaration. This is why “True” is repeated four and not five times in the last example. XML Declarations A standard XML file starts with a declaration such as the following: 442 | Chapter 10: LINQ to XML An XML declaration ensures that the file will be correctly parsed and understood by a reader. XElement and XDocument follow these rules in emitting XML declarations: • Calling Save with a filename always writes a declaration. • Calling Save with an XmlWriter writes a declaration unless the XmlWriter is instructed otherwise. • The ToString method never emits an XML declaration. You can instruct an XmlWriter not to produce a declaration by setting the OmitXmlDeclaration and ConformanceLevel proper- ties of an XmlWriterSettings object when constructing the XmlWriter. We describe this in Chapter 11. The presence or absence of an XDeclaration object has no effect on whether an XML declaration gets written. The purpose of an XDeclaration is instead to hint the XML serialization—in two ways: • What text encoding to use • What to put in the XML declaration’s encoding and standalone attributes (should a declaration be written) XDeclaration’s constructor accepts three arguments, which correspond to the attributes version, encoding, and standalone. In the following example, test.xml is encoded in UTF-16: var doc = new XDocument ( new XDeclaration ("1.0", "utf-16", "yes"), new XElement ("test", "data") ); doc.Save ("test.xml"); Whatever you specify for the XML version is ignored by the XML writer: it always writes "1.0". The encoding must use an IETF code such as "utf-16"—just as it would appear in the XML declaration. Writing a declaration to a string Suppose we want to serialize an XDocument to a string—including the XML decla- ration. Because ToString doesn’t write a declaration, we’d have to use an XmlWriter instead: var doc = new XDocument ( new XDeclaration ("1.0", "utf-8", "yes"), new XElement ("test", "data") ); var output = new StringBuilder(); Documents and Declarations | 443 LINQ to XML var settings = new XmlWriterSettings { Indent = true }; using (XmlWriter xw = XmlWriter.Create (output, settings)) doc.Save (xw); Console.WriteLine (output.ToString()); This is the result: data Notice that we got UTF-16 in the output—even though we explicitly requested UTF-8 in an XDeclaration! This might look like a bug, but in fact, XmlWriter is being remarkably smart. Because we’re writing to a string and not a file or stream, it’s impossible to apply any encoding other than UTF-16—the format in which strings are internally stored. Hence, XmlWriter writes "utf-16"—so as not to lie. This also explains why the ToString method doesn’t emit an XML declaration. Imagine that instead of calling Save, you did the following to write an XDocument to a file: File.WriteAllText ("data.xml", doc.ToString()); As it stands, data.xml would lack an XML declaration, making it incomplete but still parsable (you can infer the text encoding). But if ToString() emitted an XML dec- laration, data.xml would actually contain an incorrect declaration (encod ing="utf-16"), which might prevent it from being read at all, because WriteAll Text encodes using UTF-8. Names and Namespaces Just as .NET types can have namespaces, so too can XML elements and attributes. XML namespaces achieve two things. First, rather like namespaces in C#, they help avoid naming collisions. This can become an issue when you merge data from one XML file into another. Second, namespaces assign absolute meaning to a name. The name “nil,” for instance, could mean anything. Within the http://www.w3.org/2001/ xmlschema-instance namespace, however, “nil” means something equivalent to null in C# and comes with specific rules on how it can be applied. Because XML namespaces are a significant source of confusion, we’ll cover the topic first in general, and then move on to how they’re used in LINQ to XML. Namespaces in XML Suppose we want to define a customer element in the namespace OReilly.Nut shell.CSharp. There are two ways to proceed. The first is to use the xmlns attribute as follows: 444 | Chapter 10: LINQ to XML xmlns is a special reserved attribute. When used in this manner, it performs two functions: • It specifies a namespace for the element in question. • It specifies a default namespace for all descendant elements. This means that in the following example, address and postcode implicitly live in the OReilly.Nutshell.CSharp namespace:
If we want address and postcode to have no namespace, we’d have to do this:
Prefixes The other way to specify a namespace is with a prefix. A prefix is an alias that you assign to a namespace to save typing. There are two steps in using a prefix— defining the prefix and using it. You can do both together as follows: Two distinct things are happening here. On the right, xmlns:nut="..." defines a prefix called nut and makes it available to this element and all its descendants. On the left, nut:customer assigns the newly allocated prefix to the customer element. A prefixed element does not define a default namespace for descendants. In the fol- lowing XML, firstname has an empty namespace: Joe
To give firstname the OReilly.Nutshell.CSharp prefix, we must do this: Joe
You can also define a prefix—or prefixes—for the convenience of your descendants, without assigning any of them to the parent element itself. The following defines two prefixes, i and z, while leaving the customer element itself with an empty name- space: ... Names and Namespaces | 445 LINQ to XML If this was the root node, the whole document would have i and z at its fingertips. Prefixes are convenient when elements need to draw from a number of namespaces. Notice that both namespaces in this example are URIs. Using URIs (that you own) is standard practice: it ensures namespace uniqueness. So, in real life, our cus tomer element would more likely be: or: Attributes You can assign namespaces to attributes too. The main difference is that it always requires a prefix. For instance: Another difference is that an unqualified attribute always has an empty namespace: it never inherits a default namespace from a parent element. Attributes tend not to need namespaces because their meaning is usually local to the element. An exception is with general-purpose or metadata attributes, such as the nil attribute defined by W3C: Joe This indicates unambiguously that lastname is nil (null in C#) and not an empty string. Because we’ve used the standard namespace, a general-purpose parsing utility could know with certainty our intention. Specifying Namespaces in the X-DOM So far in this chapter, we’ve used just simple strings for XElement and XAttribute names. A simple string corresponds to an XML name with an empty namespace— rather like a .NET type defined in the global namespace. There are a couple of ways to specify an XML namespace. The first is to enclose it in braces, before the local name. For example: var e = new XElement ("{http://domain.com/xmlspace}customer", "Bloggs"); Console.WriteLine (e.ToString()); Here’s the resulting XML: Bloggs The second (and more performant) approach is to use the XNamespace and XName types. Here are their definitions: public sealed class XNamespace { public string NamespaceName { get; } 446 | Chapter 10: LINQ to XML } public sealed class XName // A local name with optional namespace { public string LocalName { get; } public XNamespace Namespace { get; } // Optional } Both types define implicit casts from string, so the following is legal: XNamespace ns = "http://domain.com/xmlspace"; XName localName = "customer"; XName fullName = "{http://domain.com/xmlspace}customer"; XNamespace also overloads the + operator, allowing you to combine a namespace and name into an XName without using braces: XNamespace ns = "http://domain.com/xmlspace"; XName fullName = ns + "customer"; Console.WriteLine (fullName); // {http://domain.com/xmlspace}customer All constructors and methods in the X-DOM that accept an element or attribute name actually accept an XName object rather than a string. The reason you can sub- stitute a string—as in all our examples to date—is because of the implicit cast. Specifying a namespace is the same whether for an element or an attribute: XNamespace ns = "http://domain.com/xmlspace"; var data = new XElement (ns + "data", new XAttribute (ns + "id", 123) ); The X-DOM and Default Namespaces The X-DOM ignores the concept of default namespaces until it comes time to actually output XML. This means that when you construct a child XElement, you must give it a namespace explicitly if needed: it will not inherit from the parent: XNamespace ns = "http://domain.com/xmlspace"; var data = new XElement (ns + "data", new XElement (ns + "customer", "Bloggs"), new XElement (ns + "purchase", "Bicycle") ); The X-DOM does, however, apply default namespaces when reading and outputting XML: Console.WriteLine (data.ToString()); OUTPUT: Bloggs Bicycle Console.WriteLine (data.Element (ns + "customer").ToString()); Names and Namespaces | 447 LINQ to XML OUTPUT: Bloggs If you construct XElement children without specifying namespaces—in other words: XNamespace ns = "http://domain.com/xmlspace"; var data = new XElement (ns + "data", new XElement ("customer", "Bloggs"), new XElement ("purchase", "Bicycle") ); Console.WriteLine (data.ToString()); you get this result instead: Bloggs Bicycle Another trap is failing to include a namespace when navigating an X-DOM: XNamespace ns = "http://domain.com/xmlspace"; var data = new XElement (ns + "data", new XElement (ns + "customer", "Bloggs"), new XElement (ns + "purchase", "Bicycle") ); XElement x = data.Element (ns + "customer"); // ok XElement y = data.Element ("customer"); // null If you build an X-DOM tree without specifying namespaces, you can subsequently assign every element to a single namespace as follows: foreach (XElement e in data.DescendantsAndSelf()) if (e.Name.Namespace == "") e.Name = ns + e.Name.LocalName; Prefixes The X-DOM treats prefixes just as it treats namespaces: purely as a serialization function. This means you can choose to completely ignore the issue of prefixes— and get by! The only reason you might want to do otherwise is for efficiency when outputting to an XML file. For example, consider this: XNamespace ns1 = "http://domain.com/space1"; XNamespace ns2 = "http://domain.com/space2"; var mix = new XElement (ns1 + "data", new XElement (ns2 + "element", "value"), new XElement (ns2 + "element", "value"), new XElement (ns2 + "element", "value") ); By default, XElement will serialize this as follows: value value 448 | Chapter 10: LINQ to XML value As you can see, there’s a bit of unnecessary duplication. The solution is not to change the way you construct the X-DOM, but instead to hint the serializer prior to writing the XML. Do this by adding attributes defining prefixes that you want to see applied. This is typically done on the root element: mix.SetAttributeValue (XNamespace.Xmlns + "ns1", ns1); mix.SetAttributeValue (XNamespace.Xmlns + "ns2", ns2); This assigns the prefix “ns1” to our XNamespace variable ns1, and “ns2” to ns2. The X-DOM automatically picks up these attributes when serializing and uses them to condense the resulting XML. Here’s the result now of calling ToString on mix: value value value Prefixes don’t change the way you construct, query, or update the X-DOM—for these activities, you ignore the presence of prefixes and continue to use full names. Prefixes come into play only when converting to and from XML files or streams. Prefixes are also honored in serializing attributes. In the following example, we record a customer’s date of birth and credit as "nil" using the W3C-standard at- tribute. The highlighted line ensures that the prefix is serialized without unnecessary namespace repetition: XNamespace xsi = "http://www.w3.org/2001/XMLSchema-instance"; var nil = new XAttribute (xsi + "nil", true); var cust = new XElement ("customers", new XAttribute (XNamespace.Xmlns + "xsi", xsi), new XElement ("customer", new XElement ("lastname", "Bloggs"), new XElement ("dob", nil), new XElement ("credit", nil) ) ); This is its XML: Bloggs For brevity, we predeclared the nil XAttribute so that we could use it twice in build- ing the DOM. You’re allowed to reference the same attribute twice because it’s automatically duplicated as required. Names and Namespaces | 449 LINQ to XML Annotations You can attach custom data to any XObject with an annotation. Annotations are intended for your own private use and are treated as black boxes by X-DOM. If you’ve ever used the Tag property on a Windows Forms or WPF control, you’ll be familiar with the concept—the difference is that you have multiple annotations, and your annotations can be privately scoped. You can create an annotation that other types cannot even see—let alone overwrite. The following methods on XObject add and remove annotations: public void AddAnnotation (object annotation) public void RemoveAnnotations() where T : class The following methods retrieve annotations: public T Annotation() where T : class public IEnumerable Annotations() where T : class Each annotation is keyed by its type, which must be a reference type. The following adds and then retrieves a string annotation: XElement e = new XElement ("test"); e.AddAnnotation ("Hello"); Console.WriteLine (e.Annotation()); // Hello You can add multiple annotations of the same type, and then use the Annotations method to retrieve a sequence of matches. A public type such as string doesn’t make a great key, however, because code in other types can interfere with your annotations. A better approach is to use an internal or (nested) private class: class X { class CustomData { internal string Message; } // Private nested type static void Test() { XElement e = new XElement ("test"); e.AddAnnotation (new CustomData { Message = "Hello" } ); Console.Write (e.Annotations().First().Message); // Hello } } To remove annotations, you must also have access to the key’s type: e.RemoveAnnotations(); Projecting into an X-DOM So far, we’ve shown how to use LINQ to get data out of an X-DOM. You can also use LINQ queries to project into an X-DOM. The source can be anything over which LINQ can query, such as: • LINQ to SQL or Entity Framework queries 450 | Chapter 10: LINQ to XML • A local collection • Another X-DOM Regardless of the source, the strategy is the same in using LINQ to emit an X-DOM: first write a functional construction expression that produces the desired X-DOM shape, and then build a LINQ query around the expression. For instance, suppose we want to retrieve customers from a database into the fol- lowing XML: Sue 3 ... We start by writing a functional construction expression for the X-DOM using sim- ple literals: var customers = new XElement ("customers", new XElement ("customer", new XAttribute ("id", 1), new XElement ("name", "Sue"), new XElement ("buys", 3) ) ); We then turn this into a projection and build a LINQ query around it: var customers = new XElement ("customers", from c in dataContext.Customers select new XElement ("customer", new XAttribute ("id", c.ID), new XElement ("name", c.Name), new XElement ("buys", c.Purchases.Count) ) ); In Entity Framework, you must call .ToList() after retrieving customers, so that the third line reads: from c in objectContext.Customers.ToList() Here’s the result: Tom 3 Harry Projecting into an X-DOM | 451 LINQ to XML 2 ... We can see how this works more clearly by constructing the same query in two steps. First: IEnumerable sqlQuery = from c in dataContext.Customers select new XElement ("customer", new XAttribute ("id", c.ID), new XElement ("name", c.Name), new XElement ("buys", c.Purchases.Count) ); This inner portion is a normal LINQ to SQL query that projects into custom types (from LINQ to SQL’s perspective). Here’s the second step: var customers = new XElement ("customers", sqlQuery); This constructs the root XElement. The only thing unusual is that the content, sqlQuery, is not a single XElement but an IQueryable—which implements IEnumerable. Remember that in the processing of XML content, collec- tions are automatically enumerated. So, each XElement gets added as a child node. This outer query also defines the line at which the query transitions from being a database query to a local LINQ to enumerable query. XElement’s constructor doesn’t know about IQueryable<>, so it forces enumeration of the database query—and execution of the SQL statement. Eliminating Empty Elements Suppose in the preceding example that we also wanted to include details of the customer’s most recent high-value purchase. We could do this as follows: var customers = new XElement ("customers", from c in dataContext.Customers let lastBigBuy = (from p in c.Purchases where p.Price > 1000 orderby p.Date descending select p).FirstOrDefault() select new XElement ("customer", new XAttribute ("id", c.ID), new XElement ("name", c.Name), new XElement ("buys", c.Purchases.Count), new XElement ("lastBigBuy", new XElement ("description", lastBigBuy == null ? null : lastBigBuy.Description), new XElement ("price", lastBigBuy == null ? 0m : lastBigBuy.Price) ) ) ); 452 | Chapter 10: LINQ to XML This emits empty elements, though, for customers with no high-value purchases. (If it was a local query rather than a database query, it would throw a NullReferenceEx ception.) In such cases, it would be better to omit the lastBigBuy node entirely. We can achieve this by wrapping the constructor for the lastBigBuy element in a con- ditional operator: select new XElement ("customer", new XAttribute ("id", c.ID), new XElement ("name", c.Name), new XElement ("buys", c.Purchases.Count), lastBigBuy == null ? null : new XElement ("lastBigBuy", new XElement ("description", lastBigBuy.Description), new XElement ("price", lastBigBuy.Price) For customers with no lastBigBuy, a null is emitted instead of an empty XElement. This is what we want, because null content is simply ignored. Streaming a Projection If you’re projecting into an X-DOM only to Save it (or call ToString on it), you can improve memory efficiency through an XStreamingElement. An XStreamingElement is a cut-down version of XElement that applies deferred loading semantics to its child content. To use it, you simply replace the outer XElements with XStreamingElements: var customers = new XStreamingElement ("customers", from c in dataContext.Customers select new XStreamingElement ("customer", new XAttribute ("id", c.ID), new XElement ("name", c.Name), new XElement ("buys", c.Purchases.Count) ) ); customers.Save ("data.xml"); The queries passed into an XStreamingElement’s constructor are not enumerated until you call Save, ToString, or WriteTo on the element; this avoids loading the whole X-DOM into memory at once. The flipside is that the queries are reevaluated, should you re-Save. Also, you cannot traverse an XStreamingElement’s child content—it does not expose methods such as Elements or Attributes. XStreamingElement is not based on XObject—or any other class—because it has such a limited set of members. The only members it has, besides Save, ToString, and WriteTo, are: • An Add method, which accepts content like the constructor • A Name property XStreamingElement does not allow you to read content in a streamed fashion—for this, you must use an XmlReader in conjunction with the X-DOM. We describe how to do this in the section “Patterns for Using XmlReader/XmlWriter” on page 469 in Chapter 11. Projecting into an X-DOM | 453 LINQ to XML Transforming an X-DOM You can transform an X-DOM by reprojecting it. For instance, suppose we want to transform an msbuild XML file, used by the C# compiler and Visual Studio to describe a project, into a simple format suitable for generating a report. An msbuild file looks like this: AnyCPU 9.0.11209 ... ... ... Let’s say we want to include only files, as follows: ObjectGraph.cs Program.cs Properties\AssemblyInfo.cs Tests\Aggregation.cs Tests\Advanced\RecursiveXml.cs The following query performs this transformation: XElement project = XElement.Load ("myProjectFile.csproj"); XNamespace ns = project.Name.Namespace; var query = new XElement ("ProjectReport", from compileItem in project.Elements (ns + "ItemGroup").Elements (ns + "Compile") let include = compileItem.Attribute ("Include") where include != null select new XElement ("File", include.Value) ); The query first extracts all ItemGroup elements, and then uses the Elements extension method to obtain a flat sequence of all their Compile subelements. Notice that we had to specify an XML namespace—everything in the original file inherits the name- space defined by the Project element—so a local element name such as ItemGroup won’t work on its own. Then, we extracted the Include attribute value and projected its value as an element. 454 | Chapter 10: LINQ to XML Advanced transformations When querying a local collection such as an X-DOM, you’re free to write custom query operators to assist with more complex queries. Suppose in the preceding example that we instead wanted a hierarchical output, based on folders: ObjectGraph.cs Program.cs AssemblyInfo.cs Aggregation.cs RecursiveXml.cs To produce this, we need to process path strings such as Tests\Advanced\Re- cursiveXml.cs recursively. The following method does just this: it accepts a sequence of path strings and emits an X-DOM hierarchy consistent with our desired output: static IEnumerable ExpandPaths (IEnumerable paths) { var brokenUp = from path in paths let split = path.Split (new char[] { '\\' }, 2) orderby split[0] select new { name = split[0], remainder = split.ElementAtOrDefault (1) }; IEnumerable files = from b in brokenUp where b.remainder == null select new XElement ("file", b.name); IEnumerable folders = from b in brokenUp where b.remainder != null group b.remainder by b.name into grp select new XElement ("folder", new XAttribute ("name", grp.Key), ExpandPaths (grp) ); return files.Concat (folders); } The first query splits each path string at the first backslash, into a name + remainder: Tests\Advanced\RecursiveXml.cs -> Tests + Advanced\RecursiveXml.cs If remainder is null, we’re dealing with a straight filename. The files query extracts these cases. Projecting into an X-DOM | 455 LINQ to XML If remainder is not null, we’ve got a folder. The folders query handles these cases. Because other files can be in the same folder, it must group by folder name to bring them all together. For each group, it then executes the same function for the sub- elements. The final result is a concatenation of files and folders. The Concat operator pre- serves order, so all the files come first, alphabetically, then all the folders, alphabetically. With this method in place, we can complete the query in two steps. First, we extract a simple sequence of path strings: IEnumerable paths = from compileItem in project.Elements (ns + "ItemGroup").Elements (ns + "Compile") let include = compileItem.Attribute ("Include") where include != null select include.Value; Then, we feed this into our ExpandPaths method for the final result: var query = new XElement ("Project", ExpandPaths (paths)); 456 | Chapter 10: LINQ to XML 11Other XML Technologies The System.Xml namespace comprises the following namespaces and core classes: System.Xml.* XmlReader and XmlWriter High-performance, forward-only cursors for reading or writing an XML stream XmlDocument Represents an XML document in a W3C-style DOM System.Xml.XPath Infrastructure and API (XPathNavigator) for XPath, a string-based language for querying XML System.Xml.XmlSchema Infrastructure and API for (W3C) XSD schemas System.Xml.Xsl Infrastructure and API (XslCompiledTransform) for performing (W3C) XSLT transformations of XML System.Xml.Serialization Supports the serialization of classes to and from XML (see Chapter 17) System.Xml.XLinq Modern, simplified, LINQ-centric version of XmlDocument (see Chapter 10) W3C is an abbreviation for World Wide Web Consortium, where the XML standards are defined. XmlConvert, the static class for parsing and formatting XML strings, is covered in Chapter 6. 457 XmlReader XmlReader is a high-performance class for reading an XML stream in a low-level, forward-only manner. Consider the following XML file: Jim Bo To instantiate an XmlReader, you call the static XmlReader.Create method, passing in a Stream, a TextReader, or a URI string. For example: using (XmlReader reader = XmlReader.Create ("customer.xml")) ... Because XmlReader lets you read from potentially slow sources (Streams and URIs), it offers asynchronous versions of most of its methods so that you can easily write non-blocking code. We’ll cover asynchrony in detail in Chapter 14. To construct an XmlReader that reads from a string: XmlReader reader = XmlReader.Create ( new System.IO.StringReader (myString)); You can also pass in an XmlReaderSettings object to control parsing and validation options. The following three properties on XmlReaderSettings are particularly useful for skipping over superfluous content: bool IgnoreComments // Skip over comment nodes? bool IgnoreProcessingInstructions // Skip over processing instructions? bool IgnoreWhitespace // Skip over whitespace? In the following example, we instruct the reader not to emit whitespace nodes, which are a distraction in typical scenarios: XmlReaderSettings settings = new XmlReaderSettings(); settings.IgnoreWhitespace = true; using (XmlReader reader = XmlReader.Create ("customer.xml", settings)) ... Another useful property on XmlReaderSettings is ConformanceLevel. Its default value of Document instructs the reader to assume a valid XML document with a single root node. This is a problem if you want to read just an inner portion of XML, containing multiple nodes: Jim Bo 458 | Chapter 11: Other XML Technologies To read this without throwing an exception, you must set ConformanceLevel to Fragment. XmlReaderSettings also has a property called CloseInput, which indicates whether to close the underlying stream when the reader is closed (there’s an analogous prop- erty on XmlWriterSettings called CloseOutput). The default value for CloseInput and CloseOutput is false. Reading Nodes The units of an XML stream are XML nodes. The reader traverses the stream in textual (depth-first) order. The Depth property of the reader returns the current depth of the cursor. The most primitive way to read from an XmlReader is to call Read. It advances to the next node in the XML stream, rather like MoveNext in IEnumerator. The first call to Read positions the cursor at the first node. When Read returns false, it means the cursor has advanced past the last node, at which point the XmlReader should be closed and abandoned. In this example, we read every node in the XML stream, outputting each node type as we go: XmlReaderSettings settings = new XmlReaderSettings(); settings.IgnoreWhitespace = true; using (XmlReader reader = XmlReader.Create ("customer.xml", settings)) while (reader.Read()) { Console.Write (new string (' ',reader.Depth*2)); // Write indentation Console.WriteLine (reader.NodeType); } The output is as follows: XmlDeclaration Element Element Text EndElement Element Text EndElement EndElement XmlReader | 459 Other XML Attributes are not included in Read-based traversal (see the sec- tion “Reading Attributes” on page 465 later in this chapter). NodeType is of type XmlNodeType, which is an enum with these members: None XmlDeclaration Element EndElement Text Attribute Comment Entity EndEntity EntityReference ProcessingInstruction CDATA Document DocumentType DocumentFragment Notation Whitespace SignificantWhitespace Two string properties on XmlReader provide access to a node’s content: Name and Value. Depending on the node type, either Name or Value (or both) is populated: XmlReaderSettings settings = new XmlReaderSettings(); settings.IgnoreWhitespace = true; settings.ProhibitDtd = false; // Must set this to read DTDs using (XmlReader r = XmlReader.Create ("customer.xml", settings)) while (r.Read()) { Console.Write (r.NodeType.ToString().PadRight (17, '-')); Console.Write ("> ".PadRight (r.Depth * 3)); switch (r.NodeType) { case XmlNodeType.Element: case XmlNodeType.EndElement: Console.WriteLine (r.Name); break; case XmlNodeType.Text: case XmlNodeType.CDATA: case XmlNodeType.Comment: case XmlNodeType.XmlDeclaration: Console.WriteLine (r.Value); break; case XmlNodeType.DocumentType: Console.WriteLine (r.Name + " - " + r.Value); break; default: break; } } To demonstrate this, we’ll expand our XML file to include a document type, entity, CDATA, and comment: ]> 460 | Chapter 11: Other XML Technologies Jim Bo &]]> Jim Bo is a &tc; An entity is like a macro; a CDATA is like a verbatim string (@"...") in C#. Here’s the result: XmlDeclaration---> version="1.0" encoding="utf-8" DocumentType-----> customer - Element----------> customer Element----------> firstname Text-------------> Jim EndElement-------> firstname Element----------> lastname Text-------------> Bo EndElement-------> lastname Element----------> quote CDATA------------> C#'s operators include: < > & EndElement-------> quote Element----------> notes Text-------------> Jim Bo is a Top Customer EndElement-------> notes Comment----------> That wasn't so bad! EndElement-------> customer XmlReader automatically resolves entities, so in our example, the entity reference &tc; expands into Top Customer. Reading Elements Often, you already know the structure of the XML document that you’re reading. To help with this, XmlReader provides a range of methods that read while presum- ing a particular structure. This simplifies your code, as well as performing some validation at the same time. XmlReader throws an XmlException if any validation fails. XmlEx ception has LineNumber and LinePosition properties indicating where the error occurred—logging this information is essential if the XML file is large! ReadStartElement verifies that the current NodeType is StartElement, and then calls Read. If you specify a name, it verifies that it matches that of the current element. ReadEndElement verifies that the current NodeType is EndElement, and then calls Read. For instance, we could read this: Jim XmlReader | 461 Other XML as follows: reader.ReadStartElement ("firstname"); Console.WriteLine (reader.Value); reader.Read(); reader.ReadEndElement(); The ReadElementContentAsString method does all of this in one hit. It reads a start element, a text node, and an end element, returning the content as a string: string firstName = reader.ReadElementContentAsString ("firstname", ""); The second argument refers to the namespace, which is blank in this example. There are also typed versions of this method, such as ReadElementContentAsInt, which parse the result. Returning to our original XML document: Jim Bo 500.00 We could read it in as follows: XmlReaderSettings settings = new XmlReaderSettings(); settings.IgnoreWhitespace = true; using (XmlReader r = XmlReader.Create ("customer.xml", settings)) { r.MoveToContent(); // Skip over the XML declaration r.ReadStartElement ("customer"); string firstName = r.ReadElementContentAsString ("firstname", ""); string lastName = r.ReadElementContentAsString ("lastname", ""); decimal creditLimit = r.ReadElementContentAsDecimal ("creditlimit", ""); r.MoveToContent(); // Skip over that pesky comment r.ReadEndElement(); // Read the closing customer tag } The MoveToContent method is really useful. It skips over all the fluff: XML declarations, whitespace, comments, and processing instructions. You can also instruct the reader to do most of this automatically through the properties on XmlReaderSettings. Optional elements In the previous example, suppose that was optional. The solution to this is straightforward: r.ReadStartElement ("customer"); string firstName = r. ReadElementContentAsString ("firstname", ""); string lastName = r.Name == "lastname" ? r.ReadElementContentAsString() : null; decimal creditLimit = r.ReadElementContentAsDecimal ("creditlimit", ""); 462 | Chapter 11: Other XML Technologies Random element order The examples in this section rely on elements appearing in the XML file in a set order. If you need to cope with elements appearing in any order, the easiest solution is to read that section of the XML into an X-DOM. We describe how to do this later in the section “Patterns for Using XmlReader/XmlWriter” on page 469. Empty elements The way that XmlReader handles empty elements presents a horrible trap. Consider the following element: In XML, this is equivalent to: And yet, XmlReader treats the two differently. In the first case, the following code works as expected: reader.ReadStartElement ("customerList"); reader.ReadEndElement(); In the second case, ReadEndElement throws an exception, because there is no separate “end element” as far as XmlReader is concerned. The workaround is to check for an empty element as follows: bool isEmpty = reader.IsEmptyElement; reader.ReadStartElement ("customerList"); if (!isEmpty) reader.ReadEndElement(); In reality, this is a nuisance only when the element in question may contain child elements (such as a customer list). With elements that wrap simple text (such as firstname), you can avoid the whole issue by calling a method such as ReadElement ContentAsString. The ReadElementXXX methods handle both kinds of empty elements correctly. Other ReadXXX methods Table 11-1 summarizes all ReadXXX methods in XmlReader. Most of these are designed to work with elements. The sample XML fragment shown in bold is the section read by the method described. Table 11-1. Read methods Members Works on NodeType Sample XML fragment Input parameters Data returned ReadContentAsXXX Text x x ReadString Text x x ReadElementString Element x x ReadElementContentAsXXX Element x x ReadInnerXml Element x x XmlReader | 463 Other XML Members Works on NodeType Sample XML fragment Input parameters Data returned ReadOuterXml Element x x ReadStartElement Element x ReadEndElement Element x ReadSubtree Element x x ReadToDescendent Element x "b" ReadToFollowing Element x "b" ReadToNextSibling Element x "b" ReadAttributeValue Attribute See “Reading Attributes” on page 465 The ReadContentAsXXX methods parse a text node into type XXX. Internally, the XmlConvert class performs the string-to-type conversion. The text node can be within an element or an attribute. The ReadElementContentAsXXX methods are wrappers around corresponding Read ContentAsXXX methods. They apply to the element node, rather than the text node enclosed by the element. The typed ReadXXX methods also include versions that read base 64 and BinHex formatted data into a byte array. ReadInnerXml is typically applied to an element, and it reads and returns an element and all its descendents. When applied to an attribute, it returns the value of the attribute. ReadOuterXml is the same as ReadInnerXml, except it includes rather than excludes the element at the cursor position. ReadSubtree returns a proxy reader that provides a view over just the current element (and its descendents). The proxy reader must be closed before the original reader can be safely read again. At the point the proxy reader is closed, the cursor position of the original reader moves to the end of the subtree. ReadToDescendent moves the cursor to the start of the first descendent node with the specified name/namespace. ReadToFollowing moves the cursor to the start of the first node—regardless of depth —with the specified name/namespace. ReadToNextSibling moves the cursor to the start of the first sibling node with the specified name/namespace. ReadString and ReadElementString behave like ReadContentAsString and ReadEle mentContentAsString, except that they throw an exception if there’s more than a 464 | Chapter 11: Other XML Technologies single text node within the element. In general, these methods should be avoided because they throw an exception if an element contains a comment. Reading Attributes XmlReader provides an indexer giving you direct (random) access to an element’s attributes—by name or position. Using the indexer is equivalent to calling GetAt tribute. Given the following XML fragment: we could read its attributes as follows: Console.WriteLine (reader ["id"]); // 123 Console.WriteLine (reader ["status"]); // archived Console.WriteLine (reader ["bogus"] == null); // True The XmlReader must be positioned on a start element in order to read attributes. After calling ReadStartElement, the attributes are gone forever! Although attribute order is semantically irrelevant, you can access attributes by their ordinal position. We could rewrite the preceding example as follows: Console.WriteLine (reader [0]); // 123 Console.WriteLine (reader [1]); // archived The indexer also lets you specify the attribute’s namespace—if it has one. AttributeCount returns the number of attributes for the current node. Attribute nodes To explicitly traverse attribute nodes, you must make a special diversion from the normal path of just calling Read. A good reason to do so is if you want to parse attribute values into other types, via the ReadContentAsXXX methods. The diversion must begin from a start element. To make the job easier, the forward- only rule is relaxed during attribute traversal: you can jump to any attribute (forward or backward) by calling MoveToAttribute. MoveToElement returns you to the start element from anyplace within the attribute node diversion. Returning to our previous example: XmlReader | 465 Other XML we can do this: reader.MoveToAttribute ("status"); string status = reader.ReadContentAsString(); reader.MoveToAttribute ("id"); int id = reader.ReadContentAsInt(); MoveToAttribute returns false if the specified attribute doesn’t exist. You can also traverse each attribute in sequence by calling the MoveToFirstAttri bute and then the MoveToNextAttribute methods: if (reader.MoveToFirstAttribute()) do { Console.WriteLine (reader.Name + "=" + reader.Value); } while (reader.MoveToNextAttribute()); // OUTPUT: id=123 status=archived Namespaces and Prefixes XmlReader provides two parallel systems for referring to element and attribute names: • Name • NamespaceURI and LocalName Whenever you read an element’s Name property or call a method that accepts a single name argument, you’re using the first system. This works well if no namespaces or prefixes are present; otherwise, it acts in a crude and literal manner. Namespaces are ignored, and prefixes are included exactly as they were written. For example: Sample fragment Name customer customer x:customer The following code works with the first two cases: reader.ReadStartElement ("customer"); The following is required to handle the third case: reader.ReadStartElement ("x:customer"); The second system works through two namespace-aware properties: NamespaceURI and LocalName. These properties take into account prefixes and default namespaces defined by parent elements. Prefixes are automatically expanded. This means that NamespaceURI always reflects the semantically correct namespace for the current element, and LocalName is always free of prefixes. 466 | Chapter 11: Other XML Technologies When you pass two name arguments into a method such as ReadStartElement, you’re using this same system. For example, consider the following XML:
... We could read this as follows: reader.ReadStartElement ("customer", "DefaultNamespace"); reader.ReadStartElement ("address", "DefaultNamespace"); reader.ReadStartElement ("city", "OtherNamespace"); Abstracting away prefixes is usually exactly what you want. If necessary, you can see what prefix was used through the Prefix property and convert it into a name- space by calling LookupNamespace. XmlWriter XmlWriter is a forward-only writer of an XML stream. The design of XmlWriter is symmetrical to XmlReader. As with XmlTextReader, you construct an XmlWriter by calling Create with an optional settings object. In the following example, we enable indenting to make the output more human-readable, and then write a simple XML file: XmlWriterSettings settings = new XmlWriterSettings(); settings.Indent = true; using (XmlWriter writer = XmlWriter.Create ("..\\..\\foo.xml", settings)) { writer.WriteStartElement ("customer"); writer.WriteElementString ("firstname", "Jim"); writer.WriteElementString ("lastname"," Bo"); writer.WriteEndElement(); } This produces the following document (the same as the file we read in the first example of XmlReader): Jim Bo XmlWriter automatically writes the declaration at the top unless you indicate other- wise in XmlWriterSettings, by setting OmitXmlDeclaration to true or ConformanceLevel to Fragment. The latter also permits writing multiple root nodes —something that otherwise throws an exception. XmlWriter | 467 Other XML The WriteValue method writes a single text node. It accepts both string and nonstring types such as bool and DateTime, internally calling XmlConvert to perform XML- compliant string conversions: writer.WriteStartElement ("birthdate"); writer.WriteValue (DateTime.Now); writer.WriteEndElement(); In contrast, if we call: WriteElementString ("birthdate", DateTime.Now.ToString()); the result would be both non-XML-compliant and vulnerable to incorrect parsing. WriteString is equivalent to calling WriteValue with a string. XmlWriter automatically escapes characters that would otherwise be illegal within an attribute or element, such as &, < >, and extended Unicode characters. Writing Attributes You can write attributes immediately after writing a start element: writer.WriteStartElement ("customer"); writer.WriteAttributeString ("id", "1"); writer.WriteAttributeString ("status", "archived"); To write nonstring values, call WriteStartAttribute, WriteValue, and then WriteEndAttribute. Writing Other Node Types XmlWriter also defines the following methods for writing other kinds of nodes: WriteBase64 // for binary data WriteBinHex // for binary data WriteCData WriteComment WriteDocType WriteEntityRef WriteProcessingInstruction WriteRaw WriteWhitespace WriteRaw directly injects a string into the output stream. There is also a WriteNode method that accepts an XmlReader, echoing everything from the given XmlReader. Namespaces and Prefixes The overloads for the Write* methods allow you to associate an element or attribute with a namespace. Let’s rewrite the contents of the XML file in our previous example. This time we will associate all the elements with the http://oreilly.com namespace, declaring the prefix o at the customer element: writer.WriteStartElement ("o", "customer", "http://oreilly.com"); writer.WriteElementString ("o", "firstname", "http://oreilly.com", "Jim"); 468 | Chapter 11: Other XML Technologies writer.WriteElementString ("o", "lastname", "http://oreilly.com", "Bo"); writer.WriteEndElement(); The output is now as follows: Jim Bo Notice how for brevity XmlWriter omits the child element’s namespace declarations when they are already declared by the parent element. Patterns for Using XmlReader/XmlWriter Working with Hierarchical Data Consider the following classes: public class Contacts { public IList Customers = new List(); public IList Suppliers = new List(); } public class Customer { public string FirstName, LastName; } public class Supplier { public string Name; } Suppose you want to use XmlReader and XmlWriter to serialize a Contacts object to XML as in the following: Jay Dee Kay Gee X Technologies Ltd The best approach is not to write one big method, but to encapsulate XML func- tionality in the Customer and Supplier types themselves by writing ReadXml and WriteXml methods on these types. The pattern in doing so is straightforward: • ReadXml and WriteXml leave the reader/writer at the same depth when they exit. • ReadXml reads the outer element, whereas WriteXml writes only its inner content. Patterns for Using XmlReader/XmlWriter | 469 Other XML Here’s how we would write the Customer type: public class Customer { public const string XmlName = "customer"; public int? ID; public string FirstName, LastName; public Customer () { } public Customer (XmlReader r) { ReadXml (r); } public void ReadXml (XmlReader r) { if (r.MoveToAttribute ("id")) ID = r.ReadContentAsInt(); r.ReadStartElement(); FirstName = r.ReadElementContentAsString ("firstname", ""); LastName = r.ReadElementContentAsString ("lastname", ""); r.ReadEndElement(); } public void WriteXml (XmlWriter w) { if (ID.HasValue) w.WriteAttributeString ("id", "", ID.ToString()); w.WriteElementString ("firstname", FirstName); w.WriteElementString ("lastname", LastName); } } Notice that ReadXml reads the outer start and end element nodes. If its caller did this job instead, Customer couldn’t read its own attributes. The reason for not making WriteXml symmetrical in this regard is twofold: • The caller might need to choose how the outer element is named. • The caller might need to write extra XML attributes, such as the element’s subtype (which could then be used to decide which class to instantiate when reading back the element). Another benefit of following this pattern is that it makes your implementation com- patible with IXmlSerializable (see Chapter 17). The Supplier class is analogous: public class Supplier { public const string XmlName = "supplier"; public string Name; public Supplier () { } public Supplier (XmlReader r) { ReadXml (r); } public void ReadXml (XmlReader r) { r.ReadStartElement(); Name = r.ReadElementContentAsString ("name", ""); r.ReadEndElement(); } 470 | Chapter 11: Other XML Technologies public void WriteXml (XmlWriter w) { w.WriteElementString ("name", Name); } } With the Contacts class, we must enumerate the customers element in ReadXml, checking whether each subelement is a customer or a supplier. We also have to code around the empty element trap: public void ReadXml (XmlReader r) { bool isEmpty = r.IsEmptyElement; // This ensures we don't get r.ReadStartElement(); // snookered by an empty if (isEmpty) return; // element! while (r.NodeType == XmlNodeType.Element) { if (r.Name == Customer.XmlName) Customers.Add (new Customer (r)); else if (r.Name == Supplier.XmlName) Suppliers.Add (new Supplier (r)); else throw new XmlException ("Unexpected node: " + r.Name); } r.ReadEndElement(); } public void WriteXml (XmlWriter w) { foreach (Customer c in Customers) { w.WriteStartElement (Customer.XmlName); c.WriteXml (w); w.WriteEndElement(); } foreach (Supplier s in Suppliers) { w.WriteStartElement (Supplier.XmlName); s.WriteXml (w); w.WriteEndElement(); } } Mixing XmlReader/XmlWriter with an X-DOM You can fly in an X-DOM at any point in the XML tree where XmlReader or XmlWriter becomes too cumbersome. Using the X-DOM to handle inner elements is an excellent way to combine X-DOM’s ease of use with the low-memory footprint of XmlReader and XmlWriter. Using XmlReader with XElement To read the current element into an X-DOM, you call XNode.ReadFrom, passing in the XmlReader. Unlike XElement.Load, this method is not “greedy” in that it doesn’t expect to see a whole document. Instead, it reads just the end of the current subtree. Patterns for Using XmlReader/XmlWriter | 471 Other XML For instance, suppose we have an XML logfile structured as follows: ... ... ... ... If there were a million logentry elements, reading the whole thing into an X-DOM would waste memory. A better solution is to traverse each logentry with an XmlReader, and then use XElement to process the elements individually: XmlReaderSettings settings = new XmlReaderSettings(); settings.IgnoreWhitespace = true; using (XmlReader r = XmlReader.Create ("logfile.xml", settings)) { r.ReadStartElement ("log"); while (r.Name == "logentry") { XElement logEntry = (XElement) XNode.ReadFrom (r); int id = (int) logEntry.Attribute ("id"); DateTime date = (DateTime) logEntry.Element ("date"); string source = (string) logEntry.Element ("source"); ... } r.ReadEndElement(); } If you follow the pattern described in the previous section, you can slot an XEle ment into a custom type’s ReadXml or WriteXml method without the caller ever know- ing you’ve cheated! For instance, we could rewrite Customer’s ReadXml method as follows: public void ReadXml (XmlReader r) { XElement x = (XElement) XNode.ReadFrom (r); FirstName = (string) x.Element ("firstname"); LastName = (string) x.Element ("lastname"); } XElement collaborates with XmlReader to ensure that namespaces are kept intact and prefixes are properly expanded—even if defined at an outer level. So, if our XML file read like this: ... the XElements we constructed at the logentry level would correctly inherit the outer namespace. 472 | Chapter 11: Other XML Technologies Using XmlWriter with XElement You can use an XElement just to write inner elements to an XmlWriter. The following code writes a million logentry elements to an XML file using XElement—without storing the whole thing in memory: using (XmlWriter w = XmlWriter.Create ("log.xml")) { w.WriteStartElement ("log"); for (int i = 0; i < 1000000; i++) { XElement e = new XElement ("logentry", new XAttribute ("id", i), new XElement ("date", DateTime.Today.AddDays (-1)), new XElement ("source", "test")); e.WriteTo (w); } w.WriteEndElement (); } Using an XElement incurs minimal execution overhead. If we amend this example to use XmlWriter throughout, there’s no measurable difference in execution time. XmlDocument XmlDocument is an in-memory representation of an XML document, which has since been superseded by the LINQ-to-XML DOM. XmlDocument’s object model and the methods that its types expose conform to a pattern defined by the W3C. So, if you’re familiar with another W3C-compliant XML DOM (e.g., in Java), you’ll be at home with XmlDocument. When compared to the more modern LINQ-to-XML DOM, how- ever, the W3C model is much clumsier to use. XmlDocument is unavailable in the Metro profile. However, a similar DOM is exposed in the WinRT namespace Win dows.Data.Xml.Dom. The base type for all objects in an XmlDocument tree is XmlNode. The following types derive from XmlNode: XmlNode XmlDocument XmlDocumentFragment XmlEntity XmlNotation XmlLinkedNode XmlLinkedNode exposes NextSibling and PreviousSibling properties and is an abstract base for the following subtypes: XmlLinkedNode XmlCharacterData XmlDeclaration XmlDocument | 473 Other XML XmlDocumentType XmlElement XmlEntityReference XmlProcesingInstruction Loading and Saving an XmlDocument To load an XmlDocument from an existing source, you instantiate an XmlDocument and then call Load or LoadXml: • Load accepts a filename, Stream, TextReader, or XmlReader. • LoadXml accepts a literal XML string. To save a document, call Save with a filename, Stream, TextWriter, or XmlWriter: XmlDocument doc = new XmlDocument(); doc.Load ("customer1.xml"); doc.Save ("customer2.xml"); Traversing an XmlDocument To illustrate traversing an XmlDocument, we’ll use the following XML file: Jim Bo The ChildNodes property (defined in XNode) allows you to descend into the tree structure. This returns an indexable collection: XmlDocument doc = new XmlDocument(); doc.Load ("customer.xml"); Console.WriteLine (doc.DocumentElement.ChildNodes[0].InnerText); // Jim Console.WriteLine (doc.DocumentElement.ChildNodes[1].InnerText); // Bo With the ParentNode property, you can ascend back up the tree: Console.WriteLine ( doc.DocumentElement.ChildNodes[1].ParentNode.Name); // customer The following properties also help traverse the document (all of which return null if the node does not exist): FirstChild LastChild NextSibling PreviousSibling The following two statements both output firstname: Console.WriteLine (doc.DocumentElement.FirstChild.Name); Console.WriteLine (doc.DocumentElement.LastChild.PreviousSibling.Name); 474 | Chapter 11: Other XML Technologies XmlNode exposes an Attributes property for accessing attributes either by name (and namespace) or by ordinal position. For example: Console.WriteLine (doc.DocumentElement.Attributes ["id"].Value); InnerText and InnerXml The InnerText property represents the concatenation of all child text nodes. The following two lines both output Jim, since our XML document contains only a single text node: Console.WriteLine (doc.DocumentElement.ChildNodes[0].InnerText); Console.WriteLine (doc.DocumentElement.ChildNodes[0].FirstChild.Value); Setting the InnerText property replaces all child nodes with a single text node. Be careful when setting InnerText to not accidentally wipe over element nodes. For example: doc.DocumentElement.ChildNodes[0].InnerText = "Jo"; // wrong doc.DocumentElement.ChildNodes[0].FirstChild.InnerText = "Jo"; // right The InnerXml property represents the XML fragment within the current node. You typically use InnerXml on elements: Console.WriteLine (doc.DocumentElement.InnerXml); // OUTPUT: JimBo InnerXml throws an exception if the node type cannot have children. Creating and Manipulating Nodes To create and add new nodes: 1. Call one of the CreateXXX methods on the XmlDocument, such as CreateElement. 2. Add the new node into the tree by calling AppendChild, PrependChild, InsertBe fore, or InsertAfter on the desired parent node. Creating nodes requires that you first have an XmlDocument— you cannot simply instantiate an XmlElement on its own like with the X-DOM. Nodes rely on a host XmlDocument for sustenance. For example: XmlDocument doc = new XmlDocument(); XmlElement customer = doc.CreateElement ("customer"); doc.AppendChild (customer); The following creates a document matching the XML we started with earlier in this chapter in the section “XmlReader” on page 458: XmlDocument doc = new XmlDocument (); doc.AppendChild (doc.CreateXmlDeclaration ("1.0", null, "yes")); XmlDocument | 475 Other XML XmlAttribute id = doc.CreateAttribute ("id"); XmlAttribute status = doc.CreateAttribute ("status"); id.Value = "123"; status.Value = "archived"; XmlElement firstname = doc.CreateElement ("firstname"); XmlElement lastname = doc.CreateElement ("lastname"); firstname.AppendChild (doc.CreateTextNode ("Jim")); lastname.AppendChild (doc.CreateTextNode ("Bo")); XmlElement customer = doc.CreateElement ("customer"); customer.Attributes.Append (id); customer.Attributes.Append (status); customer.AppendChild (lastname); customer.AppendChild (firstname); doc.AppendChild (customer); You can construct the tree in any order. In the previous example, it doesn’t matter if you rearrange the order of the lines that append child nodes. To remove a node, you call RemoveChild, ReplaceChild, or RemoveAll. Namespaces See Chapter 10 for an introduction to XML namespaces and prefixes. The CreateElement and CreateAttribute methods are overloaded to let you specify a namespace and prefix: CreateXXX (string name); CreateXXX (string name, string namespaceURI); CreateXXX (string prefix, string localName, string namespaceURI); The name parameter refers to either a local name (i.e., no prefix) or a name qualified with a prefix. The namespaceURI parameter is used if and only if you are declaring (rather than merely referring to) a namespace. Here is an example of declaring a namespace with a prefix while creating an element: XmlElement customer = doc.CreateElement ("o", "customer", "http://oreilly.com"); Here is an example of referring to a namespace with a prefix while creating an element: XmlElement customer = doc.CreateElement ("o:firstname"); In the next section, we will explain how to deal with namespaces when writing XPath queries. 476 | Chapter 11: Other XML Technologies XPath XPath is the W3C standard for XML querying. In the .NET Framework, XPath can query an XmlDocument rather like LINQ queries an X-DOM. XPath has a wider scope, though, in that it’s also used by other XML technologies, such as XML schema, XLST, and XAML. XPath is unavailable in the Metro .NET profile. XPath queries are expressed in terms of the XPath 2.0 Data Model. Both the DOM and the XPath Data Model represent an XML document as a tree. The difference is that the XPath Data Model is purely data-centric, abstracting away the formatting aspects of XML text. For example, CDATA sections are not required in the XPath Data Model, since the only reason CDATA sections exist is to enable text to contain markup char- acter sequences. The XPath specification is at http://www.w3 .org/tr/xpath20/. The examples in this section all use the following XML file: Jim Bo Thomas Jefferson You can write XPath queries within code in the following ways: • Call one of the SelectXXX methods on an XmlDocument or XmlNode. • Spawn an XPathNavigator from either: — An XmlDocument — An XPathDocument • Call an XPathXXX extension method on an XNode. The SelectXXX methods accept an XPath query string. For example, the following finds the firstname node of an XmlDocument: XmlDocument doc = new XmlDocument(); doc.Load ("customers.xml"); XmlNode n = doc.SelectSingleNode ("customers/customer[firstname='Jim']"); Console.WriteLine (n.InnerText); // JimBo The SelectXXX methods delegate their implementation to XPathNavigator, which you can also use directly—over either an XmlDocument or a read-only XPathDocument. XPath | 477 Other XML You can also execute XPath queries over an X-DOM, via extension methods defined in System.Xml.XPath: XDocument doc = XDocument.Load (@"Customers.xml"); XElement e = e.XPathSelectElement ("customers/customer[firstname='Jim']"); Console.WriteLine (e.Value); // JimBo The extension methods for use with XNodes are: CreateNavigator XPathEvaluate XPathSelectElement XPathSelectElements Common XPath Operators The XPath specification is huge. However, you can get by knowing just a few oper- ators (see Table 11-2), just as you can play a lot of songs knowing just three chords. Table 11-2. Common XPath operators Operator Description / Children // Recursively children . Current node (usually implied) .. Parent node * Wildcard @ Attribute [] Filter : Namespace separator To find the customers node: XmlNode node = doc.SelectSingleNode ("customers"); The / symbol queries child nodes. To select the customer nodes: XmlNode node = doc.SelectSingleNode ("customers/customer"); The // operator includes all child nodes, regardless of nesting level. To select all lastname nodes: XmlNodeList nodes = doc.SelectNodes ("//lastname"); The .. operator selects parent nodes. This example is a little silly because we’re starting from the root anyway, but it serves to illustrate the functionality: XmlNodeList nodes = doc.SelectNodes ("customers/customer..customers"); The * operator selects nodes regardless of name. The following selects the child nodes of customer, regardless of name: XmlNodeList nodes = doc.SelectNodes ("customers/customer/*"); 478 | Chapter 11: Other XML Technologies The @ operator selects attributes. * can be used as a wildcard. Here is how to select the id attribute: XmlNode node = doc.SelectSingleNode ("customers/customer/@id"); The [] operator filters a selection, in conjunction with the operators =, !=, <, >, not(), and, and or. In this example, we filter on firstname: XmlNode n = doc.SelectSingleNode ("customers/customer[firstname='Jim']"); The : operator qualifies a namespace. Had the customers element been qualified with the x namespace, we would access it as follows: XmlNode node = doc.SelectSingleNode ("x:customers"); XPathNavigator XPathNavigator is a cursor over the XPath Data Model representation of an XML document. It is loaded with primitive methods that move the cursor around the tree (e.g., move to parent, move to first child, etc.). The XPathNavigator’s Select* meth- ods take an XPath string to express more complex navigations or queries that return multiple nodes. Spawn instances of XPathNavigator from an XmlDocument, an XPathDocument, or another XPathNavigator. Here is an example of spawning an XPathNavigator from an XmlDoument: XPathNavigator nav = doc.CreateNavigator(); XPathNavigator jim = nav.SelectSingleNode ( "customers/customer[firstname='Jim']" ); Console.WriteLine (jim.Value); // JimBo In the XPath Data Model, the value of a node is the concatenation of the text ele- ments, equivalent to XmlDocument’s InnerText property. The SelectSingleNode method returns a single XPathNavigator. The Select method returns an XPathNodeIterator, which simply iterates over multiple XPathNavigators. For example: XPathNavigator nav = doc.CreateNavigator(); string xPath = "customers/customer/firstname/text()"; foreach (XPathNavigator navC in nav.Select (xPath)) Console.WriteLine (navC.Value); OUTPUT: Jim Thomas To perform faster queries, you can compile an XPath query into an XPathExpres sion. You then pass the compiled expression to a Select* method, instead of a string. For example: XPathNavigator nav = doc.CreateNavigator(); XPathExpression expr = nav.Compile ("customers/customer/firstname"); XPath | 479 Other XML foreach (XPathNavigator a in nav.Select (expr)) Console.WriteLine (a.Value); OUTPUT: Jim Thomas Querying with Namespaces Querying elements and attributes that contain namespaces requires some extra unintuitive steps. Consider the following XML file: Jim Bo Thomas Jefferson The following query will fail, despite qualifying the nodes with the prefix o: XmlDocument doc = new XmlDocument(); doc.Load ("customers.xml"); XmlNode n = doc.SelectSingleNode ("o:customers/o:customer"); Console.WriteLine (n.InnerText); // JimBo To make this query work, you must first create an XmlNamespaceManager instance as follows: XmlNamespaceManager xnm = new XmlNamespaceManager (doc.NameTable); You can treat NameTable as a black box (XmlNamespaceManager uses it internally to cache and reuse strings). Once we create the namespace manager, we can add prefix/ namespace pairs to it as follows: xnm.AddNamespace ("o", "http://oreilly.com"); The Select* methods on XmlDocument and XPathNavigator have overloads that accept an XmlNamespaceManager. We can successfully rewrite the previous query as follows: XmlNode n = doc.SelectSingleNode ("o:customers/o:customer", xnm); XPathDocument XPathDocument is used for read-only XML documents that conform to the W3C XPath Data Model. An XPathNavigator backed by an XPathDocument is faster than an XmlDocument, but it cannot make changes to the underlying document: XPathDocument doc = new XPathDocument ("customers.xml"); XPathNavigator nav = doc.CreateNavigator(); foreach (XPathNavigator a in nav.Select ("customers/customer/firstname")) 480 | Chapter 11: Other XML Technologies Console.WriteLine (a.Value); OUTPUT: Jim Thomas XSD and Schema Validation The content of a particular XML document is nearly always domain-specific, such as a Microsoft Word document, an application configuration document, or a web service. For each domain, the XML file conforms to a particular pattern. There are several standards for describing the schema of such a pattern, to standardize and automate the interpretation and validation of XML documents. The most widely accepted standard is XSD, short for XML Schema Definition. Its precursors, DTD and XDR, are also supported by System.Xml. Consider the following XML document: Jim Bo Thomas Jefferson We can write an XSD for this document as follows: XSD and Schema Validation | 481 Other XML As you can see, XSD documents are themselves written in XML. Furthermore, an XSD document is describable with XSD—you can find that definition at http://www .w3.org/2001/xmlschema.xsd. Performing Schema Validation You can validate an XML file or document against one or more schemas before reading or processing it. There are a number of reasons to do so: • You can get away with less error checking and exception handling. • Schema validation picks up errors you might otherwise overlook. • Error messages are detailed and informative. To perform validation, plug a schema into an XmlReader, an XmlDocument, or an X-DOM object, and then read or load the XML as you would normally. Schema validation happens automatically as content is read, so the input stream is not read twice. Validating with an XmlReader Here’s how to plug a schema from the file customers.xsd into an XmlReader: XmlReaderSettings settings = new XmlReaderSettings(); settings.ValidationType = ValidationType.Schema; settings.Schemas.Add (null, "customers.xsd"); using (XmlReader r = XmlReader.Create ("customers.xml", settings)) ... If the schema is inline, set the following flag instead of adding to Schemas: settings.ValidationFlags |= XmlSchemaValidationFlags.ProcessInlineSchema; You then Read as you would normally. If schema validation fails at any point, an XmlSchemaValidationException is thrown. Calling Read on its own validates both elements and attributes: you don’t need to navigate to each individual attribute for it to be validated. If you want only to validate the document, you can do this: using (XmlReader r = XmlReader.Create ("customers.xml", settings)) try { while (r.Read()) ; } catch (XmlSchemaValidationException ex) { ... } XmlSchemaValidationException has properties for the error Message, LineNumber, and LinePosition. In this case, it only tells you about the first error in the document. If 482 | Chapter 11: Other XML Technologies you want to report on all errors in the document, you instead must handle the ValidationEventHandler event: XmlReaderSettings settings = new XmlReaderSettings(); settings.ValidationType = ValidationType.Schema; settings.Schemas.Add (null, "customers.xsd"); settings.ValidationEventHandler += ValidationHandler; using (XmlReader r = XmlReader.Create ("customers.xml", settings)) while (r.Read()) ; When you handle this event, schema errors no longer throw exceptions. Instead, they fire your event handler: static void ValidationHandler (object sender, ValidationEventArgs e) { Console.WriteLine ("Error: " + e.Exception.Message); } The Exception property of ValidationEventArgs contains the XmlSchemaValidatio nException that would have otherwise been thrown. The System.Xml namespace also contains a class called XmlVali datingReader. This was used to perform schema validation prior to Framework 2.0, and it is now deprecated. Validating an X-DOM or XmlDocument To validate an XML file or stream while reading into an X-DOM or XmlDocument, you create an XmlReader, plug in the schemas, and then use the reader to load the DOM: XmlReaderSettings settings = new XmlReaderSettings(); settings.ValidationType = ValidationType.Schema; settings.Schemas.Add (null, "customers.xsd"); XDocument doc; using (XmlReader r = XmlReader.Create ("customers.xml", settings)) try { doc = XDocument.Load (r); } catch (XmlSchemaValidationException ex) { ... } XmlDocument xmlDoc = new XmlDocument(); using (XmlReader r = XmlReader.Create ("customers.xml", settings)) try { xmlDoc.Load (r); } catch (XmlSchemaValidationException ex) { ... } You can also validate an XDocument or XElement that’s already in memory, by calling extension methods in System.Xml.Schema. These methods accept an XmlSchemaSet (a collection of schemas) and a validation event handler: XDocument doc = XDocument.Load (@"customers.xml"); XmlSchemaSet set = new XmlSchemaSet (); set.Add (null, @"customers.xsd"); StringBuilder errors = new StringBuilder (); doc.Validate (set, (sender, args) => { errors.AppendLine XSD and Schema Validation | 483 Other XML (args.Exception.Message); } ); Console.WriteLine (errors.ToString()); To validate an XmlDocument already in memory, add the schema(s) to the XmlDocu ment’s Schemas collection and then call the document’s Validate method, passing in a ValidationEventHandler to process the errors. XSLT XSLT stands for Extensible Stylesheet Language Transformations. It is an XML lan- guage that describes how to transform one XML language into another. The quintessential example of such a transformation is transforming an XML document (that typically describes data) into an XHTML document (that describes a formatted document). Consider the following XML file: Jim Bo The following XSLT file describes such a transformation:

The output is as follows:



The System.Xml.Xsl.XslCompiledTransform transform class efficiently performs XLST transforms. It renders XmlTransform obsolete. XmlTransform works very simply: XslCompiledTransform transform = new XslCompiledTransform(); transform.Load ("test.xslt"); transform.Transform ("input.xml", "output.xml"); Generally, it’s more useful to use the overload of Transform that accepts an XmlWriter rather than an output file, so you can control the formatting. 484 | Chapter 11: Other XML Technologies 12Disposal and Garbage Collection Some objects require explicit teardown code to release resources such as open files, locks, operating system handles, and unmanaged objects. In .NET parlance, this is called disposal, and it is supported through the IDisposable interface. The managed memory occupied by unused objects must also be reclaimed at some point; this function is known as garbage collection and is performed by the CLR. Disposal differs from garbage collection in that disposal is usually explicitly insti- gated; garbage collection is totally automatic. In other words, the programmer takes care of such things as releasing file handles, locks, and operating system resources while the CLR takes care of releasing memory. This chapter discusses both disposal and garbage collection, also describing C# finalizers and the pattern by which they can provide a backup for disposal. Lastly, we discuss the intricacies of the garbage collector and other memory management options. IDisposable, Dispose, and Close The .NET Framework defines a special interface for types requiring a tear-down method: public interface IDisposable { void Dispose(); } C#’s using statement provides a syntactic shortcut for calling Dispose on objects that implement IDisposable, using a try/finally block. For example: using (FileStream fs = new FileStream ("myFile.txt", FileMode.Open)) { // ... Write to the file ... } 485 The compiler converts this to: FileStream fs = new FileStream ("myFile.txt", FileMode.Open); try { // ... Write to the file ... } finally { if (fs != null) ((IDisposable)fs).Dispose(); } The finally block ensures that the Dispose method is called even when an exception is thrown,1 or the code exits the block early. In simple scenarios, writing your own disposable type is just a matter of implement- ing IDisposable and writing the Dispose method: sealed class Demo : IDisposable { public void Dispose() { // Perform cleanup / tear-down. ... } } This pattern works well in simple cases and is appropriate for sealed classes. In “Calling Dispose from a Final- izer” on page 494, we’ll describe a more elaborate pattern that can provide a backup for consumers that forget to call Dis pose. With unsealed types, there’s a strong case for following this latter pattern from the outset—otherwise, it becomes very messy if the subtype wants to add such functionality itself. Standard Disposal Semantics The Framework follows a de facto set of rules in its disposal logic. These rules are not hard-wired to the Framework or C# language in any way; their purpose is to define a consistent protocol to consumers. Here they are: 1. Once disposed, an object is beyond redemption. It cannot be reactivated, and calling its methods or properties throws an ObjectDisposedException. 2. Calling an object’s Dispose method repeatedly causes no error. 3. If disposable object x contains or “wraps” or “possesses” disposable object y, x’s Dispose method automatically calls y’s Dispose method—unless instructed otherwise. 1. In “Interrupt and Abort” on page 909 in Chapter 22, we describe how aborting a thread can violate the safety of this pattern. This is rarely an issue in practice because aborting threads is widely discouraged for precisely this (and other) reasons. 486 | Chapter 12: Disposal and Garbage Collection These rules are also helpful when writing your own types, though not mandatory. Nothing prevents you from writing an “Undispose” method, other than, perhaps, the flak you might cop from colleagues! According to rule 3, a container object automatically disposes its child objects. A good example is a Windows container control such as a Form or Panel. The container may host many child controls, yet you don’t dispose every one of them explicitly: closing or disposing the parent control or form takes care of the whole lot. Another example is when you wrap a FileStream in a DeflateStream. Disposing the Deflate Stream also disposes the FileStream—unless you instructed otherwise in the constructor. Close and Stop Some types define a method called Close in addition to Dispose. The Framework is not completely consistent on the semantics of a Close method, although in nearly all cases it’s either: • Functionally identical to Dispose • A functional subset of Dispose An example of the latter is IDbConnection: a Closed connection can be re-Opened; a Disposed connection cannot. Another example is a Windows Form activated with ShowDialog: Close hides it; Dispose releases its resources. Some classes define a Stop method (e.g., Timer or HttpListener). A Stop method may release unmanaged resources, like Dispose, but unlike Dispose, it allows for re-Starting. With WinRT, Close is considered identical to Dispose—in fact, the runtime projects methods called Close into methods called Dispose, to make their types friendly to using statements. When to Dispose A safe rule to follow (in nearly all cases) is “if in doubt, dispose.” A disposable object —if it could talk—would say the following: When you’ve finished with me, let me know. If simply abandoned, I might cause trouble for other object instances, the application domain, the computer, the network, or the database! Objects wrapping an unmanaged resource handle will nearly always require dis- posal, in order to free the handle. Examples include Windows Forms controls, file or network streams, network sockets, GDI+ pens, brushes, and bitmaps. Con- versely, if a type is disposable, it will often (but not always) reference an unmanaged handle, directly or indirectly. This is because unmanaged handles provide the gate- way to the “outside world” of operating system resources, network connections, database locks—the primary means by which objects can create trouble outside of themselves if improperly abandoned. IDisposable, Dispose, and Close | 487 Disposal and Garbage Collection There are, however, three scenarios for not disposing: • When obtaining a shared object via a static field or property • When an object’s Dispose method does something that you don’t want • When an object’s Dispose method is unnecessary by design, and disposing that object would add complexity to your program The first category is rare. The main cases are in the System.Drawing namespace: the GDI+ objects obtained through static fields or properties (such as Brushes.Blue) must never be disposed because the same instance is used throughout the life of the application. Instances that you obtain through constructors, however (such as new SolidBrush), should be disposed, as should instances obtained through static meth- ods (such as Font.FromHdc). The second category is more common. There are some good examples in the System.IO and System.Data namespaces: Type Disposal function When not to dispose MemoryStream Prevents further I/O When you later need to read/write the stream StreamReader, StreamWriter Flushes the reader/writer and closes the underlying stream When you want to keep the underlying stream open (you must instead call Flush on a StreamWriter when you’re done) IDbConnection Releases a database connection and clears the connection string If you need to re-Open it, you should call Close instead of Dispose DataContext (LINQ to SQL) Prevents further use When you might have lazily evaluated queries connected to that context MemoryStream’s Dispose method disables only the object; it doesn’t perform any crit- ical cleanup because a MemoryStream holds no unmanaged handles or other such resources. The third category includes the following classes: WebClient, StringReader, String Writer, and BackgroundWorker (in System.ComponentModel). These types are disposa- ble under the duress of their base class rather than through a genuine need to perform essential cleanup. If you happen to instantiate and work with such an object entirely in one method, wrapping it in a using block adds little inconvenience. But if the object is longer-lasting, keeping track of when it’s no longer used so that you can dispose of it adds unnecessary complexity. In such cases, you can simply ignore object disposal. Opt-in Disposal Because IDisposable makes a type tractable with C#’s using construct, there’s a temptation to extend the reach of IDisposable to nonessential activities. For instance: public sealed class HouseManager : IDisposable { public void Dispose() 488 | Chapter 12: Disposal and Garbage Collection { CheckTheMail(); } ... } The idea is that a consumer of this class can choose to circumvent the nonessential cleanup—simply by not calling Dispose. This, however, relies on the consumer knowing what’s inside HouseManager’s Dispose method. It also breaks if essential cleanup activity is later added: public void Dispose() { CheckTheMail(); // Nonessential LockTheHouse(); // Essential } The solution to this problem is the opt-in disposal pattern: public sealed class HouseManager : IDisposable { public readonly bool CheckMailOnDispose; public Demo (bool checkMailOnDispose) { CheckMailOnDispose = checkMailOnDispose; } public void Dispose() { if (CheckMailOnDispose) CheckTheMail(); LockTheHouse(); } ... } The consumer can then always call Dispose—providing simplicity and avoiding the need for special documentation or reflection. An example of where this pattern is implemented is in the DeflateStream class, in System.IO.Compression. Here’s its constructor: public DeflateStream (Stream stream, CompressionMode mode, bool leaveOpen) The nonessential activity is closing the inner stream (the first parameter) upon dis- posal. There are times when you want to leave the inner stream open and yet still dispose the DeflateStream to perform its essential tear-down activity (flushing buf- fered data). This pattern might look simple, yet until Framework 4.5, it escaped StreamReader and StreamWriter (in the System.IO namespace). The result is messy: StreamWriter must expose another method (Flush) to perform essential cleanup for consumers not calling Dispose. (Framework 4.5 now exposes a constructor on these classes that lets you keep the stream open.) The CryptoStream class in System.Security.Cryptog raphy suffers a similar problem and requires that you call FlushFinalBlock to tear it down while keeping the inner stream open. IDisposable, Dispose, and Close | 489 Disposal and Garbage Collection You could describe this as an ownership issue. The question for a disposable object is: do I really own the underlying resource that I’m using? Or am I just renting it from someone else who manages both the underlying resource lifetime and, by some undocumented contract, my lifetime? Following the opt-in pattern avoids this problem by making the ownership contract documented and explicit. Clearing Fields in Disposal In general, you don’t need to clear an object’s fields in its Dispose method. However, it is good practice to unsubscribe from events that the object has subscribed to internally over its lifetime (see “Managed Memory Leaks” on page 501 for an ex- ample). Unsubscribing from such events avoids receiving unwanted event notifica- tions—and avoids unintentionally keeping the object alive in the eyes of the garbage collector (GC). A Dispose method itself does not cause (managed) memory to be released—this can happen only in garbage collection. It’s also worth setting a field to indicate that the object is disposed so that you can throw an ObjectDisposedException if a consumer later tries to call members on the object. A good pattern is to use a publicly readable automatic property for this: public bool IsDisposed { get; private set; } Although technically unnecessary, it can also be good to clear an object’s own event handlers (by setting them to null) in the Dispose method. This eliminates the pos- sibility of those events firing during or after disposal. Occasionally, an object holds high-value secrets, such as encryption keys. In these cases, it can make sense to clear such data from fields during disposal (to avoid discovery by less privileged assemblies or malware). The SymmetricAlgorithm class in System.Security.Cryptography does exactly this, by calling Array.Clear on the byte array holding the encryption key. Automatic Garbage Collection Regardless of whether an object requires a Dispose method for custom tear-down logic, at some point the memory it occupies on the heap must be freed. The CLR handles this side of it entirely automatically, via an automatic GC. You never deal- locate managed memory yourself. For example, consider the following method: public void Test() { byte[] myArray = new byte[1000]; 490 | Chapter 12: Disposal and Garbage Collection ... } When Test executes, an array to hold 1,000 bytes is allocated on the memory heap. The array is referenced by the variable myArray, stored on the local variable stack. When the method exits, this local variable myArray pops out of scope, meaning that nothing is left to reference the array on the memory heap. The orphaned array then becomes eligible to be reclaimed in garbage collection. In debug mode with optimizations disabled, the lifetime of an object referenced by a local variable extends to the end of the code block to ease debugging. Otherwise, it becomes eligible for collection at the earliest point at which it’s no longer used. Garbage collection does not happen immediately after an object is orphaned. Rather like garbage collection on the street, it happens periodically, although (unlike garbage collection on the street) not to a fixed schedule. The CLR bases its decision on when to collect upon a number of factors, such as the available memory, the amount of memory allocation, and the time since the last collection. This means that there’s an indeterminate delay between an object being orphaned and being released from memory. This delay can range from nanoseconds to days. The GC doesn’t collect all garbage with every collection. Instead, the memory manager divides objects into generations and the GC collects new generations (recently allocated objects) more frequently than old generations (long-lived objects). We’ll discuss this in more detail in “How the Garbage Collector Works” on page 497. Garbage Collection and Memory Consumption The GC tries to strike a balance between the time it spends doing garbage collec- tion and the application’s memory consumption (working set). Consequently, applications can consume more memory than they need, particularly if large tem- porary arrays are constructed. You can monitor a process’s memory consumption via the Windows Task Man- ager or Resource Monitor—or programmatically by querying a performance counter: // These types are in System.Diagnostics: string procName = Process.GetCurrentProcess().ProcessName; using (PerformanceCounter pc = new PerformanceCounter ("Process", "Private Bytes", procName)) Console.WriteLine (pc.NextValue()); This queries the private working set, which gives the best overall indication of your program’s memory consumption. Specifically, it excludes memory that the CLR Automatic Garbage Collection | 491 Disposal and Garbage Collection has internally deallocated and is willing to rescind to the operating system should another process need it. Roots A root is something that keeps an object alive. If an object is not directly or indirectly referenced by a root, it will be eligible for garbage collection. A root is one of the following: • A local variable or parameter in an executing method (or in any method in its call stack) • A static variable • An object on the queue that stores objects ready for finalization (see next section) It’s impossible for code to execute in a deleted object, so if there’s any possibility of an (instance) method executing, its object must somehow be referenced in one of these ways. Note that a group of objects that reference each other cyclically are considered dead without a root referee (see Figure 12-1). To put it in another way, objects that cannot be accessed by following the arrows (references) from a root object are unreachable —and therefore subject to collection. Figure 12-1. Roots 492 | Chapter 12: Disposal and Garbage Collection Garbage Collection and WinRT Windows Runtime relies on COM’s reference-counting mechanism to release mem- ory instead of depending on an automatic garbage collector. Despite this, WinRT objects that you instantiate from C# have their lifetime managed by the CLR’s garbage collector, because the CLR mediates access to the COM object through an object that it creates behind the scenes called a runtime callable wrapper (Chapter 24). Finalizers Prior to an object being released from memory, its finalizer runs, if it has one. A finalizer is declared like a constructor, but it is prefixed by the ˜ symbol: class Test { ˜Test() { // Finalizer logic... } } (Although similar in declaration to a constructor, finalizers cannot be declared as public or static, cannot have parameters, and cannot call the base class.) Finalizers are possible because garbage collection works in distinct phases. First, the GC identifies the unused objects ripe for deletion. Those without finalizers are deleted right away. Those with pending (unrun) finalizers are kept alive (for now) and are put onto a special queue. At that point, garbage collection is complete, and your program continues executing. The finalizer thread then kicks in and starts running in parallel to your program, picking objects off that special queue and running their finalization methods. Prior to each object’s finalizer running, it’s still very much alive—that queue acts as a root object. Once it’s been dequeued and the finalizer executed, the object becomes orphaned and will get deleted in the next collection (for that object’s generation). Finalizers can be useful, but they come with some provisos: • Finalizers slow the allocation and collection of memory (the GC needs to keep track of which finalizers have run). • Finalizers prolong the life of the object and any referred objects (they must all await the next garbage truck for actual deletion). • It’s impossible to predict in what order the finalizers for a set of objects will be called. • You have limited control over when the finalizer for an object will be called. • If code in a finalizer blocks, other objects cannot get finalized. • Finalizers may be circumvented altogether if an application fails to unload cleanly. Finalizers | 493 Disposal and Garbage Collection In summary, finalizers are somewhat like lawyers—although there are cases in which you really need them, in general you don’t want to use them unless absolutely nec- essary. If you do use them, you need to be 100% sure you understand what they are doing for you. Here are some guidelines for implementing finalizers: • Ensure that your finalizer executes quickly. • Never block in your finalizer (Chapter 14). • Don’t reference other finalizable objects. • Don’t throw exceptions. An object’s finalizer can get called even if an exception is thrown during construction. For this reason, it pays not to assume that fields are correctly initialized when writing a finalizer. Calling Dispose from a Finalizer A popular pattern is to have the finalizer call Dispose. This makes sense when cleanup is non-urgent and hastening it by calling Dispose is more of an optimization than a necessity. Bear in mind that with this pattern, you couple memory deal- location to resource deallocation—two things with potentially divergent interests (unless the resource is itself memory). You also increase the burden on the finalization thread. This pattern can also be used as a backup for cases when a con- sumer simply forgets to call Dispose. However, it’s then a good idea to log the failure so that you can fix the bug. There’s a standard pattern for implementing this, as follows: class Test : IDisposable { public void Dispose() // NOT virtual { Dispose (true); GC.SuppressFinalize (this); // Prevent finalizer from running. } protected virtual void Dispose (bool disposing) { if (disposing) { // Call Dispose() on other objects owned by this instance. // You can reference other finalizable objects here. // ... } 494 | Chapter 12: Disposal and Garbage Collection // Release unmanaged resources owned by (just) this object. // ... } ˜Test() { Dispose (false); } } Dispose is overloaded to accept a bool disposing flag. The parameterless version is not declared as virtual and simply calls the enhanced version with true. The enhanced version contains the actual disposal logic and is protected and vir tual; this provides a safe point for subclasses to add their own disposal logic. The disposing flag means it’s being called “properly” from the Dispose method rather than in “last-resort mode” from the finalizer. The idea is that when called with disposing set to false, this method should not, in general, reference other objects with finalizers (because such objects may themselves have been finalized and so be in an unpredictable state). This rules out quite a lot! Here are a couple of tasks it can still perform in last-resort mode, when disposing is false: • Releasing any direct references to operating system resources (obtained, per- haps, via a P/Invoke call to the Win32 API) • Deleting a temporary file created on construction To make this robust, any code capable of throwing an exception should be wrapped in a try/catch block, and the exception, ideally, logged. Any logging should be as simple and robust as possible. Notice that we call GC.SuppressFinalize in the parameterless Dispose method—this prevents the finalizer from running when the GC later catches up with it. Techni- cally, this is unnecessary, as Dispose methods must tolerate repeated calls. However, doing so improves performance because it allows the object (and its referenced objects) to be garbage-collected in a single cycle. Resurrection Suppose a finalizer modifies a living object such that it refers back to the dying object. When the next garbage collection happens (for the object’s generation), the CLR will see the previously dying object as no longer orphaned—and so it will evade garbage collection. This is an advanced scenario, and is called resurrection. To illustrate, suppose we want to write a class that manages a temporary file. When an instance of that class is garbage-collected, we’d like the finalizer to delete the temporary file. It sounds easy: public class TempFileRef { public readonly string FilePath; public TempFileRef (string filePath) { FilePath = filePath; } Finalizers | 495 Disposal and Garbage Collection ~TempFileRef() { File.Delete (FilePath); } } Unfortunately, this has a bug: File.Delete might throw an exception (due to a lack of permissions, perhaps, or the file being in use). Such an exception would take down the whole application (as well as preventing other finalizers from running). We could simply “swallow” the exception with an empty catch block, but then we’d never know that anything went wrong. Calling some elaborate error reporting API would also be undesirable because it would burden the finalizer thread, hindering garbage collection for other objects. We want to restrict finalization actions to those that are simple, reliable, and quick. A better option is to record the failure to a static collection as follows: public class TempFileRef { static ConcurrentQueue _failedDeletions = new ConcurrentQueue(); public readonly string FilePath; public Exception DeletionError { get; private set; } public TempFileRef (string filePath) { FilePath = filePath; } ~TempFileRef() { try { File.Delete (FilePath); } catch (Exception ex) { DeletionError = ex; _failedDeletions.Enqueue (this); // Resurrection } } } Enqueuing the object to the static _failedDeletions collection gives the object another referee, ensuring that it remains alive until the object is eventually dequeued. ConcurrentQueue is a thread-safe version of Queue and is defined in System.Collections.Concurrent (see Chapter 23). There are a couple of reasons for using a thread-safe collection. First, the CLR reserves the right to execute finalizers on more than one thread in parallel. This means that when accessing shared state such as a static collection, we must consider the possibility of two objects being finalized at once. Second, at some point we’re going to want to dequeue items from _failed Deletions so that we can do something about them. This also has to be done in a thread-safe fashion, because it could happen while the finalizer is concurrently enqueuing another object. 496 | Chapter 12: Disposal and Garbage Collection GC.ReRegisterForFinalize A resurrected object’s finalizer will not run a second time—unless you call GC.ReRe gisterForFinalize. In the following example, we try to delete a temporary file in a finalizer (as in the last example). But if the deletion fails, we reregister the object so as to try again in the next garbage collection. public class TempFileRef { public readonly string FilePath; int _deleteAttempt; public TempFileRef (string filePath) { FilePath = filePath; } ~TempFileRef() { try { File.Delete (FilePath); } catch { if (_deleteAttempt++ < 3) GC.ReRegisterForFinalize (this); } } } After the third failed attempt, our finalizer will silently give up trying to delete the file. We could enhance this by combining it with the previous example—in other words, adding it to the _failedDeletions queue after the third failure. Be careful to call ReRegisterForFinalize just once in the finalizer method. If you call it twice, the object will be reregistered twice and will have to undergo two more finalizations! How the Garbage Collector Works The standard CLR uses a generational mark-and-compact GC that performs auto- matic memory management for objects stored on the managed heap. The GC is considered to be a tracing garbage collector in that it doesn’t interfere with every access to an object, but rather wakes up intermittently and traces the graph of objects stored on the managed heap to determine which objects can be considered garbage and therefore collected. The GC initiates a garbage collection upon performing a memory allocation (via the new keyword) either after a certain threshold of memory has been allocated, or at other times to reduce the application’s memory footprint. This process can also be initiated manually by calling System.GC.Collect. During a garbage collection, all threads may by frozen (more on this in the next section). How the Garbage Collector Works | 497 Disposal and Garbage Collection The GC begins with its root object references, and walks the object graph, marking all the objects it touches as reachable. Once this process is complete, all objects that have not been marked are considered unused, and are subject to garbage collection. Unused objects without finalizers are immediately discarded; unused objects with finalizers are enqueued for processing on the finalizer thread after the GC is com- plete. These objects then become eligible for collection in the next GC for the object’s generation (unless resurrected). The remaining “live” objects are then shifted to the start of the heap (compacted), freeing space for more objects. This compaction serves two purposes: it avoids memory fragmentation, and it allows the GC to employ a very simple strategy when allocating new objects, which is to always allocate memory at the end of the heap. This avoids the potentially time-consuming task of maintaining a list of free memory segments. If there is insufficient space to allocate memory for a new object after garbage col- lection, and the operating system is unable to grant further memory, an OutOfMemoryException is thrown. Optimization Techniques The GC incorporates various optimization techniques to reduce the garbage collec- tion time. Generational collection The most important optimization is that the GC is generational. This takes advan- tage of the fact that although many objects are allocated and discarded rapidly, certain objects are long-lived and thus don’t need to be traced during every collection. Basically, the GC divides the managed heap into three generations. Objects that have just been allocated are in Gen0 and objects that have survived one collection cycle are in Gen1; all other objects are in Gen2. The CLR keeps the Gen0 section relatively small (a maximum of 16 MB on the 32- bit workstation CLR, with a typical size of a few hundred KB to a few MB). When the Gen0 section fills up, the GC instigates a Gen0 collection—which happens rel- atively often. The GC applies a similar memory threshold to Gen1 (which acts as a buffer to Gen2), and so Gen1 collections are relatively quick and frequent too. Full collections that include Gen2, however, take much longer and so happen infre- quently. Figure 12-2 shows the effect of a full collection. To give some very rough ballpark figures, a Gen0 collection might take less than 1 ms, which is not enough to be noticed in a typical application. A full collection, however, might take as long as 100 ms on a program with large object graphs. These figures depend on numerous factors and so may vary considerably—particularly in the case of Gen2 whose size is unbounded (unlike Gen0 and Gen1). 498 | Chapter 12: Disposal and Garbage Collection The upshot is that short-lived objects are very efficient in their use of the GC. The StringBuilders created in the following method would almost certainly be collected in a fast Gen0: string Foo() { var sb1 = new StringBuilder ("test"); sb1.Append ("..."); var sb2 = new StringBuilder ("test"); sb2.Append (sb1.ToString()); return sb2.ToString(); } The large object heap The GC uses a separate heap called the Large Object Heap (LOH) for objects larger than a certain threshold (currently 85,000 bytes). This avoids excessive Gen0 col- lections—without the LOH, allocating a series of 16 MB objects might trigger a Gen0 collection after every allocation. The LOH is not subject to compaction, because moving large blocks of memory during garbage collection would be prohibitively expensive. This has two consequences: • Allocations can be slower, because the GC can’t always simply allocate objects at the end of the heap—it must also look in the middle for gaps, and this requires maintaining a linked list of free memory blocks.2 Figure 12-2. Heap generations How the Garbage Collector Works | 499 Disposal and Garbage Collection • The LOH is subject to fragmentation. This means that the freeing of an object can create a hole in the LOH that may be hard to fill later. For instance, a hole left by an 86,000-byte object can be filled only by an object of between 85,000 bytes and 86,000 bytes (unless adjoined by another hole). The large object heap is also nongenerational: all objects are treated as Gen2. Concurrent and background collection The GC must freeze (block) your execution threads for periods during a collection. This includes the entire period during which a Gen0 or Gen1 collection takes place. The GC makes a special attempt, though, at allowing threads to run during a Gen2 collection as it’s undesirable to freeze an application for a potentially long period. This optimization applies to the workstation version of the CLR only, which is used on desktop versions of Windows (and on all versions of Windows with standalone applications). The rationale is that the latency from a blocking collection is less likely to be a problem for server applications that don’t have a user interface. A mitigating factor is that the server CLR leverages all available cores to perform GCs, so an eight-core server will perform a full GC many times faster. In effect, the server GC is tuned to max- imize throughput rather than minimize latency. The workstation optimization has historically been called concurrent collection. From CLR 4.0, it’s been revamped and renamed to background collection. Back- ground collection removes a limitation whereby a concurrent collection would cease to be concurrent if the Gen0 section filled up while a Gen2 collection was running. This means that from CLR 4.0, applications that continually allocate memory will be more responsive. GC notifications (server CLR) From Framework 3.5 SP1, the server version of the CLR can notify you just before a full GC will occur. This is intended for server farm configurations: the idea is that you divert requests to another server just before a collection. You then instigate the collection immediately and wait for it to complete before rerouting requests back to that server. To start notification, call GC.RegisterForFullGCNotification. Then start up another thread (see Chapter 14) that first calls GC.WaitForFullGCApproach. When this method returns a GCNotificationStatus indicating that a collection is near, you can reroute requests to other servers and force a manual collection (see the following section). You then call GC.WaitForFullGCComplete: when this method returns, GC is complete and you can again accept requests. You then repeat the whole cycle. 2. The same thing may occur occasionally in the generational heap due to pinning (see “The fixed Statement” on page 178 in Chapter 4). 500 | Chapter 12: Disposal and Garbage Collection Forcing Garbage Collection You can manually force a GC at any time, by calling GC.Collect. Calling GC.Col lect without an argument instigates a full collection. If you pass in an integer value, only generations to that value are collected, so GC.Collect(0) performs only a fast Gen0 collection. In general, you get the best performance by allowing the GC to decide when to collect: forcing collection can hurt performance by unnecessarily promoting Gen0 objects to Gen1 (and Gen1 objects to Gen2). It can also upset the GC’s self-tuning ability, whereby the GC dynamically tweaks the thresholds for each generation to maximize performance as the application executes. There are exceptions, however. The most common case for intervention is when an application goes to sleep for a while: a good example is a Windows Service that performs a daily activity (checking for updates, perhaps). Such an application might use a System.Timers.Timer to initiate the activity every 24 hours. After completing the activity, no further code executes for 24 hours, which means that for this period, no memory allocations are made and so the GC has no opportunity to activate. Whatever memory the service consumed in performing its activity, it will continue to consume for the following 24 hours—even with an empty object graph! The solution is to call GC.Collect right after the daily activity completes. To ensure the collection of objects for which collection is delayed by finalizers, you can take the additional step of calling WaitForPendingFinalizers and re-collecting: GC.Collect(); GC.WaitForPendingFinalizers(); GC.Collect(); Another case for calling GC.Collect is when you’re testing a class that has a finalizer. Memory Pressure The runtime decides when to initiate collections based on a number of factors, in- cluding the total memory load on the machine. If your program allocates unmanaged memory (Chapter 25), the runtime will get an unrealistically optimistic perception of its memory usage, because the CLR knows only about managed memory. You can mitigate this by telling the CLR to assume a specified quantity of unmanaged memory has been allocated, by calling GC.AddMemoryPressure. To undo this (when the unmanaged memory is released) call GC.RemoveMemoryPressure. Managed Memory Leaks In unmanaged languages such as C++, you must remember to manually deallocate memory when an object is no longer required; otherwise, a memory leak will result. In the managed world, this kind of error is impossible due to the CLR’s automatic garbage collection system. Nonetheless, large and complex .NET applications can exhibit a milder form of the same syndrome with the same end result: the application consumes more and more Managed Memory Leaks | 501 Disposal and Garbage Collection memory over its lifetime, until it eventually has to be restarted. The good news is that managed memory leaks are usually easier to diagnose and prevent. Managed memory leaks are caused by unused objects remaining alive by virtue of unused or forgotten references. A common candidate is event handlers—these hold a reference to the target object (unless the target is a static method). For instance, consider the following classes: class Host { public event EventHandler Click; } class Client { Host _host; public Client (Host host) { _host = host; _host.Click += HostClicked; } void HostClicked (object sender, EventArgs e) { ... } } The following test class contains a method that instantiates 1,000 clients: class Test { static Host _host = new Host(); public static void CreateClients() { Client[] clients = Enumerable.Range (0, 1000) .Select (i => new Client (_host)) .ToArray(); // Do something with clients ... } } You might expect that after CreateClients finishes executing, the 1,000 Client objects will become eligible for collection. Unfortunately, each client has another referee: the _host object whose Click event now references each Client instance. This may go unnoticed if the Click event doesn’t fire—or if the HostClicked method doesn’t do anything to attract attention. One way to solve this is to make Client implement IDisposable, and in the Dis pose method, unhook the event handler: public void Dispose() { _host.Click -= HostClicked; } Consumers of Client then dispose of the instances when they’re done with them: Array.ForEach (clients, c => c.Dispose()); 502 | Chapter 12: Disposal and Garbage Collection In “Weak References,” we’ll describe another solution to this problem, which can be useful in environments which tend not to use disposable objects (an example is WPF). In fact, the WPF framework offers a class called WeakEventManager that leverages a pattern employing weak references. On the topic of WPF, data binding is another common cause for memory leaks: the issue is described at http://support.microsoft .com/kb/938416. Timers Forgotten timers can also cause memory leaks (we discuss timers in Chapter 22). There are two distinct scenarios, depending on the kind of timer. Let’s first look at the timer in the System.Timers namespace. In the following example, the Foo class (when instantiated) calls the tmr_Elapsed method once every second: using System.Timers; class Foo { Timer _timer; Foo() { _timer = new System.Timers.Timer { Interval = 1000 }; _timer.Elapsed += tmr_Elapsed; _timer.Start(); } void tmr_Elapsed (object sender, ElapsedEventArgs e) { ... } } Unfortunately, instances of Foo can never be garbage-collected! The problem is the .NET Framework itself holds references to active timers so that it can fire their Elapsed events. Hence: • The .NET Framework will keep _timer alive. • _timer will keep the Foo instance alive, via the tmr_Elapsed event handler. The solution is obvious when you realize that Timer implements IDisposable. Disposing of the timer stops it and ensures that the .NET Framework no longer references the object: class Foo : IDisposable { ... public void Dispose() { _timer.Dispose(); } } Managed Memory Leaks | 503 Disposal and Garbage Collection A good guideline is to implement IDisposable yourself if any field in your class is assigned an object that implements IDis posable. The WPF and Windows Forms timers behave in exactly the same way, with respect to what’s just been discussed. The timer in the System.Threading namespace, however, is special. The .NET Frame- work doesn’t hold references to active threading timers; it instead references the callback delegates directly. This means that if you forget to dispose of a threading timer, a finalizer can fire which will automatically stop and dispose the timer. For example: static void Main() { var tmr = new System.Threading.Timer (TimerTick, null, 1000, 1000); GC.Collect(); System.Threading.Thread.Sleep (10000); // Wait 10 seconds } static void TimerTick (object notUsed) { Console.WriteLine ("tick"); } If this example is compiled in “release” mode (debugging disabled and optimizations enabled), the timer will be collected and finalized before it has a chance to fire even once! Again, we can fix this by disposing of the timer when we’re done with it: using (var tmr = new System.Threading.Timer (TimerTick, null, 1000, 1000)) { GC.Collect(); System.Threading.Thread.Sleep (10000); // Wait 10 seconds } The implicit call to tmr.Dispose at the end of the using block ensures that the tmr variable is “used” and so not considered dead by the GC until the end of the block. Ironically, this call to Dispose actually keeps the object alive longer! Diagnosing Memory Leaks The easiest way to avoid managed memory leaks is to proactively monitor memory consumption as an application is written. You can obtain the current memory con- sumption of a program’s objects as follows (the true argument tells the GC to per- form a collection first): long memoryUsed = GC.GetTotalMemory (true); If you’re practicing test-driven development, one possibility is to use unit tests to assert that memory is reclaimed as expected. If such an assertion fails, you then have to examine only the changes that you’ve made recently. If you already have a large application with a managed memory leak, the windbg.exe tool can assist in finding it. There are also friendlier graphical tools such 504 | Chapter 12: Disposal and Garbage Collection as Microsoft’s CLR Profiler, SciTech’s Memory Profiler, and Red Gate’s ANTS Memory Profiler. The CLR also exposes numerous Windows WMI counters to assist with resource monitoring. Weak References Occasionally, it’s useful to hold a reference to an object that’s “invisible” to the GC in terms of keeping the object alive. This is called a weak reference, and is imple- mented by the System.WeakReference class. To use WeakReference, construct it with a target object as follows: var sb = new StringBuilder ("this is a test"); var weak = new WeakReference (sb); Console.WriteLine (weak.Target); // This is a test If a target is referenced only by one or more weak references, the GC will consider the target eligible for collection. When the target gets collected, the Target property of the WeakReference will be null: var weak = new WeakReference (new StringBuilder ("weak")); Console.WriteLine (weak.Target); // weak GC.Collect(); Console.WriteLine (weak.Target); // (nothing) To avoid the target being collected in between testing for it being null and consuming it, assign the target to a local variable: var weak = new WeakReference (new StringBuilder ("weak")); var sb = (StringBuilder) weak.Target; if (sb != null) { /* Do something with sb */ } Once a target’s been assigned to a local variable, it has a strong root and so cannot be collected while that variable’s in use. The following class uses weak references to keep track of all Widget objects that have been instantiated, without preventing those objects from being collected: class Widget { static List _allWidgets = new List(); public readonly string Name; public Widget (string name) { Name = name; _allWidgets.Add (new WeakReference (this)); } public static void ListAllWidgets() { foreach (WeakReference weak in _allWidgets) { Weak References | 505 Disposal and Garbage Collection Widget w = (Widget)weak.Target; if (w != null) Console.WriteLine (w.Name); } } } The only proviso with such a system is that the static list will grow over time, accu- mulating weak references with null targets. So you need to implement some cleanup strategy. Weak References and Caching One use for WeakReference is to cache large object graphs. This allows memory- intensive data to be cached briefly without causing excessive memory consumption: _weakCache = new WeakReference (...); // _weakCache is a field ... var cache = _weakCache.Target; if (cache == null) { /* Re-create cache & assign it to _weakCache */ } This strategy may be only mildly effective in practice, because you have little control over when the GC fires and what generation it chooses to collect. In particular, if your cache remains in Gen0, it may be collected within microseconds (and remember that the GC doesn’t collect only when memory is low—it collects regularly under normal memory conditions). So at a minimum, you should employ a two-level cache whereby you start out by holding strong references that you convert to weak refer- ences over time. Weak References and Events We saw earlier how events can cause managed memory leaks. The simplest solution is to either avoid subscribing in such conditions, or implement a Dispose method to unsubscribe. Weak references offer another solution. Imagine a delegate that holds only weak references to its targets. Such a delegate would not keep its targets alive—unless those targets had independent referees. Of course, this wouldn’t prevent a firing delegate from hitting an unreferenced target —in the time between the target being eligible for collection and the GC catching up with it. For such a solution to be effective, your code must be robust in that scenario. Assuming that is the case, a weak delegate class can be implemented as follows: public class WeakDelegate where TDelegate : class { class MethodTarget { public readonly WeakReference Reference; public readonly MethodInfo Method; public MethodTarget (Delegate d) { Reference = new WeakReference (d.Target); Method = d.Method; 506 | Chapter 12: Disposal and Garbage Collection } } List _targets = new List(); public WeakDelegate() { if (!typeof (TDelegate).IsSubclassOf (typeof (Delegate))) throw new InvalidOperationException ("TDelegate must be a delegate type"); } public void Combine (TDelegate target) { if (target == null) return; foreach (Delegate d in (target as Delegate).GetInvocationList()) _targets.Add (new MethodTarget (d)); } public void Remove (TDelegate target) { if (target == null) return; foreach (Delegate d in (target as Delegate).GetInvocationList()) { MethodTarget mt = _targets.Find (w => d.Target.Equals (w.Reference.Target) && d.Method.MethodHandle.Equals (w.Method.MethodHandle)); if (mt != null) _targets.Remove (mt); } } public TDelegate Target { get { var deadRefs = new List(); Delegate combinedTarget = null; foreach (MethodTarget mt in _targets.ToArray()) { WeakReference target = mt.Reference; if (target != null && target.IsAlive) { var newDelegate = Delegate.CreateDelegate ( typeof (TDelegate), mt.Reference.Target, mt.Method); combinedTarget = Delegate.Combine (combinedTarget, newDelegate); } else deadRefs.Add (mt); } foreach (MethodTarget mt in deadRefs) // Remove dead references _targets.Remove (mt); // from _targets. Weak References | 507 Disposal and Garbage Collection return combinedTarget as TDelegate; } set { _targets.Clear(); Combine (value); } } } This code illustrates a number of interesting points in C# and the CLR. First, note that we check that TDelegate is a delegate type in the constructor. This is because of a limitation in C#—the following type constraint is illegal because C# considers System.Delegate a special type for which constraints are not supported: ... where TDelegate : Delegate // Compiler doesn't allow this Instead, we must choose a class constraint, and perform a runtime check in the constructor. In the Combine and Remove methods, we perform the reference conversion from tar get to Delegate via the as operator rather than the more usual cast operator. This is because C# disallows the cast operator with this type parameter—because of a potential ambiguity between a custom conversion and a reference conversion. We then call GetInvocationList because these methods might be called with mul- ticast delegates—delegates with more than one method recipient. In the Target property, we build up a multicast delegate that combines all the dele- gates referenced by weak references whose targets are alive. We then clear out the remaining (dead) references from the list—to avoid the _targets list endlessly grow- ing. (We could improve our class by doing the same in the Combine method; yet another improvement would be to add locks for thread safety [Chapter 22]). The following illustrates how to consume this delegate in implementing an event: public class Foo { WeakDelegate _click = new WeakDelegate(); public event EventHandler Click { add { _click.Combine (value); } remove { _click.Remove (value); } } protected virtual void OnClick (EventArgs e) { EventHandler target = _click.Target; if (target != null) target (this, e); } } Notice that in firing the event, we assign _click.Target to a temporary variable before checking and invoking it. This avoids the possibility of targets being collected in the interim. 508 | Chapter 12: Disposal and Garbage Collection 13Diagnostics and Code Contracts When things go wrong, it’s important that information is available to aid in diag- nosing the problem. An IDE or debugger can assist greatly to this effect—but it is usually available only during development. Once an application ships, the applica- tion itself must gather and record diagnostic information. To meet this requirement, the .NET Framework provides a set of facilities to log diagnostic information, mon- itor application behavior, detect runtime errors, and integrate with debugging tools if available. The .NET Framework also allows you to enforce code contracts. Introduced in Framework 4.0, code contracts allow methods to interact through a set of mutual obligations, and fail early if those obligations are violated. The types in this chapter are defined primarily in the System.Diagnostics and Sys tem.Diagnostics.Contracts namespaces. Conditional Compilation You can conditionally compile any section of code in C# with preprocessor direc- tives. Preprocessor directives are special instructions to the compiler that begin with the # symbol (and, unlike other C# constructs, must appear on a line of their own). Logically, they execute before the main compilation takes place (although in prac- tice, the compiler processes them during the lexical parsing phase). The preprocessor directives for conditional compilation are #if, #else, #endif, and #elif. The #if directive instructs the compiler to ignore a section of code unless a specified symbol has been defined. You can define a symbol with either the #define directive or a compilation switch. #define applies to a particular file; a compilation switch applies to a whole assembly: #define TESTMODE // #define directives must be at top of file // Symbol names are uppercase by convention. using System; class Program 509 { static void Main() { #if TESTMODE Console.WriteLine ("in test mode!"); // OUTPUT: in test mode! #endif } } If we deleted the first line, the program would compile with the Console.WriteLine statement completely eliminated from the executable, as though it was commented out. The #else statement is analogous to C#’s else statement, and #elif is equivalent to #else followed by #if. The ||, &&, and ! operators can be used to perform or, and, and not operations: #if TESTMODE && !PLAYMODE // if TESTMODE and not PLAYMODE ... Bear in mind, however, that you’re not building an ordinary C# expression, and the symbols upon which you operate have absolutely no connection to variables—static or otherwise. To define a symbol assembly-wide, specify the /define switch when compiling: csc Program.cs /define:TESTMODE,PLAYMODE Visual Studio provides an option to enter conditional compilation symbols under Project Properties. If you’ve defined a symbol at the assembly level and then want to “undefine” it for a particular file, you can do so with the #undef directive. Conditional Compilation Versus Static Variable Flags The preceding example could instead be implemented with a simple static field: static internal bool TestMode = true; static void Main() { if (TestMode) Console.WriteLine ("in test mode!"); } This has the advantage of allowing runtime configuration. So, why choose condi- tional compilation? The reason is that conditional compilation can take you places variable flags cannot, such as: • Conditionally including an attribute • Changing the declared type of variable • Switching between different namespaces or type aliases in a using directive— for example: using TestType = #if V2 510 | Chapter 13: Diagnostics and Code Contracts MyCompany.Widgets.GadgetV2; #else MyCompany.Widgets.Gadget; #endif You can even perform major refactoring under a conditional compilation directive, so you can instantly switch between old and new versions, and write libraries that can compile against multiple Framework versions, leveraging the latest Framework features where available. Another advantage of conditional compilation is that debugging code can refer to types in assemblies that are not included in deployment. The Conditional Attribute The Conditional attribute instructs the compiler to ignore any calls to a particular class or method, if the specified symbol has not been defined. To see how this is useful, suppose you write a method for logging status information as follows: static void LogStatus (string msg) { string logFilePath = ... System.IO.File.AppendAllText (logFilePath, msg + "\r\n"); } Now imagine you wanted this to execute only if the LOGGINGMODE symbol is defined. The first solution is to wrap all calls to LogStatus around an #if directive: #if LOGGINGMODE LogStatus ("Message Headers: " + GetMsgHeaders()); #endif This gives an ideal result, but it is tedious. The second solution is to put the #if directive inside the LogStatus method. This, however, is problematic should LogSta tus be called as follows: LogStatus ("Message Headers: " + GetComplexMessageHeaders()); GetComplexMessageHeaders would always get called—which might incur a perfor- mance hit. We can combine the functionality of the first solution with the convenience of the second by attaching the Conditional attribute (defined in System.Diagnostics) to the LogStatus method: [Conditional ("LOGGINGMODE")] static void LogStatus (string msg) { ... } This instructs the compiler to treat calls to LogStatus as though they were wrapped in an #if LOGGINGMODE directive. If the symbol is not defined, any calls to LogSta tus get eliminated entirely in compilation—including their argument evaluation Conditional Compilation | 511 Diagnostics and Code Contracts expressions. (Hence any side-effecting expressions will be bypassed.) This works even if LogStatus and the caller are in different assemblies. Another benefit of [Conditional] is that the conditionality check is performed when the caller is compiled, rather than when the called method is compiled. This is beneficial because it allows you to write a library containing methods such as Log Status—and build just one version of that library. The Conditional attribute is ignored at runtime—it’s purely an instruction to the compiler. Alternatives to the Conditional attribute The Conditional attribute is useless if you need to dynamically enable or disable functionality at runtime: instead, you must use a variable-based approach. This leaves the question of how to elegantly circumvent the evaluation of arguments when calling conditional logging methods. A functional approach solves this: using System; using System.Linq; class Program { public static bool EnableLogging; static void LogStatus (Func message) { string logFilePath = ... if (EnableLogging) System.IO.File.AppendAllText (logFilePath, message() + "\r\n"); } } A lambda expression lets you call this method without syntax bloat: LogStatus ( () => "Message Headers: " + GetComplexMessageHeaders() ); If EnableLogging is false, GetComplexMessageHeaders is never evaluated. Debug and Trace Classes Debug and Trace are static classes that provide basic logging and assertion capabili- ties. The two classes are very similar; the main differentiator is their intended use. The Debug class is intended for debug builds; the Trace class is intended for both debug and release builds. To this effect: All methods of the Debug class are defined with [Conditional("DEBUG")]. All methods of the Trace class are defined with [Conditional("TRACE")]. This means that all calls that you make to Debug or Trace are eliminated by the compiler unless you define DEBUG or TRACE symbols. By default, Visual Studio defines 512 | Chapter 13: Diagnostics and Code Contracts both DEBUG and TRACE symbols in a project’s debug configuration—and just the TRACE symbol in the release configuration. Both the Debug and Trace classes provide Write, WriteLine, and WriteIf methods. By default, these send messages to the debugger’s output window: Debug.Write ("Data"); Debug.WriteLine (23 * 34); int x = 5, y = 3; Debug.WriteIf (x > y, "x is greater than y"); The Trace class also provides the methods TraceInformation, TraceWarning, and TraceError. The difference in behavior between these and the Write methods depends on the active TraceListeners (we’ll cover this in “TraceListener”). Fail and Assert The Debug and Trace classes both provide Fail and Assert methods. Fail sends the message to each TraceListener in the Debug or Trace class’s Listeners collection (see the following section), which by default writes the message to the debug output as well as displaying it in a dialog: Debug.Fail ("File data.txt does not exist!"); The dialog that appears asks you whether to ignore, abort, or retry. The latter then lets you attach a debugger, which is useful in instantly diagnosing the problem. Assert simply calls Fail if the bool argument is false—this is called making an assertion and indicates a bug in the code if violated. Specifying a failure message is optional: Debug.Assert (File.Exists ("data.txt"), "File data.txt does not exist!"); var result = ... Debug.Assert (result != null); The Write, Fail, and Assert methods are also overloaded to accept a string category in addition to the message, which can be useful in processing the output. An alternative to assertion is to throw an exception if the opposite condition is true. This is a common practice when validating method arguments: public void ShowMessage (string message) { if (message == null) throw new ArgumentNullException ("message"); ... } Such “assertions” are compiled unconditionally and are less flexible in that you can’t control the outcome of a failed assertion via TraceListeners. And technically, they’re not assertions. An assertion is something that, if violated, indicates a bug in the current method’s code. Throwing an exception based on argument validation indi- cates a bug in the caller’s code. Debug and Trace Classes | 513 Diagnostics and Code Contracts We’ll see soon how code contracts extend the principles of Fail and Assert, providing more power and flexibility. TraceListener The Debug and Trace classes each have a Listeners property, comprising a static collection of TraceListener instances. These are responsible for processing the con- tent emitted by the Write, Fail, and Trace methods. By default, the Listeners collection of each includes a single listener (DefaultTrace Listener). The default listener has two key features: • When connected to a debugger such as Visual Studio, messages are written to the debug output window; otherwise, message content is ignored. • When the Fail method is called (or an assertion fails), a dialog appears asking the user whether to continue, abort, or retry (attach/debug)—regardless of whether a debugger is attached. You can change this behavior by (optionally) removing the default listener, and then adding one or more of your own. You can write trace listeners from scratch (by subclassing TraceListener) or use one of the predefined types: • TextWriterTraceListener writes to a Stream or TextWriter or appends to a file. • EventLogTraceListener writes to the Windows event log. • EventProviderTraceListener writes to the Event Tracing for Windows (ETW) subsystem in Windows Vista and later. • WebPageTraceListener writes to an ASP.NET web page. TextWriterTraceListener is further subclassed to ConsoleTraceListener, Delimited ListTraceListener, XmlWriterTraceListener, and EventSchemaTraceListener. None of these listeners display a dialog when Fail is called— only DefaultTraceListener has this behavior. The following example clears Trace’s default listener, then adds three listeners—one that appends to a file, one that writes to the console, and one that writes to the Windows event log: // Clear the default listener: Trace.Listeners.Clear(); // Add a writer that appends to the trace.txt file: Trace.Listeners.Add (new TextWriterTraceListener ("trace.txt")); // Obtain the Console's output stream, then add that as a listener: 514 | Chapter 13: Diagnostics and Code Contracts System.IO.TextWriter tw = Console.Out; Trace.Listeners.Add (new TextWriterTraceListener (tw)); // Set up a Windows Event log source and then create/add listener. // CreateEventSource requires administrative elevation, so this would // typically be done in application setup. if (!EventLog.SourceExists ("DemoApp")) EventLog.CreateEventSource ("DemoApp", "Application"); Trace.Listeners.Add (new EventLogTraceListener ("DemoApp")); (It’s also possible to add listeners via the application configuration file; this is handy in allowing testers to configure tracing after an application has been built—go to http://albahari.com/traceconfig for the MSDN article.) In the case of the Windows event log, messages that you write with the Write, Fail, or Assert method always display as “Information” messages in the Windows event viewer. Messages that you write via the TraceWarning and TraceError methods, however, show up as warnings or errors. TraceListener also has a Filter of type TraceFilter that you can set to control whether a message gets written to that listener. To do this, you either instantiate one of the predefined subclasses (EventTypeFilter or SourceFilter), or subclass Trace Filter and override the ShouldTrace method. You could use this to filter by category, for instance. TraceListener also defines IndentLevel and IndentSize properties for controlling indentation, and the TraceOutputOptions property for writing extra data: TextWriterTraceListener tl = new TextWriterTraceListener (Console.Out); tl.TraceOutputOptions = TraceOptions.DateTime | TraceOptions.Callstack; TraceOutputOptions are applied when using the Trace methods: Trace.TraceWarning ("Orange alert"); DiagTest.vshost.exe Warning: 0 : Orange alert DateTime=2007-03-08T05:57:13.6250000Z Callstack= at System.Environment.GetStackTrace(Exception e, Boolean needFileInfo) at System.Environment.get_StackTrace() at ... Flushing and Closing Listeners Some listeners, such as TextWriterTraceListener, ultimately write to a stream that is subject to caching. This has two implications: • A message may not appear in the output stream or file immediately. • You must close—or at least flush—the listener before your application ends; otherwise, you lose what’s in the cache (up to 4 KB, by default, if you’re writing to a file). The Trace and Debug classes provide static Close and Flush methods that call Close or Flush on all listeners (which in turn calls Close or Flush on any underlying writers Debug and Trace Classes | 515 Diagnostics and Code Contracts and streams). Close implicitly calls Flush, closes file handles, and prevents further data from being written. As a general rule, call Close before an application ends and call Flush anytime you want to ensure that current message data is written. This applies if you’re using stream- or file-based listeners. Trace and Debug also provide an AutoFlush property, which, if true, forces a Flush after every message. It’s a good policy to set AutoFlush to true on Debug and Trace if you’re using any file- or stream-based listeners. Otherwise, if an unhandled exception or critical error occurs, the last 4 KB of diagnostic information may be lost. Code Contracts Overview We mentioned previously the concept of an assertion, whereby you check that cer- tain conditions are met throughout your program. If a condition fails, it indicates a bug, which is typically handled by invoking a debugger (in debug builds) or throwing an exception (in release builds). Assertions follow the principle that if something goes wrong, it’s best to fail early and close to the source of the error. This is usually better than trying to continue with invalid data—which can result in incorrect results, undesired side-effects, or an exception later on in the program (all of which are harder to diagnose). Historically, there have been two ways to enforce assertions: • By calling the Assert method on Debug or Trace • By throwing exceptions (such as ArgumentNullException) Framework 4.0 introduced a new feature called code contracts, which replaces both of these approaches with a unified system. That system allows you to make not only simple assertions but also more powerful contract-based assertions. Code contracts derive from the principle of “Design by Contract” from the Eiffel programming language, where functions interact with each other through a system of mutual obligations and benefits. Essentially, a function specifies preconditions that must be met by the client (caller), and in return guarantees postconditions which the client can depend on when the function returns. The types for code contracts live in the System.Diagnostics.Contracts namespace. Although the types that support code contracts are built into the .NET Framework, the binary rewriter and the static check- ing tools are available as a separate download at the Microsoft DevLabs site (http://msdn.microsoft.com/devlabs). You must install these tools before you can use code contracts in Visual Studio. 516 | Chapter 13: Diagnostics and Code Contracts Why Use Code Contracts? To illustrate, we’ll write a method that adds an item to a list only if it’s not already present—with two preconditions and a postcondition: public static bool AddIfNotPresent (IList list, T item) { Contract.Requires (list != null); // Precondition Contract.Requires (!list.IsReadOnly); // Precondition Contract.Ensures (list.Contains (item)); // Postcondition if (list.Contains(item)) return false; list.Add (item); return true; } The preconditions are defined by Contract.Requires and are verified when the method starts. The postcondition is defined by Contract.Ensures and is verified not where it appears in the code, but when the method exits. Preconditions and postconditions act like assertions and, in this case, detect the following errors: • Calling the method with a null or read-only list • A bug in the method whereby we forgot to add the item to the list Preconditions and postconditions must appear at the start of the method. This is conducive to good design: if you fail to fulfill the contract in subsequently writing the method, the error will be detected. Moreover, these conditions form a discoverable contract for that method. AddIfNot Present advertises to consumers: • “You must call me with a non-null writable list.” • “When I return, that list will contain the item you specified.” These facts can be emitted into the assembly’s XML documentation file (you can do this in Visual Studio by going to the Code Contracts tab of the Project Properties window, enabling the building of a contracts reference assembly, and checking “Emit Contracts into XML doc file”). Tools such as SandCastle can then incorporate contract details into documentation files. Contracts also enable your program to be analyzed for correctness by static contract validation tools. If you try to call AddIfNotPresent with a list whose value might be null, for example, a static validation tool could warn you before you even run the program. Another benefit of contracts is ease of use. In our example, it’s easier to code the postcondition upfront than at both exit points. Contracts also support object invar- iants—which further reduce repetitive coding and make for more reliable enforcement. Code Contracts Overview | 517 Diagnostics and Code Contracts Conditions can also be placed on interface members and abstract methods, some- thing that is impossible with standard validation approaches. And conditions on virtual methods cannot be accidentally circumvented by subclasses. Yet another benefit of code contracts is that contract violation behavior can be cus- tomized easily and in more ways than if you rely on calling Debug.Assert or throwing exceptions. And it’s possible to ensure that contract violations are always recorded —even if contract violation exceptions are swallowed by exception handlers higher in the call stack. The disadvantage of using code contracts is that the .NET implementation relies on a binary rewriter—a tool that mutates the assembly after compilation. This slows the build process, as well as complicating services that rely on calling the C# com- piler (whether explicitly or via the CSharpCodeProvider class). The enforcing of code contracts may also incur a runtime performance hit, although this is easily mitigated by scaling back contract checking in release builds. Another limitation of code contracts is that you can’t use them to enforce security-sensitive checks, because they can be cir- cumvented at runtime (by handling the ContractFailed event). Contract Principles Code contracts comprise preconditions, postconditions, assertions, and object invar- iants. These are all discoverable assertions. They differ based on when they are verified: • Preconditions are verified when a function starts. • Postconditions are verified before a function exits. • Assertions are verified wherever they appear in the code. • Object invariants are verified after every public function in a class. Code contracts are defined entirely by calling (static) methods in the Contract class. This makes contracts language-independent. Contracts can appear not only in methods, but in other functions as well, such as constructors, properties, indexers, and operators. Compilation Almost all methods in the Contract class are defined with the [Conditional("CON TRACTS_FULL")] attribute. This means that unless you define the CONTRACTS_FULL symbol, (most) contract code is stripped out. Visual Studio defines the CON TRACTS_FULL symbol automatically if you enable contract checking in the Code Con- tracts tab of the Project Properties page. (For this tab to appear, you must download and install the Contracts tools from the Microsoft DevLabs site.) 518 | Chapter 13: Diagnostics and Code Contracts Removing the CONTRACTS_FULL symbol might seem like an easy way to disable all contract checking. However, it doesn’t apply to Requires conditions (which we’ll describe in detail soon). The only way to disable contracts in code that uses Requires is to enable the CONTRACTS_FULL symbol and then get the binary rewriter to strip out contract code by choosing an enforcement level of “none.” The binary rewriter After compiling code that contains contracts, you must call the binary rewriter tool, ccrewrite.exe (Visual Studio does this automatically if contract checking is enabled). The binary rewriter moves postconditions (and object invariants) into the right place, calls any conditions and object invariants in overridden methods, and replaces calls to Contract with calls to a contracts runtime class. Here’s a (simplified) version of what our earlier example would look like after rewriting: static bool AddIfNotPresent (IList list, T item) { __ContractsRuntime.Requires (list != null); __ContractsRuntime.Requires (!list.IsReadOnly); bool result; if (list.Contains (item)) result = false; else { list.Add (item); result = true; } __ContractsRuntime.Ensures (list.Contains (item)); // Postcondition return result; } If you fail to call the binary rewriter, Contract won’t get replaced with __ContractsRuntime and the former will end up throwing exceptions. The __ContractsRuntime type is the default contracts runtime class. In advanced scenarios, you can specify your own contracts runtime class via the /rw switch or Visual Studio’s Code Con- tracts tab in Project Properties. Because __ContractsRuntime is shipped with the binary rewriter (which is not a standard part of the .NET Framework), the bi- nary rewriter actually injects the __ContractsRuntime class into your compiled assembly. You can examine its code by disas- sembling any assembly that enables code contracts. The binary rewriter also offers switches to strip away some or all contract checking: we describe these in “Selectively Enforcing Contracts” on page 531. You typically Code Contracts Overview | 519 Diagnostics and Code Contracts enable full contract checking in debug build configurations and a subset of contract checking in release configurations. Asserting versus throwing on failure The binary rewriter also lets you choose between displaying a dialog and throwing a ContractException upon contract failure. The former is typically used for debug builds; the latter for release builds. To enable the latter, specify /throwonfailure when calling the binary rewriter, or uncheck the “Assert on contract failure” check- box in Visual Studio’s Code Contracts tab in Project Properties. We’ll revisit this topic in more detail in “Dealing with Contract Fail- ure” on page 529. Purity All functions that you call from arguments passed to contract methods (Requires, Assumes, Assert, etc.) must be pure—that is, side-effect-free (they must not alter the values of fields). You must signal to the binary rewriter that any functions you call are pure by applying the [Pure] attribute: [Pure] public static bool IsValidUri (string uri) { ... } This makes the following legal: Contract.Requires (IsValidUri (uri)); The contract tools implicitly assume that all property get accessors are pure, as are all C# operators (+, *, %, etc.) and members on selected Framework types, including string, Contract, Type, System.IO.Path, and LINQ’s query operators. It also assumes that methods invoked via delegates marked with the [Pure] attribute are pure (the Comparison and Predicate attributes are marked with this attribute). Preconditions You can define code contract preconditions by calling Contract.Requires, Con tract.Requires or Contract.EndContractBlock. Contract.Requires Calling Contract.Requires at the start of a function enforces a precondition: static string ToProperCase (string s) { Contract.Requires (!string.IsNullOrEmpty(s)); ... } This is like making an assertion, except that the precondition forms a discoverable fact about your function that can be extracted from the compiled code and con- sumed by documentation or static checking tools (so that they can warn you should 520 | Chapter 13: Diagnostics and Code Contracts they see some code elsewhere in your program that tries to call ToProperCase with a null or empty string). A further benefit of preconditions is that subclasses that override virtual methods with preconditions cannot prevent the base class method’s preconditions from being checked. And preconditions defined on interface members will be implicitly woven into the concrete implementations (see “Contracts on Interfaces and Abstract Meth- ods” on page 528). Preconditions should access only members that are at least as accessible as the function itself—this ensures that callers can make sense of the contract. If you need to read or call less ac- cessible members, it’s likely that you’re validating internal state rather than enforcing the calling contract, in which case you should make an assertion instead. You can call Contract.Requires as many times as necessary at the start of the method to enforce different conditions. What Should You Put in Preconditions? The guideline from the Code Contracts team is that preconditions should: • Be possible for the client (caller) to easily validate. • Rely only on data & functions at least as accessible as the method itself. • Always indicate a bug if violated. A consequence of the last point is that a client should never specifically “catch” a contract failure (the ContractException type, in fact, is internal to help enforce that principle). Instead, the client should call the target properly; if it fails, this indicates a bug that should be handled via your general exception backstop (which may include terminating the application). In other words, if you decide control-flow or do other things based on a precondition failure, it’s not really a contract because you can continue executing if it fails. This leads to the following advice, when choosing between preconditions and throwing ordinary exceptions: • If failure always indicates a bug in the client, favor a precondition. • If failure indicates an abnormal condition, which may mean a bug in the client, throw a (catchable) exception instead. To illustrate, suppose we’re writing the Int32.Parse function. It’s reasonable to assume that a null input string always indicates a bug in the caller, so we’d enforce this with a precondition: public static int Parse (string s) { Contract.Requires (s != null); } Preconditions | 521 Diagnostics and Code Contracts Next, we need to check that the string contains only digits and symbols such as + and – (in the right place). It would place an unreasonable burden on the caller to validate this and so we’d enforce it not as a precondition, but a manual check that throws a (catchable) FormatException if violated. To illustrate the member accessibility issue, consider the following code, which often appears in types implementing the IDisposable interface: public void Foo() { if (_isDisposed) // _isDisposed is a private field throw new ObjectDisposedException(); ... } This check should not be made into a precondition unless we make _isDisposed accessible to the caller (by refactoring it into a publicly readable property, for instance). Finally, consider the File.ReadAllText method. The following would be inappro- priate use of a precondition: public static string ReadAllText (string path) { Contract.Requires (File.Exists (path)); ... } The caller cannot reliably know that the file exists before calling this method (it could be deleted between making that check and calling the method). So, we’d enforce this in the old-fashioned way—by throwing a catchable FileNotFoundEx ception instead. Contract.Requires The introduction of code contracts challenges the following deeply entrenched pat- tern established in the .NET Framework from version 1.0: static void SetProgress (string message, int percent) // Classic approach { if (message == null) throw new ArgumentNullException ("message"); if (percent < 0 || percent > 100) throw new ArgumentOutOfRangeException ("percent"); ... } static void SetProgress (string message, int percent) // Modern approach { Contract.Requires (message != null); Contract.Requires (percent >= 0 && percent <= 100); ... } 522 | Chapter 13: Diagnostics and Code Contracts If you have a large assembly that enforces classic argument checking, writing new methods with preconditions will create an inconsistent library: some methods will throw argument exceptions whereas others will throw a ContractException. One solution is to update all existing methods to use contracts, but this has two problems: • It’s time-consuming. • Callers may have come to depend on an exception type such as ArgumentNullEx ception being thrown. (This almost certainly indicates bad design, but may be the reality nonetheless.) The solution is to call the generic version of Contract.Requires. This lets you specify an exception type to throw upon failure: Contract.Requires (message != null, "message"); Contract.Requires (percent >= 0 && percent <= 100, "percent"); (The second argument gets passed to the constructor of the exception class). This results in the same behavior as with old-fashioned argument checking, while delivering the benefits of contracts (conciseness, support for interfaces, implicit documentation, static checking and runtime customization). The specified exception is thrown only if you specify /throwon failure when rewriting the assembly (or uncheck the Assert on Contract Failure checkbox in Visual Studio). Otherwise, a dia- log box appears. It’s also possible to specify a contract-checking level of ReleaseRequires in the binary rewriter (see “Selectively Enforcing Contracts” on page 531). Calls to the generic Contract.Requires then remain in place while all other checks are stripped away: this results in an assembly that behaves just as in the past. Contract.EndContractBlock The Contract.EndContractBlock method lets you get the benefit of code contracts with traditional argument-checking code—avoiding the need to refactor code writ- ten prior to Framework 4.0. All you do is call this method after performing manual argument checks: static void Foo (string name) { if (name == null) throw new ArgumentNullException ("name"); Contract.EndContractBlock(); ... } The binary rewriter then converts this code into something equivalent to: static void Foo (string name) { Contract.Requires (name != null, "name"); Preconditions | 523 Diagnostics and Code Contracts ... } The code that precedes EndContractBlock must comprise simple statements of the form: if throw ; You can mix traditional argument checking with code contract calls: simply put the latter after the former: static void Foo (string name) { if (name == null) throw new ArgumentNullException ("name"); Contract.Requires (name.Length >= 2); ... } Calling any of the contract-enforcing methods implicitly ends the contract block. The point is to define a region at the beginning of the method where the contract rewriter knows that every if statement is part of a contract. Calling any of the con- tract-enforcing methods implicitly extends the contract block, so you don’t need to use EndContractBlock if you use another method such as Contract.Ensures, etc. Preconditions and Overridden Methods When overriding a virtual method, you cannot add preconditions, because doing so would change the contract (by making it more restrictive)—breaking the principles of polymorphism. (Technically, the designers could have allowed overridden methods to weaken pre- conditions; they decided against this because the scenarios weren’t sufficiently com- pelling to justify adding this complexity). The binary rewriter ensures that a base method’s preconditions are always enforced in subclasses—whether or not the overrid- den method calls the base method. Postconditions Contract.Ensures Contract.Ensures enforces a postcondition: something which must be true when the method exits. We saw an example earlier: static bool AddIfNotPresent (IList list, T item) { Contract.Requires (list != null); // Precondition Contract.Ensures (list.Contains (item)); // Postcondition if (list.Contains(item)) return false; list.Add (item); 524 | Chapter 13: Diagnostics and Code Contracts return true; } The binary rewriter moves postconditions to the exit points of the method. Post- conditions are checked if you return early from a method (as in this example)—but not if you return early via an unhandled exception. Unlike preconditions, which detect misuse by the caller, postconditions detect an error in the function itself (rather like assertions). Therefore, postconditions may access private state (subject to the caveat stated shortly, in “Postconditions and Overridden Methods” on page 526). Postconditions and Thread Safety Multithreaded scenarios (Chapter 14) challenge the usefulness of postconditions. For instance, suppose we wrote a thread-safe wrapper for a List with a method as follows: public class ThreadSafeList { List _list = new List(); object _locker = new object(); public bool AddIfNotPresent (T item) { Contract.Ensures (_list.Contains (item)); lock (_locker) { if (_list.Contains(item)) return false; _list.Add (item); return true; } } public void Remove (T item) { lock (_locker) _list.Remove (item); } } The postcondition in the AddIfNotPresent method is checked after the lock is re- leased—at which point the item may no longer exist in the list if another thread called Remove right then. There is currently no workaround for this problem, other than to enforce such conditions as assertions (see next section) rather than post- conditions. Contract.EnsuresOnThrow Occasionally, it’s useful to ensure that a certain condition holds should a particular type of exception be thrown. The EnsuresOnThrow method does exactly this: Contract.EnsuresOnThrow (this.ErrorMessage != null); Postconditions | 525 Diagnostics and Code Contracts Contract.Result and Contract.ValueAtReturn Because postconditions are not evaluated until a function ends, it’s reasonable to want to access the return value of a method. The Contract.Result method does exactly that: Random _random = new Random(); int GetOddRandomNumber() { Contract.Ensures (Contract.Result() % 2 == 1); return _random.Next (100) * 2 + 1; } The Contract.ValueAtReturn method fulfills the same function—but for ref and out parameters. Contract.OldValue Contract.OldValue returns the original value of a method parameter. This is use- ful with postconditions because the latter are checked at the end of a function. Therefore, any expressions in postconditions that incorporate parameters will read the modified parameter values. For example, the postcondition in the following method will always fail: static string Middle (string s) { Contract.Requires (s != null && s.Length >= 2); Contract.Ensures (Contract.Result().Length < s.Length); s = s.Substring (1, s.Length - 2); return s.Trim(); } Here’s how we can correct it: static string Middle (string s) { Contract.Requires (s != null && s.Length >= 2); Contract.Ensures (Contract.Result().Length < Contract.OldValue (s).Length); s = s.Substring (1, s.Length - 2); return s.Trim(); } Postconditions and Overridden Methods An overridden method cannot circumvent postconditions defined by its base, but it can add new ones. The binary rewriter ensures that a base method’s postconditions are always checked—even if the overridden method doesn’t call the base implementation. 526 | Chapter 13: Diagnostics and Code Contracts For the reason just stated, postconditions on virtual methods should not access private members. Doing so will result in the binary rewriter weaving code into the subclass that will try to access private members in the base class—causing a runtime error. Assertions and Object Invariants In addition to preconditions and postconditions, the code contracts API lets you make assertions and define object invariants. Assertions Contract.Assert You can make assertions anywhere in a function by calling Contract.Assert. You can optionally specify an error message if the assertion fails: ... int x = 3; ... Contract.Assert (x == 3); // Fail unless x is 3 Contract.Assert (x == 3, "x must be 3"); ... The binary rewriter doesn’t move assertions around. There are two reasons for fa- voring Contract.Assert over Debug.Assert: • You can leverage the more flexible failure-handling mechanisms offered by code contracts • Static checking tools can attempt to validate Contract.Asserts Contract.Assume Contract.Assume behaves exactly like Contract.Assert at run-time, but has slightly different implications for static checking tools. Essentially, static checking tools won’t challenge an assumption, whereas they may challenge an assertion. This is useful in that there will always be things a static checker is unable to prove, and this may lead to it “crying wolf” over a valid assertion. Changing the assertion to an assumption keeps the static checker quiet. Object Invariants For a class, you can specify one or more object invariant methods. These methods run automatically after every public function in the class, and allow you to assert that the object is in an internally consistent state. Assertions and Object Invariants | 527 Diagnostics and Code Contracts Support for multiple object invariant methods was included to make object invariants work well with partial classes. To define an object invariant method, write a parameterless void method and an- notate it with the [ContractInvariantMethod] attribute. In that method, call Con tract.Invariant to enforce each condition that should hold true if your object is a valid state: class Test { int _x, _y; [ContractInvariantMethod] void ObjectInvariant() { Contract.Invariant (_x >= 0); Contract.Invariant (_y >= _x); } public int X { get { return _x; } set { _x = value; } } public void Test1() { _x = −3; } void Test2() { _x = −3; } } The binary rewriter translates the X property, Test1 method and Test2 method to something equivalent to this: public void X { get { return _x; } set { _x = value; ObjectInvariant(); } } public void Test1() { _x = −3; ObjectInvariant(); } void Test2() { _x = −3; } // No change because it's private Object invariants don’t prevent an object from entering an in- valid state: they merely detect when that condition has occurred. Contract.Invariant is rather like Contract.Assert, except that it can appear only in a method marked with the [ContractInvariantMethod] attribute. And conversely, a contract invariant method can only contain calls to Contract.Invariant. A subclass can introduce its own object invariant method, too, and this will be checked in addition to the base class’s invariant method. The caveat, of course, is that the check will take place only after a public method is called. Contracts on Interfaces and Abstract Methods A powerful feature of code contracts is that you can attach conditions to interface members and abstract methods. The binary rewriter then automatically weaves these conditions into the members’ concrete implementations. 528 | Chapter 13: Diagnostics and Code Contracts A special mechanism lets you specify a separate contract class for interfaces and abstract methods, so that you can write method bodies to house the contract con- ditions. Here’s how it works: [ContractClass (typeof (ContractForITest))] interface ITest { int Process (string s); } [ContractClassFor (typeof (ITest))] sealed class ContractForITest : ITest { int ITest.Process (string s) // Must use explicit implementation. { Contract.Requires (s != null); return 0; // Dummy value to satisfy compiler. } } Notice that we had to return a value when implementing ITest.Process to satisfy the compiler. The code that returns 0 will not run, however. Instead, the binary rewriter extracts just the conditions from that method, and weaves them into the real implementations of ITest.Process. This means that the contract class is never actually instantiated (and any constructors that you write will not execute). You can assign a temporary variable within the contract block to make it easier to reference other members of the interface. For instance, if our ITest interface also defined a Message property of type string, we could write the following in ITest.Process: int ITest.Process (string s) { ITest test = this; Contract.Requires (s != test.Message); ... } This is easier than: Contract.Requires (s != ((ITest)this).Message); (Simply using this.Message won’t work because Message must be explicitly imple- mented.) The process of defining contract classes for abstract classes is exactly the same, except that the contract class should be marked abstract instead of sealed. Dealing with Contract Failure The binary rewriter lets you specify what happens when a contract condition fails, via the /throwonfailure switch (or the “Assert on Contract Failure” checkbox in Visual Studio’s Contracts tab in Project Properties). Dealing with Contract Failure | 529 Diagnostics and Code Contracts If you don’t specify /throwonfailure—or check “Assert on Contract Failure”—a dialog appears upon contract failure, allowing you to abort, debug or ignore the error. There are a couple of nuances to be aware of: • If the CLR is hosted (i.e., in SQL Server or Exchange), the host’s escalation policy is triggered instead of a dialog ap- pearing. • Otherwise, if the current process can’t pop up a dialog box to the user, Environment.FailFast is called. The dialog is useful in debug builds for a couple of reasons: • It makes it easy to diagnose and debug contract failures on the spot—without having to re-run the program. This works regardless of whether Visual Studio is configured to break on first-chance exceptions. And unlike with exceptions in general, contract failure almost certainly means a bug in your code. • It lets you know about contract failure—even if a caller higher up in the stack “swallows” exceptions as follows: try { // Call some method whose contract fails } catch { } The code above is considered an antipattern in most scenarios because it masks failures, including conditions that the author never anticipated. If you specify the /throwonfailure switch—or uncheck “Assert on Contract Failure” in Visual Studio—a ContractException is thrown upon failure. This is desirable for: • Release builds—where you would let the exception bubble up the stack and be treated like any other unexpected exception (perhaps by having a top-level ex- ception handler log the error or invite the user to report it). • Unit testing environments— where the process of logging errors is automated. ContractException cannot appear in a catch block because this type is not public. The rationale is that there’s no reason that you’d want to specifically catch a ContractException—you’d want to catch it only as part of a general exception backstop. 530 | Chapter 13: Diagnostics and Code Contracts The ContractFailed Event When a contract fails the static Contract.ContractFailed event fires before any fur- ther action is taken. If you handle this event, you can query the event arguments object for details of the error. You can also call SetHandled to prevent a ContractEx ception from being subsequently thrown (or a dialog appearing). Handling this event is particularly useful when /throwonfailure is specified, because it lets you log all contract failures—even if code higher in the call stack swallows exceptions as we described just before. A great example is with automated unit testing: Contract.ContractFailed += (sender, args) => { string failureMessage = args.FailureKind + ": " + args.Message; // Log failureMessage with unit testing framework: // ... args.SetUnwind(); }; This handler logs all contract failures, while allowing the normal ContractExcep tion (or contract failure dialog) to run its course after the event handler has finished. Notice that we also call SetUnwind: this neutralizes the effect of any calls to SetHan dled from other event subscribers. In other words, it ensures that a ContractExcep tion (or dialog) will always follow after all event handlers have run. If you throw an exception from within this handler, any other event handlers will still execute. The exception that you threw then populates the InnerException prop- erty of the ContractException that’s eventually thrown. Exceptions Within Contract Conditions If an exception is thrown within a contract condition itself, then that exception propagates like any other—regardless of whether /throwonfailure is specified. The following method throws a NullReferenceException if called with a null string: string Test (string s) { Contract.Requires (s.Length > 0); ... } This precondition is essentially faulty. It should instead be: Contract.Requires (!string.IsNullOrEmpty (s)); Selectively Enforcing Contracts The binary rewriter offers two switches that strip away some or all contract check- ing: /publicsurface and /level. You can control these from Visual Studio via the Code Contracts tab of Project Properties. The /publicsurface switch tells the re- writer to check contracts only on public members. Selectively Enforcing Contracts | 531 Diagnostics and Code Contracts The /level switch has the following options: None (Level 0) Strips out all contract verification ReleaseRequires (Level 1) Enables only calls to the generic version of Contract.Requires Preconditions (Level 2) Enables all preconditions (Level 1 plus normal preconditions) Pre and Post (Level 3) Enables Level 2 checking plus postconditions Full (Level 4) Enables Level 3 checking plus object invariants and assertions (i.e., everything) You typically enable full contract checking in debug build configurations. Contracts in Release Builds When it comes to making release builds, there are two general philosophies: • Favor safety and enable full contract checking • Favor performance and disable all contract checking If you’re building a library for public consumption, though, the second approach creates a problem. Imagine that you compile and distribute library L in release mode with contract checking disabled. A client then builds project C in debug mode that references library L. Assembly C can then call members of L incorrectly without contract violations! In this situation, you actually want to enforce the parts of L’s contract that ensure correct usage of L—in other words, the preconditions in L’s public members. The simplest way to resolve this is to enable /publicsurface checking in L with a level of Preconditions or ReleaseRequires. This ensures that the essential precondi- tions are enforced for the benefit of consumers, while incurring the performance cost of only those preconditions. In extreme cases, you might not want to pay even this small performance price—in which case you can take the more elaborate approach of call-site checking. Call-Site Checking Call-site checking moves precondition validation from called methods into calling methods (call sites). This solves the problem just described—by en