SQL 袖珍参考手册(第3版)


SQL Pocket Guide THIRD EDITION SQL Pocket Guide Jonathan Gennick Beijing • Cambridge • Farnham • Köln • Sebastopol • Tokyo SQL Pocket Guide, Third Edition by Jonathan Gennick Copyright © 2011 Jonathan Gennick. 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 promo- tional use. Online editions are also available for most titles (http://my.safari booksonline.com). For more information, contact our corporate/institutional sales department: (800) 998-9938 or corporate@oreilly.com. Editor: Julie Steele Copyeditor: Teresa Elsey Production Editor: Teresa Elsey Proofreader: Emily Quill Indexer: Ellen Troutman Zaig Cover Designer: Karen Montgomery Interior Designer: David Futato Illustrator: Robert Romano Printing History: March 2004: First Edition. April 2006: Second Edition. November 2010: Third Edition. Nutshell Handbook, the Nutshell Handbook logo, and the O’Reilly logo are registered trademarks of O’Reilly Media, Inc. The Pocket Guide series desig- nations, SQL Pocket Guide, the image of a chameleon, 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 author assume no responsibility for errors or omissions, or for damages resulting from the use of the information contained herein. ISBN: 978-1-449-39409-7 [TG] 1288815268 Contents SQL Pocket Guide 1 Introduction 1 Analytic Functions 7 CASE Expressions: Simple 7 CASE Expressions: Searched 7 CAST Function 8 CONNECT BY Queries 8 Data Type Conversion 15 Data Types: Binary Integer 15 Data Types: Character String 15 Data Types: Datetime 16 Data Types: Decimal 19 Datetime Conversions: DB2 21 Datetime Conversions: MySQL 24 Datetime Conversions: Oracle 28 Datetime Conversions: PostgreSQL 31 Datetime Conversions: SQL Server 34 Datetime Functions: DB2 38 Datetime Functions: MySQL 39 Datetime Functions: Oracle 40 Datetime Functions: PostgreSQL 43 Datetime Functions: SQL Server 45 v Deleting Data 47 EXTRACT Function 51 GREATEST 52 Grouping and Summarizing 52 Hierarchical Queries 62 Indexes, Creating 66 Indexes, Removing 67 Inserting Data 67 Joining Tables 72 LEAST 82 Literals 82 Merging Data 86 Nulls 88 Numeric Conversions: DB2 93 Numeric Conversions: MySQL 95 Numeric Conversions: Oracle 95 Numeric Conversions: PostgreSQL 97 Numeric Conversions: SQL Server 98 Numeric/Math Functions 99 OLAP Functions 101 Pivoting and Unpivoting 101 Predicates 109 Recursive Queries 112 Regular Expressions 113 Selecting Data 124 String Functions 134 Subqueries 139 Tables, Creating 143 Tables, Dropping 148 Tables, Modifying 149 Transaction Management 154 Union Queries 162 vi | Table of Contents Updating Data 168 Window Functions 173 Index 181 Table of Contents | vii SQL Pocket Guide Introduction This book is an attempt to cram the most useful information about SQL into a pocket-size guide. It covers commonly used syntax for the following platforms: IBM DB2 Release 9.7, MySQL 5.1, Oracle Database 11g Release 2, PostgreSQL 9.0, and Microsoft SQL Server 2008 Release 2. Not all syntax will work on all platforms, and some features may not be available in earlier releases of these products. Whenever possible, I’ve tried to note any product or release dependencies. Organization of This Book Topics are organized alphabetically, with many section names carefully chosen to correspond to relevant SQL keywords. For example, see “Inserting Data” on page 67 for help with the INSERT statement. Platform notes MySQL requires the leading parenthesis in a function invoca- tion to immediately follow the function name. For example, upper (name) will generate an error message because of the space between upper and (name). 1 Conventions The following typographical conventions are used in this book: UPPERCASE Indicates an SQL keyword lowercase Indicates a user-defined item in an SQL statement Italic Indicates emphasis or a new technical term Constant width Used for code examples and for in-text references to table names, column names, expressions, and so forth Constant width bold Indicates user input in input/output code examples Constant width italic Indicates an element of syntax you need to supply [] Denotes an optional element of syntax {} Denotes a required choice | Separates choices in syntax Example Data All example SQL statements in this book execute against a set of tables and data that you can download from this book’s catalog page at http://oreilly.com/catalog/9781449394097/. Figure 1 illustrates the relationships between the core tables, which give information on waterfalls in Michigan’s Upper Pen- insula. Some examples also use tables based on or derived from those in Figure 1. The terms datum, zone, northing, and easting refer to Universal Transverse Mercator (UTM) grid coordinates, such as those 2 | SQL Pocket Guide you might use with a topographical map or GPS device. For more, see http://erg.usgs.gov/isb/pubs/factsheets/fs07701.html. Some SQL examples in this book use a pivot table, which is nothing more than a single-column table containing sequen- tially numbered rows—in this case, 1,000 rows. The name of the table is pivot. (Exceptions! In SQL Server, pivot is a re- served word, so the SQL Server example script creates the table as pivvot, with two vs. In the MySQL script, the table dual is named duel.) 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 docu- mentation. You do not need to contact us for permission unless you’re reproducing a significant portion of the code. For ex- ample, 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 per- mission. 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 permission. We appreciate, but do not require, attribution. An attribution usually includes the title, author, publisher, and ISBN. For ex- ample: “SQL Pocket Guide, by Jonathan Gennick (O’Reilly). Copyright 2011 Jonathan Gennick, 9781449394097.” 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. Introduction | 3 Figure 1. Example schema for this book 4 | SQL Pocket Guide 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, exam- ples, and any additional information. You can access this page at: http://oreilly.com/catalog/9781449394097 To comment or ask technical questions about this book, send email to: bookquestions@oreilly.com For more information about our books, conferences, Resource Centers, and the O’Reilly Network, see our website at: http://oreilly.com Safari® Books Online Safari Books Online is an on-demand digital li- brary that lets you easily search over 7,500 tech- nology and creative reference books and videos to find the answers you need quickly. With a subscription, you can read any page and watch any video from our library online. Read books on your cell phone and mobile devices. Access new titles before they are available for print, and get exclusive access to manuscripts in develop- ment and post feedback for the authors. Copy and paste code samples, organize your favorites, download chapters, Introduction | 5 bookmark key sections, create notes, print out pages, and ben- efit from tons of other time-saving features. O’Reilly Media has uploaded this book to the Safari Books Online service. To have full digital access to this book and others on similar topics from O’Reilly and other publishers, sign up for free at http://my.safaribooksonline.com. Acknowledgments My heartiest thanks to the following people for their support, encouragement, and assistance: Grant Allen; Don Bales; Vladimir Begun; Tugrul Bingol; John Blake; Michel Cadot; Dias Costa; Chris Date; Bruno Denuit; Doug Doole; Chris Eaton; Stéphane Faroult; Iggy Fernandez; Bobby Fielding; Donna, Jenny, and Jeff Gennick; K. Gopalakrishnan; Jonah Harris; John Haydu; Kelvin Ho; Brand Hunt; Ken Jacobs; Chris Kempster; Stephen Lee; Peter Linsley; Jim Melton; Anthony Molinaro; Ari Mozes; Arup Nanda; Tanel Poder; Ted Rexstrew; Brandon Rich; Serge Rielau; Debby Russell; Andrew and Aaron Sears; Jeff Smith; Nuno Souto; Richard Swagerman; April Wells; and Fred Zemke. 6 | SQL Pocket Guide Analytic Functions Analytic function is Oracle’s term for what the SQL standard refers to as a window function. See the section “Window Func- tions” on page 173 for more on this extremely useful class of function. CASE Expressions: Simple Simple CASE expressions correlate a list of values to a list of alternatives. For example: SELECT u.name, CASE u.open_to_public WHEN 'y' THEN 'Welcome!' WHEN 'n' THEN 'Go Away!' ELSE 'Bad code!' END AS column_alias FROM upfall u; Simple CASE expressions are useful when you can directly link an input value to a WHEN clause by means of an equality con- dition. If no WHEN clause is a match, and no ELSE is specified, the expression returns null. CASE Expressions: Searched Searched CASE expressions associate a list of alternative return values with a list of true/false conditions. They also allow you to implement an IS NULL test. For example: SELECT u.name, CASE WHEN u.open_to_public = 'y' THEN 'Welcome!' WHEN u.open_to_public = 'n' THEN 'Go Away!' WHEN u.open_to_public IS NULL THEN 'Null!' ELSE 'Bad code!' END AS column_alias FROM upfall u; CASE Expressions: Searched | 7 Null is returned when no condition is TRUE and no ELSE is specified. If multiple conditions are TRUE, the first-listed con- dition takes precedence. CAST Function CAST explicitly converts a value to a new type. For example: SELECT * FROM upfall u WHERE u.id = CAST('1' AS INTEGER); When converting from text to numeric or date types, CAST offers little flexibility in dealing with different input data for- mats. For example, if the value you are casting is a string, the contents must conform to your database’s default text repre- sentation of the target data type. NOTE Most database brands have more useful conversion functions than CAST. SQL Server’s CONVERT function is one such example. See the sections on Datetime Con- versions and Numeric Conversions. CONNECT BY Queries Oracle Database supports CONNECT BY syntax for executing hierarchical queries. Beginning in Oracle Database 11g Release 2, you should consider the WITH clause, which in that release supports ISO standard syntax for recursive queries. See “Hierarchical Queries” on page 62. NOTE DB2 optionally supports CONNECT BY for compati- bility with Oracle. There are some limitations, and support needs to be enabled through B2_COMPATI- BILITY_VECTOR. 8 | SQL Pocket Guide Core CONNECT BY Syntax To return data in a hierarchy, specify a starting node using START WITH, and specify the parent-child relationship using CONNECT BY: SELECT id, name, type, parent_id FROM gov_unit START WITH parent_id IS NULL CONNECT BY parent_id = PRIOR id; ID NAME TYPE PARENT_ID ----- ---------- -------- --------- 3 Michigan state 2 Alger county 3 1 Munising city 2 4 Munising township 2 5 Au Train township 2 6 Baraga county 3 7 Ontonagon county 3 8 Interior township 7 9 Dickinson county 3 10 Gogebic county 3 11 Delta county 3 12 Masonville township 11 ... The START WITH clause identifies the row(s) Oracle consid- ers to be at the top of the tree(s). There is only one tree in this example, and it is for the state of Michigan. Alger County is a subdivision of Michigan. Munising and Au Train Townships are subdivisions of Alger County. Each entity’s parent_id points to its enclosing entity. Your START WITH condition does not necessarily need to involve the columns that link parent to child nodes. For ex- ample, use the following to generate a tree for each county: START WITH type = 'county' In a CONNECT BY query, the keyword PRIOR represents an operator that returns a column’s value from either the parent or a child row, depending on whether you are walking the tree top-down or bottom-up. PRIOR is often used to define the re- cursive relationship, but you can also use PRIOR in SELECT CONNECT BY Queries | 9 lists, WHERE clauses, or anywhere else a column reference is valid. Creative CONNECT BY CONNECT BY is not limited to hierarchical data. Any data linked in a recursive fashion is a candidate for CONNECT BY queries. For instance, the tour stops in this book’s example schema are linked in a fashion that CONNECT BY handles very well. The following query uses CONNECT BY to list each stop in its proper order: SELECT t.name tour_name, t.stop FROM trip t START WITH parent_stop IS NULL CONNECT BY parent_stop = PRIOR stop AND name = PRIOR name; Because some waterfalls appear in more than one tour, CON- NECT BY also includes a condition on tour_name to avoid loops. Output from the query is as follows: TOUR_NAME STOP ---------- ---------------------- M-28 3 M-28 1 M-28 8 M-28 9 M-28 10 M-28 11 Munising 1 Munising 2 Munising 6 Munising 4 Munising 3 Munising 5 US-2 14 US-2 12 US-2 11 US-2 13 You can also use CONNECT BY as a row generator. For ex- ample, to generate 100 rows (credit to Mikito Harakiri and Tom Kyte for showing me this clever trick), specify: 10 | SQL Pocket Guide SELECT level x FROM dual CONNECT BY level <= 100; Some older releases of Oracle have a bug that you can avoid by placing the logic into a subquery: SELECT x FROM ( SELECT level x FROM dual CONNECT BY level <= 100); You can also see the real-life case study “Finding Flight Legs” at http://gennick.com/flight.html. WHERE Clauses with CONNECT BY You can write WHERE clauses in CONNECT BY queries to restrict the results to specific rows of interest. The conditions in the CONNECT BY clause control which trees are processed by your query, and those trees in turn represent a candidate pool of rows. Conditions in the WHERE clause winnow down that candidate pool to only those rows that you wish the query to return. Joins with CONNECT BY A CONNECT BY query may involve a join, in which case the following order of operations applies: 1. The join is materialized first, which means that any join predicates are evaluated first. 2. The CONNECT BY processing is applied to the rows re- turned from the join operation. 3. Any filtering predicates from the WHERE clause are ap- plied to the results of the CONNECT BY operation. The following is an adaptation of the CONNECT BY query listing tour stops, which now incorporates a join to bring in the waterfall names: SELECT t.name tour_name, t.stop, u.name falls_name FROM trip t INNER JOIN upfall u ON t.stop = u.id CONNECT BY Queries | 11 START WITH parent_stop IS NULL CONNECT BY t.parent_stop = PRIOR t.stop AND t.name = PRIOR t.name; Be careful! Don’t write joins that inadvertently eliminate nodes from the hierarchy you are querying. Sorting CONNECT BY Results Oracle’s CONNECT BY syntax implies an ordering in which, given a top-down walk of the tree, each parent node is followed by its immediate children, each child is followed by its imme- diate children, and so on. It’s rare to write a standard ORDER BY clause into a CONNECT BY query, because the resulting sort destroys the hierarchical ordering of the data. However, beginning in Oracle9i Database, you can use the new ORDER SIBLINGS BY clause to sort each level independently without destroying the hierarchy: SELECT id, name, type, parent_id FROM gov_unit START WITH parent_id IS NULL CONNECT BY parent_id = PRIOR id ORDER SIBLINGS BY type, name; ID NAME TYPE PARENT_ID -- ---------- -------- ---------------------- 3 Michigan state 2 Alger county 3 1 Munising city 2 5 Au Train township 2 4 Munising township 2 6 Baraga county 3 ... Baraga County follows Alger County because both are at the same level and Baraga County comes later in the sorting order. Within Alger County, the city is listed before the two town- ships because the sort is on type first, followed by name. The two townships are then sorted in alphabetical order. Each level in the hierarchy is sorted independently, yet each parent is still followed by its immediate children. Thus, the hierarchy remains intact. 12 | SQL Pocket Guide Loops in Hierarchical Data Hierarchical data can sometimes be malformed in that a row’s child may also be that row’s parent or ancestor. Such a situa- tion leads to a loop. You can simulate a loop in the trip table by omitting AND t.name = PRIOR t.name from the CONNECT BY clause of the query to list tour stops. You can then detect that loop by adding NOCYCLE to the CONNECT BY clause and the CONNECT_BY_ISCYCLE pseudocolumn to the SELECT list: SELECT t.name tour_name, t.stop, u.name falls_name, CONNECT_BY_ISCYCLE FROM trip t INNER JOIN upfall u ON t.stop = u.id START WITH parent_stop IS NULL CONNECT BY NOCYCLE t.parent_stop = PRIOR t.stop; NOCYCLE prevents Oracle from following recursive loops in the data. CONNECT_BY_ISCYCLE returns 1 for any row having a child that is also a parent or ancestor. Here are the preceding query’s results: TOUR_NAME STOP FALLS_NAME CONNECT_BY_ISCYCLE --------- ---- -------------- ------------------ Munising 1 Munising Falls 0 Munising 2 Tannery Falls 0 Munising 6 Miners Falls 0 Munising 4 Wagner Falls 0 Munising 3 Alger Falls 1 ... The 1 in the fourth column indicates that a loop arises from the node for stop 3. If you look carefully at the data in the trip table, you’ll see two nodes where stop = 3. These nodes are for different tours. Without the restriction on t.name, one branch of recursive processing will go from stop 3 on the Mu- nising tour to stop 1 on the M-28 tour (child of a stop 3) to stop 2 on the Munising tour (child of a stop 1). Eventually, you’ll come again to stop 3 on the Munising tour, thereby creating the loop. CONNECT BY Queries | 13 Supporting Functions and Operators Oracle implements a number of helpful functions and opera- tors to use in writing CONNECT BY queries: CONNECT_BY_ISCYCLE Returns 1 when a row’s child is also its ancestor; other- wise, it returns 0. Use with CONNECT BY NOCYCLE. (Oracle Database 10g and higher.) CONNECT_BY_ISLEAF Returns 1 for leaf rows, 0 for rows with children. (Oracle Database 10g and higher.) CONNECT_BY_ROOT(column) Returns a value from the root row. See PRIOR. (Oracle Database 10g and higher.) LEVEL Returns 0 for the root node of a hierarchy, 1 for nodes just below the root, 2 for the next level of nodes, and so forth. LEVEL is commonly used in SQL*Plus to indent hierarch- ical results via an incantation such as the following: RPAD(' ', 2*(LEVEL-1)) || first_column PRIOR(column) or PRIOR column Returns a value from a row’s parent. See also CONNECT_BY_ROOT. SYS_CONNECT_BY_PATH (column , delimiter) Returns a concatenated list of column values in the path from the root to the current node. Each value is preceded by a delimiter, which you must specify as a string constant. Add SYS_CONNECT_BY_PATH(u.name,';') to the SELECT list of the tour query shown in “Joins with CONNECT BY” on page 11, and you’ll get results such as these: ;Alger Falls, ;Alger Falls;Munising Falls, ;Alger Falls;Munis ing Falls;Scott Falls, and so forth. (Oracle9i Database and higher.) 14 | SQL Pocket Guide Data Type Conversion See the following topics for help on type conversion: CAST Function EXTRACT Function Datetime Conversions for your chosen platform Numeric Conversions for your chosen platform Most platforms allow implicit conversion from one data type to another. Here’s an example in Oracle: SELECT * FROM upfall WHERE id = '1'; It’s often better to use explicit type conversion so that you know for sure which value is getting converted and how. Data Types: Binary Integer Except for Oracle, the platforms support the following binary integer types: SMALLINT INTEGER BIGINT These types correspond to 2-byte, 4-byte, and 8-byte integers, respectively. Ranges are −32,768 to 32,767; −2,147,483,648 to 2,147,483,647; and −9,223,372,036,854,775,808 to 9,223,372,036,854,775,807, respectively. Data Types: Character String For all platforms except Oracle, use the VARCHAR type to store character data: VARCHAR(max_bytes) MySQL allows TEXT as a synonym for VARCHAR: TEXT (max_bytes) Data Types: Character String | 15 In Oracle, append a 2 to get VARCHAR2: VARCHAR2(max_bytes) Oracle Database 9i and higher allows you to specify explicitly whether the size refers to bytes or characters: VARCHAR2(max_bytes BYTE) VARCHAR2(max_characters CHAR) Using Oracle’s CHAR option means that all indexing into the string (such as with SUBSTR) is performed in terms of charac- ters, not bytes. Maximums are 4,000 bytes (Oracle), 32,672 bytes (DB2), 8,000 bytes (SQL Server), 65,532 bytes (MySQL), and 1 GB (PostgreSQL). Data Types: Datetime Datetime support varies wildly among platforms; com- monality is virtually nonexistent. DB2 DB2 supports the following datetime types: DATE TIME TIMESTAMP TIMESTAMP(0to12default6) DATE stores year, month, and day. TIME stores hour, minute, and second. TIMESTAMP stores both date and time, to a frac- tional position of up to 12 digits. The range of valid values is from 1 A.D. through 9999 A.D. MySQL MySQL supports the following datetime types: DATE TIME 16 | SQL Pocket Guide DATETIME TIMESTAMP DATE stores dates from 1-Jan-1000 through 31-Dec-9999. TIME stores hour/minute/second values from −838:59:59 through 838:59:59. DATETIME stores both date and time of day (with the same range as DATE and TIME except that hours max out at 23). TIMESTAMP stores Unix timestamp values. The first TIMESTAMP column in a row is set to the current time in any INSERT or UPDATE, unless you specify explicitly a value of your own. Oracle Oracle supports the following datetime types: DATE TIMESTAMP TIMESTAMP WITH TIME ZONE TIMESTAMP WITH LOCAL TIME ZONE TIMESTAMP(0to9default6) ... DATE stores date and time to the second. TIMESTAMP adds fractional seconds. WITH TIME ZONE adds the time zone. WITH LOCAL TIME ZONE assumes each value to be in the same time zone as the database server, with time zone trans- lation taking place automatically between server and session time zones. The range of valid datetime values is from 4712 B.C. through 9999 A.D. You can specify a fractional precision of up to nine digits for any TIMESTAMP type. PostgreSQL PostgreSQL supports the following datetime types: DATE TIME [WITH[OUT] TIME ZONE] TIMESTAMP [WITH[OUT] TIME ZONE] TIME(0to6or0to10) ... TIMESTAMP(0to6) ... DATE stores a date only. TIME types store time of day. TIME- STAMP types store both date and time. The default is to Data Types: Datetime | 17 exclude time zone. The range of years is from 4713 B.C. through 294,276 A.D. (TIMESTAMPs) and 5,874,897 A.D. (DATEs). TIME and TIMESTAMP allow you to limit the number of pre- cision digits for fractional seconds. The range depends on whether PostgreSQL stores time using a DOUBLE PRECI- SION floating point (0 to 6) or BIGINT (0 to 10). The default is DOUBLE PRECISION. The choice is a compile-time option. Using BIGINT drops the high end of the TIMESTAMP year range to 294,276 A.D. SQL Server SQL Server supports the following datetime types: DATE DATETIME DATETIME2 DATETIME2(precision) DATETIMEOFFSET DATETIMEOFFSET(precision) SMALLDATETIME TIME TIME(precision) DATE stores date only from 1-Jan-0001 through 31-Dec-9999. DATETIME stores date and time of day to an increment of 3.33 milliseconds, with a range of 1-Jan-1753 through 31- Dec-9999. DATETIME2 is a combination of DATE and TIME. DATETIMEOFFSET extends DATETIME2 with a time zone offset. SMALLDATETIME stores date and time of day to the minute, with a range of 1-Jan-1900 through 6-Jun-2079. TIME stores time of day. DATETIME2, DATETIMEOFFSET, and TIME take an op- tional parameter to specify the decimal precision of the seconds value. The default precision is to store seconds to seven decimal places. The valid range is from 0 through 7. 18 | SQL Pocket Guide NOTE SQL Server supports a type called TIMESTAMP. It has nothing whatsoever to do with storing datetime values. Data Types: Decimal Decimal data types are rather more consistent across platforms than the datetime types. The following sections describe the more commonly used decimal types. DB2’s DECFLOAT Type DB2 9.5 and higher support a new DECFLOAT type that is based on the IEEE 754r standard. DB2 supports two precision choices: DECFLOAT(16) DECFLOAT(34) DECFLOAT(16) gives 16 digits of precision, requiring eight bytes of storage; DECFLOAT(34) gives 34 digits and requires 16 bytes of storage. The range for DECFLOAT(16) is: from –9.999999999999999 × 10384 to –1.0 × 10–383, and from 1.0 × 10–383 to 9.999999999999999 × 10384. The range for DECFLOAT(34) is: from –9.999999999999999999999999999999999 × 106144 to –1.0 × 10–6143, and from 1.0 × 10–6143 to 9.999999999999999999999999999999999 × 106144. Data Types: Decimal | 19 The DECFLOAT type supports five rounding modes: ROUND_CEILING Rounds upward, always in a positive direction. ROUND_FLOOR Rounds downward, always in a negative direction. ROUND_HALF_UP Rounds to the nearest up or down value. Values of 0.5 round upward. ROUND_HALF_EVEN Rounds to the nearest value. Values of 0.5 round up or down so as to make the final digit an even digit. ROUND_DOWN Rounds toward zero. You specify the rounding mode at the database level, using the parameter decflt_rounding. You must restart the database for any change to take effect. DECIMAL/NUMBER Type All platforms support the use of DECIMAL for storing numeric base-10 data (such as monetary amounts): DECIMAL DECIMAL(precision) DECIMAL(precision, scale) In Oracle, DECIMAL is a synonym for NUMBER, and you should generally use NUMBER instead. DECIMAL(precision) is a decimal integer of up to precision dig- its. DECIMAL(precision, scale) is a fixed-point decimal number of precision digits with scale digits to the right of the decimal point. For example, DECIMAL(9,2) can store values up to 9,999,999.99. 20 | SQL Pocket Guide NOTE In Oracle, declaring a column as DECIMAL without specifying precision or scale results in a decimal float- ing-point column. In DB2, the same declaration is in- terpreted as DECIMAL(5,0). In SQL Server, the effect is the same as DECIMAL(18,0). Maximum precision/scale values are: 38/127 (Oracle), 31/31 (DB2), 38/38 (SQL Server), 65/30 (MySQL), and 1,000/1,000 (PostgreSQL). Datetime Conversions: DB2 DB2 recently added a great deal of support to emulate Oracle’s TO_CHAR and TO_DATE functions. If compatibility with Oracle is important, test to see whether the functions described under “Datetime Conversions: Oracle” on page 28 will work for you. Otherwise, use the following functions to convert to and from dates, times, and timestamps. In the syntax, datetime can be a date, time, or timestamp; date must be either a date or a time- stamp; time must be either a time or a timestamp; and time stamp must be a timestamp. Similarly, dateduration must be a date or timestamp duration; timeduration must be either a time or timestamp duration; and timestampduration must be a time- stamp duration. Valid string representations of all of these types are allowed as well: BIGINT(datetime) CHAR(datetime, [ISO|USA|EUR|JIS|LOCAL]) DATE(date) DATE(integer) DATE('yyyyddd') DAY(date) DAY(dateduration) DAYNAME(date) DAYOFWEEK(date) DAYOFWEEK_ISO(date) Datetime Conversions: DB2 | 21 DAYOFYEAR(date) DAYS(date) DECIMAL(datetime[,precision[,scale]]) GRAPHIC(datetime, [ISO|USA|EUR|JIS|LOCAL]) HOUR(time) HOUR(timeduration) INTEGER(date_only) INTEGER(time_only) JULIAN_DAY(date) MICROSECOND(timestamp) MICROSECOND(timestampduration) MIDNIGHT_SECONDS(time) MINUTE(time) MINUTE(timeduration) MONTH(date) MONTH(dateduration) MONTHNAME(date) QUARTER(date) SECOND(time) SECOND(timeduration) TIME(time) TIMESTAMP(timestamp) TIMESTAMP(date, time) TIMESTAMP_FORMAT(string, 'YYYY-MM-DD HH24:MI:SS') TIMESTAMP_ISO(datetime) TO_CHAR(timestamp, 'YYYY-MM-DD HH24:MI:SS') TO_DATE(string, 'YYYY-MM-DD HH24:MI:SS') VARCHAR(datetime) VARCHAR_FORMAT(timestamp, 'YYYY-MM-DD HH24:MI:SS') VARGRAPHIC(datetime, [ISO|USA|EUR|JIS|LOCAL]) WEEK(date) WEEK_ISO(date) YEAR(date) YEAR(dateduration) The following example combines the use of several functions to produce a text representation of confirmed_date: SELECT u.id, MONTHNAME(u.confirmed_date) || ' ' || RTRIM(CHAR(DAY(u.confirmed_date))) || ',' || RTRIM(CHAR(YEAR(u.confirmed_date))) confirmed FROM upfall u; 22 | SQL Pocket Guide ID CONFIRMED ----------- --------------- 1 December 8,2005 2 December 8,2005 3 December 8,2005 4 December 8,2005 Functions requiring date, time, or timestamp arguments also accept character strings that can be converted implicitly into values of those types. For example: SELECT DATE('2003-11-7') , TIME('21:25:00'), TIMESTAMP('2003-11-7 21:25:00.00') FROM pivot WHERE x = 1; Use the CHAR function’s second argument to exert some con- trol over the output format of dates, times, and timestamps: SELECT CHAR(current_date, ISO), CHAR(current_date, LOCAL), CHAR(current_date, USA) FROM pivot WHERE x=1; 2003-11-06 11-06-2003 11/06/2003 Use the DATE function to convert an integer to a date. Valid integers range from 1 to 3,652,059, where 1 represents 1- Jan-0001. The DAYS function converts in the reverse direction: SELECT DATE(716194), DAYS('1961-11-15') FROM pivot WHERE x=1; 11/15/1961 716194 Use the DECIMAL and BIGINT functions to return dates, times, and timestamps as decimal and 8-byte integer values, which will take the forms yyyymmdd, hhmmss, and yyyymmddhhmmss.nnnnnnn, respectively: SELECT DECIMAL(current_date), DECIMAL(current_time), DECIMAL(current_timestamp) FROM pivot WHERE x=1; 20031106. 213653. 20031106213653.088001 Datetime Conversions: DB2 | 23 The JULIAN_DAY function returns the number of days since 1-Jan-4713 B.C. (which is the same as 1-Jan in the astronomical year −4712), counting that date as day 0. There is no function to convert in the reverse direction. Datetime Conversions: MySQL MySQL implements a variety of datetime conversion func- tions, including some in support of Unix timestamps. The available functions are described in the following subsections. Date and Time Elements MySQL supports the following functions to return specific date and time elements: DAYOFWEEK(date) WEEKDAY(date) DAYOFMONTH(date) DAYOFYEAR(date) MONTH(date) DAYNAME(date) MONTHNAME(date) QUARTER(date) WEEK(date) WEEK(date, first) YEAR(date) YEARWEEK(date) YEARWEEK(date, first) HOUR(time) MINUTE(time) SECOND(time) For example, to return the current date in text form, specify: SELECT CONCAT(DAYOFMONTH(CURRENT_DATE), '-', MONTHNAME(CURRENT_DATE), '-', YEAR(CURRENT_DATE)); 2-January-2004 For functions taking a first argument, you can specify whether weeks begin on Sunday (first = 0) or on Monday (first = 1). 24 | SQL Pocket Guide TO_DAYS and FROM_DAYS Use TO_DAYS to convert a date into the number of days since the beginning of the Christian calendar (1-Jan-0001 is consid- ered day 1): SELECT TO_DAYS(CURRENT_DATE); 731947 Use FROM_DAYS to convert in the reverse direction: SELECT FROM_DAYS(731947); 2004-01-02 These functions are designed for use only with Gregorian dates, which begin on 15-Oct-1582. TO_DAYS and FROM_DAYS functions will not return correct results for earlier dates. Unix Timestamp Support The following functions convert to and from Unix timestamps: UNIX_TIMESTAMP([ date ]) Returns a Unix timestamp, which is an unsigned integer with the number of seconds since 1-Jan-1970. With no argument, you generate the current timestamp. The date argument may be a date string, a datetime string, a time- stamp, or a numeric equivalent. FROM_UNIXTIME(unix_timestamp [, format ]) Converts a Unix timestamp into a displayable date and time using the format you specify, if any. See Table 1 for a list of valid format elements. For example, to convert 4-Jan-2004 at 7:18 PM into the num- ber of seconds since 1-Jan-1970, specify: SELECT UNIX_TIMESTAMP(20040104191800); 1073261880 Datetime Conversions: MySQL | 25 To convert that timestamp into a human-readable format, specify: SELECT FROM_UNIXTIME(1073261880, '%M %D, %Y at %h:%i:%r'); January 4th, 2004 at 07:18:07:18:00 PM The format argument is optional. The default format for the datetime given in this example is 2004-01-04 19:18:00. Seconds in the Day Two MySQL functions let you work in terms of seconds in the day: SEC_TO_TIME(seconds) Converts seconds past midnight into a string of the form hh:mi:ss. TIME_TO_SEC(time) Converts a time into seconds past midnight. For example: SELECT TIME_TO_SEC('19:18'), SEC_TO_TIME(69480); 69480 19:18:00 DATE_FORMAT and TIME_FORMAT These two functions provide a great deal of flexibility in con- versions to text. Use DATE_FORMAT to convert dates to text and TIME_FORMAT to convert times: SELECT DATE_FORMAT(CURRENT_DATE, '%W, %M %D, %Y'); Sunday, January 4th, 2004 The second argument to both functions is a format string. For- mat elements in that format string are replaced with their re- spective datetime elements, as described in Table 1. Other text in the format string, such as the commas and spaces in this example, is left in place as part of the function’s return value. 26 | SQL Pocket Guide Table 1. MySQL date format elements Specifier Description %a Weekday abbreviation: Sun, Mon, Tue,… %b Month abbreviation: Jan, Feb, Mar,… %c Month number: 1, 2, 3,… %D Day of month with suffix: 1st, 2nd, 3rd,… %d Day of month, two digits: 01, 02, 03,… %e Day of month: 1, 2, 3,… %f Microseconds: 000000–999999 %H Hour, two digits, 24-hour clock: 00…23 %h Hour, two digits, 12-hour clock: 01…12 %I Hour, two digits, 12-hour clock: 01…12 %i Minutes: 00, 01,…59 %j Day of year: 001…366 %k Hour, 24-hour clock: 0, 1,…23 %l Hour, 12-hour clock: 1, 2,…12 %M Month name: January, February,… %m Month number: 01, 02,…12 %p Meridian indicator: AM or PM %r Time of day on a 12-hour clock, e.g., 12:15:05 PM %S Seconds: 00, 01,…59 %s Same as %S %T Time of day on a 24-hour clock, e.g., 12:15:05 (for 12:15:05 PM) %U Week with Sunday as the first day: 00, 01,…53 %u Week with Monday as the first day: 00, 01,…53 %V Week with Sunday as the first day, beginning with 01 and corresponding to %X: 01, 02,…53 %v Week with Monday as the first day, beginning with 01 and corresponding to %x: 01, 02,…53 %W Weekday name: Sunday, Monday,… Datetime Conversions: MySQL | 27 Specifier Description %w Numeric day of week: 0=Sunday, 1=Monday,… %X Year for the week, four digits, with Sunday as the first day and corresponding to %V %x Year for the week, four digits, with Monday as the first day and corresponding to %v %Y Four-digit year: 2003, 2004,… %y Two-digit year: 03, 04,… %% Places the percent sign (%) in the output Datetime Conversions: Oracle You can convert to and from datetime types in Oracle by using the following functions: TO_CHAR({datetime|interval}, format) TO_DATE(string, format) TO_TIMESTAMP(string, format) TO_TIMESTAMP_TZ(string, format) TO_DSINTERVAL('D HH:MI:SS') TO_YMINTERVAL('Y-M') NUMTODSINTERVAL(number, 'unit_ds') NUMTOYMINTERVAL(number, 'unit_ym') unit_ds ::= {DAY|HOUR|MINUTE|SECOND} unit_ym ::= {YEAR|MONTH} The format argument allows great control over text represen- tation. For example, you can specify precisely the display format for dates: SELECT name, TO_CHAR(confirmed_date, 'dd-Mon-yyyy') cdate FROM upfall; Munising Falls 08-Dec-2005 Tannery Falls 08-Dec-2005 Alger Falls 08-Dec-2005 … 28 | SQL Pocket Guide And to convert in the other direction: INSERT INTO upfall (id, name, confirmed_date) VALUES (15, 'Tahquamenon', TO_TIMESTAMP('29-Jan-2006','dd-Mon-yyyy')); Table 2 lists the format elements that you can use in creating a format mask. Output from many of the elements depends on your session’s current language setting (e.g., if your session language is French, you’ll get month names in French). When converting to text, the case of alphabetic values, such as month abbreviations, is determined by the case of the format element. Thus, 'Mon' yields 'Jan' and 'Feb', 'mon' yields 'jan' and 'feb', and 'MON' yields 'JAN' and 'FEB'. When con- verting from text, case is irrelevant. The format mask is always optional. You can omit it when your input value conforms to the default format specified by the following: NLS_DATE_FORMAT (dates) for dates, NLS_TIMESTAMP_FORMAT for timestamps, and NLS_TIMESTAMP_TZ_FORMAT for timestamps with time zones. You can query the NLS_SESSION_PARAMETERS view to check your NLS settings. Table 2. Oracle datetime format elements Element Description AM or PM A.M. or P.M. Meridian indicator. BC or AD B.C. or A.D. B.C. or A.D. indicator. CC Century. Output-only. D Day in the week. DAY, Day, or day Name of day. DD Day in the month. DDD Day in the year. DL Long date format. Output-only. Combines only with TS. Datetime Conversions: Oracle | 29 Element Description DS Short date format. Output-only. Combines only with TS. DY, Dy, or dy Abbreviated name of day. E Abbreviated era name for Japanese Imperial, ROC Official, and Thai Buddha calendars. Input-only. EE Full era name. FF, FF1…FF9 Fractional seconds. Only for TIMESTAMP values. Always use two Fs. FF1…FF9 work in Oracle Database 10g and higher. FM Toggles blank suppression. Output-only. FX Requires exact pattern matching on input. HH or HH12 Hour in the day, from 1–12. HH12 is output-only. HH24 Hour in the day, from 0–23. IW ISO week in the year. Output-only. IYY, IY, or I Last three, two, or one digits of ISO year. Output-only. IYYY ISO year. Output-only. J Julian date. January 1, 4712 B.C. is day 1. MI Minutes. MM Month number. MON, Mon, or mon Abbreviated name of month. MONTH, Month, or month Name of month. Q Quarter of year. Output-only. RM or rm Roman numeral month number. RR Last two digits of year. Sliding window for hundreds value: 00–49 = 20xx, 50–99 = 19xx. RRRR Four-digit year; also accepts two digits on input. Sliding window just like RR. SCC Century. B.C. dates negative. Output-only. SP Suffix that converts a number to its spelled format. SPTH Suffix that converts a number to its spelled and ordinal formats. SS Seconds. 30 | SQL Pocket Guide Element Description SSSSS Seconds since midnight. SYEAR, SYear, or syear Year in words. B.C. dates negative. Output-only. SYYYY Four-digit year. B.C. dates negative. TH or th Suffix that converts a number to ordinal format. TS Short time format. Output-only. Combine only with DL or DS. TZD Abbreviated time zone name. Input-only. TZH Time zone hour displacement from UTC (Coordinated Universal Time). TZM Time zone minute displacement from UTC. TZR Time zone region. W Week in the month, from 1 through 5. Week 1 starts on the first day of the month and ends on the seventh. Output-only. WW Week in the year, from 1 through 53. Output-only. X Local radix character used to denote the decimal point. This is a period in American English. Y,YYY Four-digit year with comma. YEAR, Year, or year Year in words. Output-only. YYY, YY, or Y Last three, two, or one digits of year. YYYY Four-digit year. Datetime Conversions: PostgreSQL Convert between datetimes and character strings using the fol- lowing functions: TO_CHAR({timestamp|interval}, format) TO_DATE(string, format) TO_TIMESTAMP(string, format) Datetime Conversions: PostgreSQL | 31 For example, to convert a date to the character representation of a timestamp, specify: SELECT u.name, TO_CHAR(u.confirmed_date, 'dd-Mon-YYYY') FROM upfall u; name | to_char -----------------+------------- Munising Falls | 08-Dec-2005 Tannery Falls | 08-Dec-2005 Alger Falls | 08-Dec-2005 ... To convert in the other direction (a character representation of a timestamp to a date), specify: SELECT TO_DATE('8-Dec-2005', 'dd-mon-yyyy'); PostgreSQL closely follows Oracle in its support for format elements. Table 3 lists those available in PostgreSQL. Case fol- lows form for alphabetic elements: use MON to yield JAN, FEB; Mon to yield Jan, Feb; and mon to yield jan, feb. WARNING You cannot apply TO_CHAR to values of type TIME. You can also use TO_TIMESTAMP to convert a Unix epoch value to a PostgreSQL timestamp: SELECT TO_TIMESTAMP(0); Unix time begins at midnight, at the beginning of 1-Jan-1970, Coordinated Universal Time (UTC). Table 3. PostgreSQL datetime format elements Element Description AM or PM A.M. or P.M. Meridian indicator. BC or AD B.C. or A.D. B.C. or A.D. indicator. 32 | SQL Pocket Guide Element Description CC Century. Output-only. D Day in the week. DAY, Day, or day Name of day. DD Day in the month. DDD Day in the year. DY, Dy, or dy Abbreviated name of day. FM Toggles blank suppression. Output-only. FX Requires exact pattern matching on input. HH or HH12 Hour in the day, from 1–12. HH12 is output-only. HH24 Hour in the day, from 0–23. IW ISO week in the year. Output-only. IYY, IY, or I Last three, two, or one digits of ISO standard year. Output-only. IYYY ISO standard year. Output-only. J Julian date. January 1, 4712 B.C. is day 1. MI Minutes. MM Month number. MON, Mon, or mon Abbreviated name of month. MONTH, Month, or month Name of month. MS Milliseconds. Q Quarter of year. Output-only. RM or rm Month number in Roman numerals. SP Suffix that converts a number to its spelled format (not implemented). SS Seconds. SSSS Seconds since midnight. TH or th Suffix that converts a number to ordinal format. TZ or tz Time zone name. US Microseconds. Datetime Conversions: PostgreSQL | 33 Element Description W Week in the month, from 1 through 5. Week 1 starts on the first day of the month and ends on the seventh. Output-only. WW Week in the year, from 1 through 53. Output-only. Y,YYY Four-digit year with comma. YYY, YY, or Y Last three, two, or one digits of year. YYYY Four-digit year. Datetime Conversions: SQL Server In SQL Server, you can choose one of four overall approaches to datetime conversion. The CONVERT function is a good general choice, although DATENAME and DATEPART pro- vide a great deal of flexibility when converting to text. CAST and SET DATEFORMAT SQL Server supports the standard CAST function and also allows you to specify a datetime format using the SET DATE- FORMAT command: SET DATEFORMAT dmy SELECT CAST('1/12/2004' AS datetime) 2004-12-01 00:00:00.000 For dates in unambiguous formats, you may not need to worry about the DATEFORMAT setting: SET DATEFORMAT dmy SELECT CAST('12-Jan-2004' AS datetime) 2004-01-12 00:00:00.000 When using SET DATEFORMAT, you can specify any of the following arguments: mdy, dmy, ymd, myd, dym. 34 | SQL Pocket Guide CONVERT You can use the CONVERT function for general datetime conversions: CONVERT(datatype[(length)], expression[, style]) The optional style argument allows you to specify the target and source formats for datetime values, depending on whether you are converting to or from a character string. Table 4 lists the supported styles. For example, you can convert to and from text: SELECT CONVERT(VARCHAR, CONVERT(DATETIME, '15-Nov-1961', 106), 106) 15 Nov 1961 Use the length argument if you want to specify the length of the resulting character string type. Subtract 100 from most style numbers for two-digit years: SELECT CONVERT(DATETIME, '1/1/50', 1) 1950-01-01 00:00:00.000 SELECT CONVERT(DATETIME, '49.1.1', 2) 2049-01-01 00:00:00.000 SQL Server uses the year 2049 as a cutoff. Years 50–99 are interpreted as 1950–1999. Years 00–49 are treated as 2000– 2049. You can see this behavior in the preceding example. Be aware that your DBA can change the cutoff value using the two digit year cutoff configuration option. Datetime Conversions: SQL Server | 35 Table 4. SQL Server datetime styles Style Description 0, 100 Default: mon dd yyyy hh:miAM (or PM) 101a USA: mm/dd/yyyy 102a ANSI: yyyy.mm.dd 103a British/French: dd/mm/yyyy 104a German: dd.mm.yyyy 105a Italian: dd-mm-yyyy 106a dd mon yyyy 107a mon dd, yyyy 108a hh:mm:ss 9, 109 Default with milliseconds: mon dd yyyy hh:mi:ss: mmmAM (or PM) 110a USA: mm-dd-yyyy 111a Japan: yyyy/mm/dd 112a ISO: yyyymmdd 13, 113 Europe default with milliseconds and 24-hour clock: dd mon yyyy hh:mm:ss:mmm 114a hh:mi:ss:mmm with 24-hour clock 20, 120 ODBC canonical, 24-hour clock: yyyy-mm-dd hh:mi:ss 21, 121 ODBC canonical with milliseconds, 24-hour clock: yyyy-mm-dd hh:mi:ss.mmm 126 ISO8601, no spaces: yyyy-mm-yyThh:mm:ss:mmm 127 Time with time zone (literal T separating the date from the time): yyyy-mm- ddThh:mi:ss.mmm 130 Hijri: dd mon yyyy hh:mi:ss:mmmAM 131 Hijri: dd/mm/yyyy hh:mi:ss:mmmAM a Subtract 100 to get a two-digit year. DATENAME and DATEPART Use the DATENAME and DATEPART functions to extract specific elements from datetime values: 36 | SQL Pocket Guide DATENAME(datepart, datetime) DATEPART(datepart, datetime) DATENAME returns a textual representation, whereas DATEPART returns a numeric representation. For example: SELECT DATENAME(month, GETDATE()), DATEPART(month, GETDATE()) January 1 Some elements, such as year and day, are always represented as numbers; however, the two functions give you the choice of getting back a string or an actual numeric value. Both of the following function calls return the year, but DATENAME returns the string '2004', whereas DATEPART returns the number 2004: SELECT DATENAME(year, GETDATE()), DATEPART(year, GETDATE()); SQL Server supports the following datepart keywords: year, yy, yyyy, quarter, qq, q, month, mm, m, dayofyear, dy, y, day, dd, d, week, wk, ww, weekday, dw, hour, hh, minute, mi, n, second, ss, s, millisecond, ms, microsecond, mcs, nanosecond, ns, TZoffset, tz, ISO_Week, isowk, isoww. DAY, MONTH, and YEAR SQL Server also supports a few functions to extract specific values from dates: DAY(datetime) MONTH(datetime) YEAR(datetime) For example: SELECT DAY(CURRENT_TIMESTAMP), MONTH(CURRENT_TIMESTAMP), YEAR(CURRENT_TIMESTAMP) 11 11 2003 Datetime Conversions: SQL Server | 37 Datetime Functions: DB2 DB2 implements the following special registers to return date- time information: CURRENT DATE or CURRENT_DATE Returns the current date on the server. CURRENT TIME or CURRENT_TIME Returns the current time on the server. CURRENT TIMESTAMP or CURRENT_TIMESTAMP Returns the current date and time as a timestamp. CURRENT TIMEZONE or CURRENT_TIMEZONE Returns the current time zone as a decimal number rep- resenting the time zone offset—in hours, minutes, and seconds—from UTC. The first two digits are the hours, the second two digits are the minutes, and the last two digits are the seconds. DB2 also supports labeled durations. For example: CURRENT_DATE + 1 YEARS - 3 MONTHS + 10 DAYS Valid labels are YEAR, YEARS, MONTH, MONTHS, DAY, DAYS, HOUR, HOURS, MINUTE, MINUTES, SECOND, SECONDS, MICROSECOND, and MICROSECONDS. NOTE DB2 9.7 and higher now support many of the same functions as Oracle, notably: ROUND, TRUNC, ADD_MONTHS, LAST_DAY, NEXT_DAY, and MONTHS_BETWEEN. See “Datetime Functions: Ora- cle” on page 40 for details. 38 | SQL Pocket Guide Datetime Functions: MySQL MySQL implements the following functions to return the cur- rent date and time: CURDATE() or CURRENT_DATE Returns the current date as a string ('YYYY-MM-DD') or a number (YYYYMMDD), depending on the context. CURTIME() or CURRENT_TIME Returns the current time as a string ('HH:MI:SS') or a number (HHMISS), depending on the context. NOW(), SYSDATE(), or CURRENT_TIMESTAMP Returns the current date and time as a string ('YYYY-MM-DD HH:MI:SS') or a number (YYYYMMDDHHMISS), depending on the context. UNIX_TIMESTAMP Returns the number of seconds since the beginning of 1- Jan-1970 as an integer. MySQL also implements the following functions for adding and subtracting intervals from dates. DATE_ADD(date , INTERVAL value units) Adds value number of units to the date. You can use ADDDATE as a synonym for DATE_ADD. DATE_SUB(date , INTERVAL value units) Subtracts value number of units from the date. You can use SUBDATE as a synonym for DATE_SUB. For example, to add one month to the current date: SELECT DATE_ADD(CURRENT_DATE, INTERVAL 1 MONTH); Or, to subtract one year and two months: SELECT DATE_SUB(CURRENT_DATE, INTERVAL '1-2' YEAR_MONTH); Valid interval keywords for numeric intervals include SEC- OND, MINUTE, HOUR, DAY, MONTH, and YEAR. You can also use the string-based formats shown in Table 5. Datetime Functions: MySQL | 39 Table 5. MySQL string-based interval formats Keyword Format DAY_HOUR 'dd hh' DAY_MINUTE 'dd hh:mi' DAY_SECOND 'dd hh:mi:ss' HOUR_MINUTE 'HH:MI' HOUR_SECOND 'hh:mi:ss' MINUTE_SECOND 'MI:SS' YEAR_MONTH 'yy-mm' Datetime Functions: Oracle Oracle implements a wide variety of helpful functions for working with dates and times. Getting Current Date and Time It is common to invoke SYSDATE to return the current date and time in the server’s time zone. For example: SELECT SYSDATE FROM dual; 2006-02-07 09:32:32 You can use ALTER SESSION to specify a default date format for your session using the date format elements described in Table 2. ALTER SESSION SET NLS_DATE_FORMAT = 'dd-Mon-yyyy hh: mi:ss'; The following Oracle functions return current datetime information: CURRENT_DATE Returns the current date in the session time zone as a value of type DATE. 40 | SQL Pocket Guide CURRENT_TIMESTAMP[(precision)] Returns the current date and time in the session time zone as a value of type TIMESTAMP WITH TIME ZONE. The precision is the number of decimal digits used to express fractional seconds; it defaults to 6. LOCALTIMESTAMP[(precision)] The same as CURRENT_TIMESTAMP, but it returns a TIMESTAMP value with no time zone offset. SYSDATE Returns the server date and time as a DATE. SYSTIMESTAMP[(precision)] Returns the current server date and time as a TIME- STAMP WITH TIME ZONE value. DBTIMEZONE Returns the database server time zone as an offset from UTC in the form '[+|-]hh:mi'. SESSIONTIMEZONE Returns the session time zone as an offset from UTC in the form '[+|-]hh:mi'. Rounding and Truncating Oracle allows you to round and truncate DATE values to spe- cific datetime elements. The following example illustrates rounding and truncating to the nearest month: SELECT SYSDATE, ROUND(SYSDATE,'Mon'), TRUNC(SYSDATE,'Mon') FROM dual; SYSDATE ROUND(SYSDA TRUNC(SYSDA ----------- ----------- ----------- 31-Dec-2003 01-Jan-2004 01-Dec-2003 Rounding is implemented to the nearest occurrence of the ele- ment you specify. My input date was closer to 1-Jan-2004 than it was to 1-Dec-2003, so my date was rounded up to the nearest month. Datetime Functions: Oracle | 41 Truncation simply sets any element of lesser significance than the one you specify to its minimum value. The minimum day value is 1, so 31-Dec was truncated to 1-Dec. Use the date format elements from Table 2 to specify the ele- ment for which you want to round or truncate a date. Avoid esoteric elements such as RM (Roman numerals) and J (Julian day); stick to easily understood elements such as MM (month), Q (quarter), and so forth. If you omit the second argument to ROUND or TRUNC, the date is rounded or truncated to the day (the DD element). Other Oracle Datetime Functions The following functions work with, and usually return, values of type DATE: ADD_MONTHS(date , integer) Adds integer months to date. If date is the last day of its month, the result is forced to the last day of the target month. If the target month has fewer days than date’s month, the result is also forced to the last of the month. LAST_DAY(date) Returns the last day of the month that contains a specified date. NEXT_DAY(date , weekday) Returns the first specified weekday following a given date. The weekday must be a valid weekday name or ab- breviation in the current date language for the session. (You can query NLS_SESSION_PARAMETERS to check this value.) Even when date falls on weekday, the function will still return the next occurrence of weekday. MONTHS_BETWEEN(later_date , earlier_date) Computes the number of months between two dates. The math corresponds to later_date – earlier_date. The in- put dates can actually be in either order, but if the second date is later, the result will be negative. 42 | SQL Pocket Guide The result will be an integer number of months for any case in which both dates correspond to the same day of the month, or for any case in which both dates correspond to the last day of their respective months. Otherwise, Ora- cle calculates a fractional result based on a 31-day month, also considering any time-of-day components of the input dates. None of these functions is overloaded to handle TIMESTAMP values. Any timestamp inputs are converted implicitly to type DATE and consequently lose any fractional second and time zone information. Datetime Functions: PostgreSQL The following subsections demonstrate some of PostgreSQL’s more useful datetime functions. Getting Current Date and Time PostgreSQL implements the following functions to return the current date and time: CURRENT_DATE CURRENT_TIME CURRENT_TIMESTAMP CURRENT_TIME [(precision)] CURRENT_TIMESTAMP [(precision)] LOCALTIME LOCALTIMESTAMP LOCALTIME [(precision)] LOCALTIMESTAMP [(precision)] NOW() The function NOW() is equivalent to CURRENT_TIME- STAMP. The CURRENT functions return values with a time zone. The LOCAL functions return values without a time zone. Datetime Functions: PostgreSQL | 43 For example: SELECT TO_CHAR(CURRENT_TIMESTAMP, 'HH:MI:SS tz'), TO_CHAR(LOCALTIMESTAMP, 'HH:MI:SS tz'); 05:02:00 est | 05:02:00 Some functions accept an optional precision argument. You can omit the argument to receive the fullest possible precision. Alternatively, you can use the argument to round to precision digits to the right of the decimal. For example: SELECT CURRENT_TIME, CURRENT_TIME(1); 17:10:07.490077-05 | 17:10:07.50-05 None of the previously listed functions advance their return values during a transaction. You will always get the date and time at which the current transaction began. The function TIMEOFDAY() is an exception to this rule: SELECT TIMEOFDAY(); Sun Feb 05 17:11:39.659280 2006 EST TIMEOFDAY() returns wall-clock time, advances during a transaction, and returns a character-string result. Rounding and Truncating PostgreSQL does not support the rounding of datetime values; however, it does provide a DATE_TRUNC function for trun- cating a datetime: SELECT CURRENT_DATE, DATE_TRUNC('YEAR', CURRENT_DATE); 2006-02-05 | 2006-01-01 00:00:00-05 The result is either a TIMESTAMP or an INTERVAL, depend- ing on what type of value is being truncated. The following are valid values for DATE_TRUNC’s first argument: MICROSEC- ONDS, MILLISECONDS, SECOND, MINUTE, HOUR, DAY, WEEK, MONTH, YEAR, DECADE, CENTURY, and 44 | SQL Pocket Guide MILLENNIUM. Pass one of these values as a text string; case does not matter. Other PostgreSQL Datetime Functions Use AT TIME ZONE either to apply a time zone to a datetime without one or to convert a datetime from one time zone to another. For example: SELECT CURRENT_TIMESTAMP; 2006-02-05 17:28:38.534286-05 SELECT CURRENT_TIMESTAMP AT TIME ZONE 'PST'; 2006-02-05 14:28:38.541632 You can achieve the same results as those in the previous example through the TIMEZONE function: SELECT TIMEZONE('PST', CURRENT_TIMESTAMP); PostgreSQL supports an Ingres-inspired function called DATE_PART that provides the same functionality as the ISO- standard EXTRACT. For example, to extract the current minute value as a number, specify: SELECT DATE_PART('minute', CURRENT_TIME); 36 DATE_PART accepts all of the same datetime element names as EXTRACT. See “EXTRACT Function” on page 51. Datetime Functions: SQL Server SQL Server 2008 introduces a set of high-precision functions to return current datetime information: SYSDATETIME() Returns date and time as a DATETIME2 value. Datetime Functions: SQL Server | 45 SYSDATETIMEOFFSET() Returns date, time, and time zone offset as a DATETIMEOFFSET value. SYSUTCDATETIME Returns current UTC time as a DATETIME2 value. SQL Server continues to support the following functions from previous releases: CURRENT_TIMESTAMP or GETDATE() Returns the current date and time on the server as a date- time value. GETUTCDATE() Returns the current UTC date and time, as derived from the server’s time and time zone setting. SQL Server implements two functions for date arithmetic: DATEADD(datepart, interval,date) Adds interval (expressed as an integer) to date. Specify a negative interval to perform subtraction. The datepart ar- gument is a keyword specifying the portion of the date to increment, and it may be any of the following: year, yy, yyyy, quarter, qq, q, month, mm, m, dayofyear, dy, y, day, dd, d, week, wk, ww, hour, hh, minute, mi, n, second, ss, s, millisecond, ms. For example, to add one day to the cur- rent date, use DATEADD(day, 1, GETDATE()). DATEDIFF(datepart , startdate , enddate) Returns enddate – startdate expressed in terms of the units you specify for the datepart argument. For example, to compute the number of minutes between the current time and UTC time, use DATEDIFF(mi, GETUTCDATE(), GETDATE()). SQL Server 2008 introduces new functions to work with time zone offsets: SWITCHOFFSET(datetimeoffset, new_offset) Inserts a new time zone offset into a DATETIMEOFFSET value and returns that new value. 46 | SQL Pocket Guide TODATETIMEOFFSET(datetime2, new_offset) Creates a DATETIMEOFFSET value from a DATETIME2 and an offset that you specify. Specify time zone offsets in string form. For example, to con- vert the current time into U.S. Eastern Standard Time: SELECT SWITCHOFFSET ( SYSDATETIMEOFFSET(), '-05:00'); Negative offsets count westward from the prime meridian; positive offsets count eastward. Deleting Data Use the DELETE statement to delete rows from a table: DELETE FROM data_source WHERE predicates For example, you may want to delete states for which you don’t know the population: DELETE FROM state s WHERE s.population IS NULL; SQL Server, MySQL, and PostgreSQL 8.1 and earlier do not allow the alias on the target table. See the section “Predi- cates” on page 109 for more details on the different kinds of predicates that you can write. Deleting in Order MySQL requires that you include an ORDER BY clause in your DELETE statement when deleting multiple rows from a table having a self-referential foreign-key constraint. This is to en- sure that child rows are deleted before their parents. Because MySQL checks for constraint violations during statement execution, this is a MySQL-only issue. Deleting Data | 47 NOTE The ISO SQL standard allows constraint checking to be done either at the end of each statement’s execution or at the end of a transaction, but never during statement execution. In the section “Subquery Inserts” on page 69, you will find an INSERT INTO…SELECT FROM statement that creates a new tour in the trip table called J's Tour. If you wish to delete J's Tour, you must issue a statement such as: DELETE FROM trip WHERE name = 'J''s Tour' ORDER BY CASE stop WHEN 1 THEN 1 WHEN 2 THEN 2 WHEN 6 THEN 3 WHEN 4 THEN 4 WHEN 3 THEN 5 WHEN 5 THEN 6 END DESC; The CASE expression in this statement’s ORDER BY clause hardcodes a child-first delete order. Obviously, this completely defeats the purpose of a multirow DELETE statement. If you’re lucky, you’ll have a sortable column that will yield a child-first delete order without its having to be hardcoded. In the case of this book’s example schema and data, I wasn’t so lucky. Deleting All Rows Omit the WHERE clause to remove all rows from a table: DELETE FROM township; Many database systems also implement a TRUNCATE TABLE statement that empties a table instantly, without logging, and thus with no hope of rolling back: TRUNCATE TABLE township; 48 | SQL Pocket Guide Oracle provides a form that preserves any space allocated to the table (which is useful if you plan to reload the table right away): TRUNCATE TABLE township REUSE STORAGE; Deleting from Views and Subqueries All platforms allow deletes from views, but with restrictions. Oracle and DB2 allow deletes from a subquery (also known as an inline view). For example, to delete any states not referenced by the gov_unit table, you can specify: DELETE FROM ( SELECT * FROM state s WHERE s.id NOT IN ( SELECT g.id FROM gov_unit g WHERE g.type = 'State')); In PostgreSQL, a view that is the target of a DELETE must have an associated ON DELETE DO INSTEAD rule. PostgreSQL does not allow deleting from subqueries. Various restrictions are placed on deletions from views and subqueries because, ultimately, a database system must be able to resolve a DELETE against a view or a subquery to a set of rows in an underlying table. Returning Deleted Data: DB2 DB2 provides a very powerful option for retrieving the rows affected by a DELETE statement. Simply SELECT from the DELETE statement. For example: SELECT * FROM OLD TABLE ( DELETE FROM state WHERE name = 'Michigan' ); Specify FROM OLD TABLE, and wrap your DELETE in parentheses. Deleting Data | 49 Returning Deleted Data: Oracle Oracle’s solution to returning just-deleted rows is a RETURN- ING clause to specify the data to be returned and where it will be placed: DELETE FROM ... WHERE ... RETURNING expression [,expression ...] [BULK COLLECT] INTO variable [,variable ...] For DELETEs of more than one row, the target variables must also be PL/SQL collection types, and you must use the BULK COLLECT keywords: DECLARE TYPE county_id_array IS ARRAY(100) OF NUMBER; county_ids county_id_array; BEGIN DELETE FROM county_copy RETURNING id BULK COLLECT INTO county_ids; END; / Rather than specifying a target variable for each source expression, your target can be a record containing the appro- priate number and type of fields. Returning Deleted Data: SQL Server SQL Server implements the OUTPUT clause for returning deleted rows from a query. For example: DELETE FROM state OUTPUT DELETED.id AS state_id, DELETED.name; You can use the syntax OUTPUT DELETED.* to return all columns. You can specify expressions such as UPPER(DELETED.name). You can specify column aliases as in any query, with or without the optional AS keyword. 50 | SQL Pocket Guide Double-FROM SQL Server supports an extension to DELETE that lets you delete from a table based on values from a joined table. For example, to delete counties from gov_unit for which you do not know the population, specify: DELETE FROM gov_unit FROM gov_unit g JOIN county c ON g.id = c.id WHERE c.population IS NULL; The first FROM clause identifies the ultimate target of the DELETE. The second FROM clause specifies a table join. Then predicates in the WHERE clause can evaluate columns from both tables in the join. In this example, rows are deleted from the gov_unit table based on a corresponding population from the county table. EXTRACT Function DB2 (9.7 and higher), MySQL, Oracle, and PostgreSQL support the standard EXTRACT function to retrieve specific elements from a datetime value. In MySQL, for example: SELECT EXTRACT(DAY FROM CURRENT_DATE); The result will be a number. Valid elements are SECOND, MINUTE, HOUR, DAY, MONTH, and YEAR. Oracle supports the following additional elements: TIMEZONE_HOUR, TIMEZONE_MINUTE, TIME- ZONE_REGION, and TIMEZONE_ABBR. The latter two Oracle elements are exceptions and return string values. PostgreSQL also supports additional elements: CENTURY, DECADE, DOW (day of week), DOY (day of year), EPOCH (number of seconds in an interval, or since 1-Jan-1970 for a date), MICROSECONDS, MILLENNIUM, MILLISEC- ONDS, QUARTER, TIMEZONE (offset from UTC, in seconds), TIMEZONE_HOUR (hour part of UTC offset), TIMEZONE_MINUTE (minute part of offset), and WEEK. EXTRACT Function | 51 GREATEST DB2 (9.5 onward), MySQL, Oracle, and PostgreSQL imple- ment the GREATEST function to return the largest value from a list of values: GREATEST(value [, value ...]) The input values may be numbers, datetimes, or strings. On some platforms, if even one input value is null, then the func- tion returns null. Grouping and Summarizing SQL enables you to collect rows into groups and to summarize those groups in various ways, ultimately returning just one row per group. You do this using the GROUP BY and HAVING clauses, as well as various aggregate functions. Aggregate Functions An aggregate function takes a group of values, one from each row in a group of rows, and returns one value as output. One of the most common aggregate functions is COUNT, which counts non-null values in a column. For example, to count the number of waterfalls associated with a county, specify: SELECT COUNT(u.county_id) AS county_count FROM upfall u; 16 Add DISTINCT to the preceding query to count the number of counties containing waterfalls: SELECT COUNT(DISTINCT u.county_id) AS county_count FROM upfall u; 6 52 | SQL Pocket Guide The ALL behavior is the default, counting all values: COUNT(expression) is equivalent to COUNT(ALL expression). COUNT is a special case of an aggregate function because you can pass the asterisk (*) to count rows rather than column values: SELECT COUNT(*) FROM upfall; Nullity is irrelevant when COUNT(*) is used because the concept of null applies only to columns, not to rows as a whole. All other aggregate functions ignore nulls. Table 6 lists some commonly available aggregate functions. However, most database vendors implement aggregate func- tions well beyond those shown. Table 6. Common aggregate functions Function Description AVG(x) Returns the mean. COUNT(x) Counts non-null values. MAX(x) Returns the greatest value. MEDIAN(x) Returns the median, or middle value, which may be interpolated. (Oracle only.) MIN(x) Returns the least value. STDDEV(x) Returns the standard deviation. Use STDEV (only one D) in SQL Server. SUM(x) Sums all numbers. VARIANCE(x) Returns the statistical variance. Use VAR in SQL Server. Not available in MySQL. GROUP BY Aggregate functions come into their own when you apply them to groups of rows rather than to all rows in a table. To do this, use the GROUP BY clause. The following query counts the number of waterfalls in each of the predefined tours: SELECT t.name AS tour_name, COUNT(*) FROM upfall u INNER JOIN trip t Grouping and Summarizing | 53 ON u.id = t.stop GROUP BY t.name; When you execute a query like this one, the result-set rows are grouped as specified by the GROUP BY clause: TOUR_NAME FALL_NAME ---------- --------------- M-28 Munising Falls M-28 Alger Falls M-28 Scott Falls M-28 Canyon Falls M-28 Agate Falls M-28 Bond Falls Munising Munising Falls Munising Tannery Falls Munising Alger Falls Munising Wagner Falls Munising Horseshoe Falls Munising Miners Falls US-2 Bond Falls US-2 Fumee Falls US-2 Kakabika Falls US-2 Rapid River Fls After the groups have been created, any aggregate functions are applied once to each group. In this example, COUNT(*) is eval- uated separately for each group: TOUR_NAME COUNT(*) ---------- --------------- M-28 6 M-28 M-28 M-28 M-28 M-28 Munising 6 Munising Munising Munising Munising Munising 54 | SQL Pocket Guide US-2 4 US-2 US-2 US-2 Any columns to which an aggregate function has not been ap- plied are now “collapsed” into one value: TOUR_NAME COUNT(*) ---------- --------------- M-28 6 Munising 6 US-2 4 In practical terms, this collapsing of many detail rows into one aggregate row means that you must apply an aggregate function to any column not listed in your GROUP BY clause. NOTE Grouping usually implies a limited sort operation to sort the rows into their groups. Listing the Detail Values Oracle implements the LISTAGG function to aggregate detail values for a column into a single value per group. The result is a delimited list of values. The following example extends the previous section’s query to return a column named stop with a comma-delimited list of falls on a given tour: SELECT t.name AS tour_name, LISTAGG (u.name, ',') WITHIN GROUP (ORDER BY u.name ASC) AS stop FROM upfall u INNER JOIN trip t ON u.id = t.stop GROUP BY t.name; TOUR_NAME STOP ---------- ------------------------------ M-28 Agate Falls,Alger Falls,Bond Falls,Canyon Falls,Munising Falls,Scott Falls Grouping and Summarizing | 55 Munising Alger Falls,Horseshoe Falls,Miners Falls,Munising Falls,Tannery Falls,Wagner Falls The parameters to LISTAGG specify the column to aggregate and the delimiter to use in creating the list. The WITHIN GROUP keywords are mandatory. The ORDER BY clause in parentheses is also mandatory, and in this case it sorts the list alphabetically. You can sort the list on any detail column. For example, you can sort the list of falls in the stop column by u.confirmed_date, by u.lat_lon, and so forth. Reducing the GROUP BY List Sometimes you want to list a column in the SELECT list of a GROUP BY query without having to list that same column in the GROUP BY clause. In the following query, a given county number implies a county name: SELECT c.id AS county_id, c.name AS county_name, COUNT(*) AS waterfall_count FROM upfall u INNER JOIN county c ON u.county_id = c.id GROUP BY c.id, c.name; Rather than grouping by the c.id and c.name columns, it might be more efficient to group by the c.id column only, which yields a much shorter sort key. The grouping sort will poten- tially run faster and use less scratch space on disk. One ap- proach to doing this is specified as follows: SELECT c.id AS county_id, MAX(c.name) AS county_name, COUNT(*) AS waterfall_count FROM upfall u INNER JOIN county c ON u.county_id = c.id GROUP BY c.id; This query drops c.name from the GROUP BY clause. To com- pensate for that, the query arbitrarily applies the MAX function to that same column in the SELECT list. Because all county 56 | SQL Pocket Guide names within a group of similar c.id values are the same, MAX can return only that one name. Grouping Before the Join The GROUP BY examples in the preceding section involve a join that is performed before the grouping operation. Using a subquery, it’s possible to restate the query in a way that causes the join to occur after the aggregation instead: SELECT c.id AS county_id, c.name AS county_name, agg.falls_count FROM county c INNER JOIN ( SELECT u.county_id, COUNT(*) AS falls_count FROM upfall u GROUP BY u.county_id) agg ON c.id = agg.county_id; The advantage here is that the join involves far fewer rows be- cause the aggregation occurs prior to the join, not after it. An- other advantage is a potential reduction in scratch disk and memory requirements, as the rows involved in the GROUP BY operation and subsequent aggregation do not include any data from the county table. HAVING The HAVING clause is used to place restrictions on the rows returned from a GROUP BY query. For example, to list only those tours having at least six stops, specify the following: SELECT t.name AS tour_name, COUNT(*) FROM upfall u INNER JOIN trip t ON u.id = t.stop GROUP BY t.name HAVING COUNT(*) >= 6; Never put a condition in the HAVING clause that does not involve an aggregation. Such conditions are evaluated much more efficiently in the WHERE clause. Grouping and Summarizing | 57 ROLLUP The ROLLUP operation supported in DB2, MySQL, Oracle, and SQL Server generates a summary row for each group. For example, to roll up tour stops by county in DB2, Oracle, or SQL Server 2008, specify the following: SELECT t.name AS tour_name, c.name AS county_name, COUNT(*) AS falls_count FROM upfall u INNER JOIN trip t ON u.id = t.stop INNER JOIN county c ON u.county_id = c.id GROUP BY ROLLUP(t.name, c.name); Use the WITH ROLLUP syntax in MySQL or SQL Server 2005: SELECT t.name AS tour_name, c.name AS county_name, COUNT(*) as falls_count FROM upfall u INNER JOIN trip t ON u.id = t.stop INNER JOIN county c ON u.county_id = c.id GROUP BY t.name, c.name WITH ROLLUP; Following is the output from the preceding queries. The rows in boldface are generated as a result of using ROLLUP: TOUR_NAME COUNTY_NAME FALLS_COUNT ---------- ----------- ---------------------- M-28 Alger 3 M-28 Baraga 1 M-28 Ontonagon 2 M−28 6 US-2 Delta 1 US-2 Gogebic 1 US-2 Dickinson 1 US-2 Ontonagon 1 US−2 4 Munising Alger 6 Munising 6 16 The GROUP BY operation generates the normal summary by tour and county. The ROLLUP operation adds in summaries 58 | SQL Pocket Guide for all other possible levels by tour name and for the entire set of rows. M-28’s six stops, for example, comprise three stops in Alger County, two in Ontonagon County, and one in Baraga County. There are 16 tour stops total across all tours. CUBE CUBE takes things a step further. It generates summaries for all possible combinations of the columns you specify, as well as a grand total. The following is the CUBE version of the pre- ceding section’s ROLLUP query for DB2, Oracle, and SQL Server 2008, but restricted to the Munising tour: SELECT t.name AS tour_name, c.name county_name, COUNT(*) AS falls_count FROM upfall u INNER JOIN trip t ON u.id = t.stop INNER JOIN county c ON u.county_id = c.id WHERE t.name = 'Munising' GROUP BY CUBE(t.name, c.name); MySQL 5.1 does not support CUBE. SQL Server 2005 requires the WITH CUBE syntax: SELECT t.name AS tour_name, c.name AS county_name, COUNT(*) AS falls_count FROM upfall u INNER JOIN trip t ON u.id = t.stop INNER JOIN county c ON u.county_id = c.id WHERE t.name = 'Munising' GROUP BY t.name, c.name WITH CUBE; The results in DB2, Oracle, and SQL Server are the same: TOUR_NAME COUNTY_NAME FALLS_COUNT ---------- ----------- ---------------------- 6 Alger 6 Munising 6 Munising Alger 6 Grouping and Summarizing | 59 These results are only for the Munising tour. CUBE generates far more rows than ROLLUP does. GROUPING SETS Oracle, DB2, and SQL Server implement the GROUPING SETS function to let you specify the groupings that you want. For example: SELECT t.name AS tour_name, c.name AS county_name, COUNT(*) AS falls_count FROM upfall u INNER JOIN trip t ON u.id = t.stop INNER JOIN county c ON u.county_id = c.id GROUP BY GROUPING SETS(t.name, c.name); TOUR_NAME COUNTY_NAME FALLS_COUNT ---------- ----------- ----------- Munising 6 M-28 6 US-2 4 Delta 1 Ontonagon 3 Gogebic 1 Baraga 1 Alger 9 Dickinson 1 This data is a subset of the results generated by CUBE, but if it’s all you need, then using GROUPING SETS is more convenient. Related Functions The following functions are helpful when using CUBE, ROLLUP, and GROUPING SETS: GROUPING(column) Returns 1 if a null column value was the result of a CUBE, ROLLUP, or GROUPING SETS operation; otherwise, it returns 0. (DB2, Oracle, SQL Server.) 60 | SQL Pocket Guide GROUPING_ID(column , column , ...) Behaves similarly to GROUPING, but this generates a bit vector of 1s and 0s, depending on whether the corre- sponding columns contain nulls generated by an extended GROUP BY feature. (Available only in Oracle9i Database and higher.) GROUP_ID() Enables you to distinguish between duplicate rows in the output from CUBE, ROLLUP, and GROUPING SETS. The function returns 0 through n−1 for each row in a set of n duplicates. You can use that return value to decide how many duplicates to retain. Use HAVING GROUP_ID()=0 to eliminate all duplicates. (Oracle only.) Following is an example of the GROUPING function, using SQL Server’s WITH CUBE syntax. The function returns a 1 whenever a null is the result of the CUBE operation. SELECT t.name AS tour_name, c.name AS county_name, COUNT(*) AS falls_count, GROUPING(t.name) AS n1, GROUPING(c.name) n2 FROM upfall u INNER JOIN trip t ON u.id = t.stop INNER JOIN county c ON u.county_id = c.id WHERE t.name = 'Munising' GROUP BY t.name, c.name WITH CUBE; tour_name county_name falls_count n1 n2 ---------- ----------- ----------- ---- ---- Munising Alger 6 0 0 Munising NULL 6 0 1 NULL NULL 6 1 1 NULL Alger 6 1 0 The n1 and n2 columns indicate when null tour and county names are the result of the CUBE operation rather than of data actually contained within the tables being queried. Grouping and Summarizing | 61 Hierarchical Queries DB2, Oracle, and SQL Server support the recursive use of WITH as defined in the ISO SQL standard for querying hier- archical and recursive data. PostgreSQL supports recursive WITH, but with a slight syntax difference. NOTE Oracle also supports a proprietary CONNECT BY syn- tax. See “CONNECT BY Queries” on page 8. Recursive WITH Following is an example recursive query that generates a hier- archical list of governmental units. States will be listed first, then counties, then townships. WITH recursiveGov (depth, id, parent_id, name, type) AS (SELECT 1, parent.id, parent.parent_id, parent.name, parent.type FROM gov_unit parent WHERE parent.parent_id IS NULL UNION ALL SELECT parent.depth+1, child.id, child.parent_id, child.name, child.type FROM recursiveGOV parent, gov_unit child WHERE child.parent_id = parent.id) SELECT depth, id, parent_id, name, type FROM recursiveGOV; PostgreSQL requires that you specify that the WITH clause is to be recursive by including the RECURSIVE keyword: WITH RECURSIVE recursiveGov ... Most of the preceding statement consists of a subquery named recursiveGOV that is specified using the WITH clause. The subquery consists of two SELECTs unioned together. Consider 62 | SQL Pocket Guide the first SELECT as the union query’s starting point. It includes a predicate to treat rows having null parent_ids as the tree roots. Consider the second SELECT as defining the recursive link between parent and child rows. The second SELECT brings in the children of the first. Because the second SELECT references the named subquery that it is part of (itself), it recursively brings back children of the rows it returned (and so forth until the end). The main SELECT kicks off all this recursion by simply selecting from the named subquery. NOTE For a more in-depth explanation of what happens when a recursive WITH executes, read the article “Under- standing the WITH Clause” at http://gennick.com/with .html. The output from the preceding query will look like this: DEPTH ID PARENT_ID NAME TYPE ----- -- --------- -------- -------- 1 3 Michigan state 2 2 3 Alger county 2 6 3 Baraga county ... Tracking Your Depth To keep track of your depth in a hierarchy, create a depth col- umn as shown in the example query. Have the first SELECT return 1 as the value for that column and have the second SELECT return parent.depth+1. Then the root node will be depth 1, the root’s immediate children will be depth 2, and so on, down to the bottom of the hierarchy. Hierarchical Queries | 63 Breadth-First Versus Depth-First Sorting Results are returned by default in the following breadth-first order, which differs from the order you’ll get using Oracle’s CONNECT BY syntax (described in “CONNECT BY Quer- ies” on page 8): 1. The root node 2. The root’s immediate children 3. The children of the root’s immediate children 4. And so forth Oracle lets you specify whether you prefer depth- or breadth- first sorting via a search clause. You can also specify how you want siblings ordered. Look at what follows the SEARCH key- word that precedes SELECT in the following example: WITH recursiveGov (depth, id, parent_id, name, type) AS (SELECT 1, parent.id, parent.parent_id, parent.name, parent.type FROM gov_unit parent WHERE parent.parent_id IS NULL UNION ALL SELECT parent.depth+1, child.id, child.parent_id, child.name, child.type FROM recursiveGOV parent, gov_unit child WHERE child.parent_id = parent.id) SEARCH DEPTH FIRST BY name ASC NULLS FIRST SET ordering_column SELECT depth, id, parent_id, name, type FROM recursiveGOV ORDER BY ordering_column; You can specify either SEARCH DEPTH FIRST or SEARCH BREADTH FIRST. The preceding query returns results in depth-first order: DEPTH ID PARENT_ID NAME TYPE ----- -- --------- -------- -------- 1 3 Michigan State 64 | SQL Pocket Guide 2 2 3 Alger County 3 5 2 Au Train Township 3 1 2 Munising City ... Key to these results is the SET ordering_column clause. That clause adds an extra column of output containing a value that you can sort on in the main query to ensure that rows are, in fact, returned in the specified breadth- or depth-first order. You can name the ordering column anything you wish. You can also include that column in your main query’s select list, in the WHERE clause, and anywhere else in the main query a column name is allowed. Detecting Recursive Loops Oracle also supports syntax to detect loops in recursive data. For instance, the example data for this book has been carefully crafted to include a loop in the trip table. If you look closely at the raw data, you’ll find that two stops share the same pa- rent, leading to a loop. The following query uses the CYCLE clause to detect that problem and the resulting loop: WITH recursiveTrip (name, stop, parent_stop) AS (SELECT parent.name, parent.stop, parent.parent_stop FROM trip parent WHERE parent.parent_stop IS NULL UNION ALL SELECT child.name, child.stop, child.parent_stop FROM recursiveTrip parent, trip child WHERE child.parent_stop = parent.stop) SEARCH DEPTH FIRST BY stop ASC NULLS FIRST SET ordering_column CYCLE stop SET cycle_flag TO 'y' DEFAULT 'n' SELECT name, stop, parent_stop, cycle_flag FROM recursiveTrip; A cycle in this case is any instance of a parent and child row sharing the same stop number. Hence the specification of Hierarchical Queries | 65 CYCLE stop in this query. The SET clause adds a new column to the query output, specifying that the column be set to y in the event that the row in question is the cause of a loop in the data. Here’s what the output looks like: NAME STOP PARENT_STOP CYCLE_FLAG ---------- ---------- ----------- ---------- Munising 1 n Munising 2 1 n Munising 6 2 n Munising 4 6 n Munising 3 4 n M-28 1 3 y Munising 5 3 n ... If you omit the CYCLE clause and a loop results, Oracle will throw an ORA-32044: cycle detected while executing recur sive WITH query error. When you specify both SEARCH and CYCLE clauses, you must specify them in that order. You can base loop-detection on a combination of multiple columns by specifying a comma-delimited list of columns following the CYCLE keyword. Indexes, Creating The basic CREATE INDEX statement syntax is: CREATE INDEX falls_name ON upfall (name, open_to_public); In this syntax, falls_name is the name of the index. The table to be indexed is upfall. The index is on the combined values of name and open_to_public. Oracle and PostgreSQL allow you to assign an index to a tablespace: CREATE INDEX falls_name ON upfall (name, open_to_public) TABLESPACE users; Oracle, PostgreSQL, and SQL Server also allow you to index column expressions: 66 | SQL Pocket Guide CREATE INDEX falls_name ON upfall (UPPER(name), open_to_public); This particular index is useful for resolving queries in which the WHERE clause predicates involve the expression UPPER(name). Indexes on expressions are subject to various restrictions. SQL Server requires that such expressions return “precise” results, thus ruling out expressions returning or involving floating- point data types. Such expressions must also be deterministic, meaning that a given input always returns the same output, no matter the server or environmental settings. Indexes, Removing In DB2, Oracle, and PostgreSQL, you remove an index by naming it in a DROP INDEX statement: DROP INDEX falls_name; MySQL and SQL Server require you to also specify the table name: DROP INDEX falls_name ON upfall; Inserting Data Use the INSERT statement to insert new rows in a table. You can insert one row, many rows (DB2 and MySQL), or the results of a subquery. Single-Row Inserts The following example adds a county to the gov_unit table. The values in the VALUES clause correspond to the columns listed after the table name: INSERT INTO gov_unit (id, parent_id, name, type) VALUES (13, 3, 'Chippewa', 'County'); Inserting Data | 67 Any columns you omit from an INSERT statement take on their default values specified at table-creation time. If you do not specify a default value at table-creation, then a null is used. Use the DEFAULT keyword to specify explicitly that a column should take on its default value. Use the null keyword to insert a null value explicitly in a column that might otherwise default to a non-null value. For example: INSERT INTO gov_unit (id, parent_id, name, type) VALUES (14, DEFAULT, 'Mackinac', NULL); If your VALUES list contains a value for each of the table’s columns in the order specified at table creation, you can omit the column list: INSERT INTO gov_unit VALUES (15, DEFAULT, 'Luce', 'County'); For anything other than an ad-hoc insert (in other words, for inserts you embed in your scripts and programs), it’s safer to specify a list of columns. Otherwise, such queries can fail the moment a new column is added to the target table. Multirow Inserts DB2, MySQL, and SQL Server 2008 provide the ability to insert multiple rows via repeated value lists in the VALUES clause: INSERT INTO gov_unit (id, parent_id, name, type) VALUES (16, 3, 'Menominee', 'County'), (17, 3, 'Iron', 'County'), (18, 3, 'Keweenaw', 'County'); Insert Targets All platforms allow inserts into a view. DB2 and Oracle also allow inserts into subqueries (or inline views): INSERT INTO (SELECT id, name, type FROM gov_unit) 68 | SQL Pocket Guide (id, name, type) VALUES (19, 'Keweenaw', 'County'); PostgreSQL requires views that are the targets of inserts to have an associated ON INSERT DO INSTEAD rule. Subquery Inserts Using a subquery to feed an INSERT statement, it’s possible to insert a number of rows at one time. For example, to create a duplicate of the Munising tour, but with a different name, specify: INSERT INTO trip (name, stop, parent_stop) (SELECT 'J''s Tour', stop, parent_stop FROM trip WHERE name = 'Munising'); The SELECT statement in this form of INSERT must return an expression corresponding to each column listed after the target table. Some platforms let you get away without the parentheses surrounding the subquery, but it’s safer to include them. The subquery can be any valid SELECT statement. It may return zero, one, or many rows. MySQL requires that you sort your source rowset in a way that avoids loading any child row ahead of its parent. For example: INSERT INTO trip (name, stop, parent_stop) (SELECT 'J''s Tour', stop, parent_stop FROM trip WHERE name = 'Munising' ORDER BY CASE stop WHEN 1 THEN 1 WHEN 2 THEN 2 WHEN 6 THEN 3 WHEN 4 THEN 4 WHEN 3 THEN 5 WHEN 5 THEN 6 END); The issue with MySQL is that constraints are not checked at the end of a statement as in all other platforms, but rather they are checked as a statement executes—in this case, for each row inserted. Also, in this particular case, there is no column in the Inserting Data | 69 trip table by which you can sort to prevent a constraint viola- tion. Thus, you are forced to enumerate each row in the ORDER BY clause, which virtually eliminates the benefit of using INSERT INTO…SELECT FROM. See the section “Deleting in Order” on page 47 for a DELETE that will remove rows for J's Trip in the reverse order from their insertion. Returning Inserted Values: DB2 DB2 allows you to query newly inserted data by simply select- ing from the INSERT statement. For example: SELECT * FROM NEW TABLE ( INSERT INTO gov_unit (id, name, type) VALUES (20, 'Limestone', 'Township') ); Specify FROM NEW TABLE, and wrap your INSERT in parentheses. Returning Inserted Values: Oracle Oracle supports a RETURNING clause to specify the data to be returned and where it will be placed. The following example works in Oracle SQL*Plus: VARIABLE pid VARCHAR2(10); INSERT INTO gov_unit (id, name, type) VALUES (19, 'Houghton', 'County') RETURNING parent_id INTO :pid; VARIABLE is an SQL*Plus command used to create a bind variable. This example returns one column. You can return more than one column by simply separating column names and result variables with commas: RETURNING col1, col2 ... INTO var1, var2 ... See “Returning Deleted Data: Oracle” on page 50 for an ex- ample of the RETURNING clause showing the use of array variables to return data from multiple rows. 70 | SQL Pocket Guide Returning Inserted Data: SQL Server You can use SQL Server’s OUTPUT clause to return values from newly inserted rows. For example: INSERT INTO gov_unit (id, name, type) OUTPUT INSERTED.parent_id AS pid VALUES (19, 'Houghton', 'County'); You can use the syntax OUTPUT INSERTED.* to return all columns. You can specify expressions such as UPPER(INSERTED.name). You can specify column aliases as in any query, with or without the optional AS keyword. Multitable Inserts Using Oracle, you can issue INSERTs that affect multiple ta- bles at once. You can insert the results of a subquery uncon- ditionally into several tables, or you can write predicates that control which rows are inserted into which table. If you choose to write predicates, you can choose whether evaluation stops with one success or whether a row is considered for insertion into more than one table. Unconditional multitable insert Use INSERT ALL to insert the results of a subquery in more than one target table: INSERT ALL INTO fall_description (id, name, description) VALUES (id, name, description) INTO fall_location (id, datum, zone, northing, easting) VALUES (id, datum, zone, northing, easting) SELECT id, name, description, datum, zone, northing, easting FROM upfall; This example inserts location and description data in two sep- arate tables. Different data is inserted in each table, but all data Inserting Data | 71 comes from the subquery. Each row returned by the subquery results in two new rows—one in each table. Conditional multitable insert Use WHEN clauses to insert conditionally in multiple tables. The following statement splits township and state data into two separate tables: INSERT FIRST WHEN type = 'Township' THEN INTO township (id, name) VALUES (id, name) WHEN type = 'State' THEN INTO state (id, name) VALUES (id, name) ELSE INTO other_unit (id, name) VALUES (id, name) SELECT * FROM gov_unit; The ELSE clause in this statement causes all rows that do not meet any other criteria to be added to the other_unit table. The ELSE clause is optional; you can omit it to ignore rows that do not meet at least one WHEN condition. ALL versus FIRST In a conditional multitable INSERT, the keyword ALL causes each row returned by the subquery to be evaluated against each WHEN clause. Thus, a row meeting criteria in two clauses can be inserted in more than one table. Use INSERT FIRST to stop evaluating a row after the first matching WHEN clause. Joining Tables Joins allow you to combine data from multiple tables into a single result-set row. There are two fundamental types of join: inner and outer. There are also two join syntaxes—the syntax introduced in the 1992 SQL standard, which depends on a JOIN clause, and an older syntax in which you separate table names with commas. 72 | SQL Pocket Guide The Concept of a Join The concept of a join is best explained by beginning with the earlier syntax. To join related rows from two tables, begin by listing two table expressions separated by a comma in your FROM clause. For example, to retrieve a list of waterfalls and their county names, you could begin by writing the following: SELECT u.name AS fall, c.name AS county FROM upfall u, county c; FALL COUNTY --------------- ---------- Munising Falls Alger Munising Falls Baraga Munising Falls Ontonagon ... This result is a Cartesian product, which is all possible com- binations of rows from the two tables. Conceptually, all joins begin as Cartesian products. From there, it’s up to you to sup- ply conditions to narrow down the results to only those rows that make sense. Using the older join syntax, you supply those conditions in the WHERE clause: SELECT u.name AS fall, c.name AS county FROM upfall u, county c WHERE u.county_id = c.id; FALL COUNTY --------------- ---------- Munising Falls Alger Tannery Falls Alger Alger Falls Alger ... These results are much more useful. The process I’ve just described is purely conceptual; database systems will rarely or never form a Cartesian product when executing a join. However, thinking in these conceptual terms will help you write correct join queries and understand their results. Regardless of how the join operation is optimized, join results must match the conceptual results in the end. Joining Tables | 73 Cross Joins The SQL standard uses the term cross join to describe a Car- tesian product. Generate a cross join as follows: SELECT * FROM upfall CROSS JOIN county; One case in which cross joins are useful is when joining to a single-row result set. For example, to generate a report of tablespaces that include database_name, an Oracle database administrator could specify the following: SELECT d.name database_name, t.name tablespace_name FROM v$tablespace T CROSS JOIN v$database d; DATABASE_NAME TABLESPACE_NAME ------------- ------------------------------ DB01 SYSTEM DB01 UNDOTBS1 ... This cross join has the effect of replicating the v$database in- formation to every row of the result set. Because there is only ever one row in v$database, the result set will still return one row per tablespace. NOTE DB2 began to support the CROSS JOIN syntax in release 9.5. Cross joins are also useful in conjunction with DB2’s LAT- ERAL clause: SELECT u.name, ings.direction, ings.meters FROM upfall u, LATERAL (VALUES ('Northing', u.northing), ('Easting', u.easting)) AS ings(direction, meters); This query results in two rows per waterfall, one with the northing value and the other with the easting. For example: 74 | SQL Pocket Guide NAME DIRECTION METERS --------------- --------- ----------- Munising Falls Northing 5141184 Munising Falls Easting 528971 Olson Falls Northing 5140000 Olson Falls Easting 528808 ... LATERAL generates a new table with values from the table to its left. The cross join expands each waterfall row into two rows. Inner Joins An inner join brings together corresponding rows from two tables. For example, you could list each waterfall in its corre- sponding county: SELECT u.name AS fall, c.name AS county FROM upfall u INNER JOIN county c ON u.county_id = c.id; FALL COUNTY --------------- ---------- Munising Falls Alger Tannery Falls Alger Alger Falls Alger ... The keywords INNER JOIN between the two tables specify that the join should be an inner join. The ON clause specifies the join condition, or the condition that must apply in order for two rows to be considered related. Conceptually, as described in the preceding section, a Cartesian product is formed and the join condition is then applied to filter out unwanted combina- tions of rows. The order of tables in an inner join is irrelevant. The INNER keyword is optional. A WHERE clause is still valid in join queries. For example, to report only on counties with a popu- lation above 10,000, use the following: SELECT u.name AS fall, c.name AS county FROM county c INNER JOIN upfall u Joining Tables | 75 ON u.county_id = c.id WHERE c.population > 10000; Conceptually, the join results are materialized first, and the WHERE clause then restricts the results to those joined rows that satisfy the WHERE conditions. In reality, your database will find a more optimal approach to producing the results. The USING Clause MySQL, Oracle, and PostgreSQL support the USING clause. When the columns defining a join between two tables are iden- tically named, and when the join condition would be an equal- ity condition requiring that each set of identically named columns contain the same value (an equi-join), you can write the join more simply by replacing the ON clause with the USING clause. Here’s an example: SELECT * FROM fall_description JOIN fall_location USING (id); There is a subtle issue to be aware of when using the USING clause. Consider the following query: SELECT fd.id, fl.id FROM fall_description fd JOIN fall_location fl USING (id); This version of the query will work in MySQL and PostgreSQL, but will fail in Oracle with the following error: ORA-25154: column part of USING clause cannot have qualifier In Oracle, the USING clause merges the two id columns, and the result will have only one column named id (not fd.id and not fl.id). That column is associated with neither table, so neither table alias applies: SELECT id FROM fall_description fd JOIN fall_location fl USING (id); 76 | SQL Pocket Guide MySQL and PostgreSQL merge the two id columns into one (thus conforming to the ISO SQL standard), but they still allow you to select both an fl.id value and an fd.id value. However, a SELECT * against the join in MySQL or PostgreSQL will yield only one id column in the result. Natural Joins There is yet another shortcut beyond the USING clause, and that is the NATURAL JOIN syntax supported by MySQL, Oracle, and PostgreSQL. If two tables should be joined based on all columns they have in common with the same name and the join is an equi-join, you can use the NATURAL JOIN key- words without specifying explicitly the join conditions. In Oracle, you cannot qualify a NATURAL JOIN column with an alias. For example, in the following query only one id col- umn is returned, and it is not associated with either table (and hence not with any fd or fl alias): SELECT id FROM fall_description fd NATURAL JOIN fall_location fl; MySQL and PostgreSQL, on the other hand, give you the op- tion to qualify join columns: SELECT fd.id, fl.id FROM fall_description fd NATURAL JOIN fall_location fl; Be wary of using NATURAL JOIN, especially in queries that you encapsulate within program code. The simple addition of a column to one table, if it has a name that happens to match a column in a joined table, can suddenly change the semantics of a NATURAL JOIN query. If you do use NATURAL JOIN, use it only for ad-hoc queries—and even then, be careful! Joining Tables | 77 Non-Equi-Joins So far, all the joins illustrated have been equi-joins, which in- volve corresponding columns from two tables that have the same values in two corresponding rows. Equi-joins are proba- bly the most common type of joins, but it is sometimes useful and even necessary to write join conditions that are not equality-based. Such joins are sometimes referred to as non- equi-joins. For example, the following statement creates a table of years via a subquery, and then it joins the table of months to upfall based on the value of upfall’s confirmed_date. This particular example runs on MySQL and Oracle (because they support date literals): SELECT u.name, y.year_num FROM upfall u JOIN (SELECT 2005 AS year_num, DATE '2005-1-1' AS year_begin, DATE '2005-12-31' AS year_end FROM dual UNION SELECT 2006 AS year_num, DATE '2006-1-1' AS year_begin, DATE '2006-12-31' AS year_end FROM dual) y ON u.confirmed_date BETWEEN y.year_begin AND y.year_end; The result is an association of waterfalls to the year in which their data was confirmed: NAME YEAR_NUM --------------- ---------------------- Munising Falls 2005 Tannery Falls 2005 Alger Falls 2005 ... Tahquamenon 2006 There are definitely easier ways to obtain this result, but this example does illustrate that not all joins need to be equi-joins. 78 | SQL Pocket Guide Outer Joins In an outer join, each row in the result set does not necessarily have to contain a row from both tables being joined; one or both tables are treated as optional. If you want a join to be done only when possible, and you want rows back regardless, use an outer join. Left outer joins Use a left outer join when you want all rows from one table, regardless of whether corresponding rows exist in the other table. Consider the possibility of a waterfall for which the owner is unknown: a row in upfall with a null owner_id. You want to list all waterfalls with their owners if possible, but even when no corresponding owner row exists, you still want to list all of the waterfalls. You can do that using a left outer join: SELECT u.name AS fall, o.name AS owner FROM upfall u LEFT OUTER JOIN owner o ON u.owner_id = o.id; FALL OWNER --------------- --------------- Miners Falls Pictured Rocks Munising Falls Pictured Rocks Tannery Falls Michigan Nature Tahquamenon Rapid River Fls Kakabika Falls ... A left outer join designates the leftmost table as the required table. In this case, the leftmost table is the upfall table. Thus, each row in the final result from the query must correspond to a row from the upfall table. The owner table is the optional table. If an owner row exists that corresponds to a row from upfall, the result is the same as that from an inner join: a row with values from both tables. If no owner row corresponds to a given upfall row, a row is returned with data from upfall, but with nulls in place of all the owner values. Joining Tables | 79 NOTE The USING and NATURAL clauses, as well as paren- theses, can be used with outer joins in the same way that they are used with inner joins. Interpreting nulls in an outer join When interpreting the results from the left outer join in the preceding section, you can’t really be certain that there is no owner listed for Kakabika Falls, for instance, because there could be an owner row but with a null name. A “safer” version of the query includes the primary key column from owner: SELECT u.name AS fall, o.name AS owner, o.id FROM upfall u LEFT OUTER JOIN owner o ON u.owner_id = o.id; The o.id column is a primary key, meaning it cannot ever le- gitimately be null in the owner table. Therefore, if you see a null o.id value in the result of this query, you can rest assured that it is a result of the left outer join, not a result of finding a null in the corresponding owner row. Another way to address the problem of interpreting an outer join is to always include the join columns in the result. If the join columns from the required table (e.g., u.owner_id) are not null, but the join columns from the optional table (e.g., o.id) are null, you have a case in which there is no corresponding row from the optional table (e.g., owner). Right outer joins A right outer join is the same as a left outer join except that the required table is the rightmost table—the second table to be listed. For example, the following two joins are semantically equivalent: SELECT u.name AS fall, o.name AS owner FROM upfall u LEFT OUTER JOIN owner o ON u.owner_id = o.id; 80 | SQL Pocket Guide SELECT u.name AS fall, o.name AS owner FROM owner o RIGHT OUTER JOIN upfall u ON u.owner_id = o.id; In either case, upfall is the required table and owner is the op- tional table. The results of the two queries are identical. To avoid confusion between left and right, some SQL program- mers write all such joins as LEFT OUTER JOINs. Full outer joins Sometimes you want an outer join in which both tables are optional. Such a join is a full outer join, and such joins are sup- ported by DB2, Oracle, PostgreSQL, and SQL Server. You can write one as follows: SELECT u.name AS fall, o.name AS owner FROM upfall u FULL OUTER JOIN owner o ON u.owner_id = o.id; FALL OWNER --------------- --------------- Little Miners Pictured Rocks Agate Falls Horseshoe Falls This query returns falls without recorded owners and “owners” who own no waterfalls (i.e., potential owners), all in addition to the standard inner join results of falls and their correspond- ing owners. Vendor-specific outer join syntax In the past, database vendors have developed different ways to write outer joins. In Oracle, you used to identify the optional table by adding the suffix (+) to the optional table’s column reference in all of the join conditions for the given join. For example: SELECT u.name fall, o.name owner FROM upfall u, owner o WHERE u.owner_id = o.id(+); Joining Tables | 81 Reversing the order of the columns in the predicate gives the same result. What matters is the location of the (+) operator: SELECT u.name fall, o.name owner FROM upfall u, owner o WHERE o.id(+) = u.owner_id; Older versions of SQL Server required the use of *= and =* in equality conditions to designate left and right outer joins, re- spectively. For example: SELECT u.name fall, o.name owner FROM upfall u, owner o WHERE o.id *= u.owner_id; Oracle and SQL Server still support these syntaxes (although in SQL Server, you must set your compatibility level to 80 or lower, using stored procedure sp_dbcmptlevel). However, your queries will be much easier to understand and debug if you write your joins using the JOIN clause. LEAST DB2 (9.5 onward), MySQL, Oracle, and PostgreSQL imple- ment the LEAST function to return the smallest value from a list of values: LEAST(value [, value ...]) The input values may be numbers, datetimes, or strings. Literals All database systems make provisions for embedding literal values in SQL statements. Text and numeric literals are usually quite simple, but there are some nuances of which you should be aware. Date and time literals tend to be more complex. 82 | SQL Pocket Guide Text Literals The ISO SQL standard for text literals is to enclose them in single quotes: 'This is a text literal' Use two adjacent single quotes when you need to embed a sin- gle quote in a string: 'Isn''t SQL fun?' SQL will treat the two adjacent single quotes as a single quote within the literal: Isn't SQL fun? Oracle Database 10g and higher allow you to specify alterna- tive quoting delimiters, which are always two characters and always include leading and trailing single quotes. Introduce delimiters by prefacing them with a Q or a q. For example, to use '[ and ]' as delimiters, specify: Q'[This isn't as bad as it looks]' q'[This isn't as bad as it looks]' The (, [, and { characters are special cases in that their corre- sponding closing delimiters must be ), ], and }, respectively. Otherwise, use the same character to close the string that you use to open it: Q'|This string is delimited by vertical bars|' You can’t use space, tab, or return characters to delimit a string in this manner. PostgreSQL allows you to specify alternative quoting delimit- ers using a dollar-sign syntax, producing a dollar-quoted string constant. For example: $tag$This is a dollar-quoted string constant$tag$ Replace tag with any desired sequence of characters. Your quoting delimiter is then $tag$. If you like, you can even use $ $ without any intervening tag text. Escape sequences (see Table 7) do not have any effect in dollar-quoted string Literals | 83 constants; they are treated as literal character sequences. $$\t$$ yields the string \t, not a tab character. MySQL allows you to include the escape sequences shown in Table 7 in string literals. PostgreSQL allows the escape sequences shown in Table 8. Table 7. MySQL string literal escape sequences Escape Description \0 NULL character (ASCII zero) \' Single quote \" Double quote \b Backspace \n Newline \r Carriage return \t Tab \z ASCII 26 or the Ctrl-Z character \\ Backslash \% Percent sign \_ Underscore Table 8. PostgreSQL string literal escape sequences Escape Description \b Backspace \f Form feed \n Newline \r Carriage return \t Tab \octal Character corresponding to the given octal value \xhexadecimal Character corresponding to the given hexadecimal value \\ Backslash 84 | SQL Pocket Guide Numeric Literals Numeric literals follow standard conventions for writing numbers: 123 123.45 +123 −123.45 Numbers written without a decimal point are generally treated as integers. Oracle allows for a trailing F, f, D, or d to indicate FLOAT or DOUBLE, respectively: 123D 123.45F +123d −123.45f You can also use scientific notation to write floating-point constants: 123.45E+23 123.45e−23 These literals are interpreted respectively as 123.45 × 1023 and 123.45 ÷ 1023. Datetime Literals The SQL standard defines the following formats for date, time, and timestamp literals, with hours specified according to a 24- hour clock: DATE 'yyyy-mm-dd' TIME 'hh:mi:ss [{+|-}hh:mi]' TIMESTAMP 'yyyy-mm-dd hh:mi:ss [{+|-}hh:mi]' For example, the following specifications refer to 19-Dec-2005, 8:00 PM, and 8:00 PM U.S. Eastern Standard Time on 19- Dec-2005: DATE '2005-12-19' TIME '20:00:00' TIMESTAMP '2005-12-19 20:00:00 −5:00' SQL Server does not support these literals. DB2 does not sup- port specifying the time zone. Literals | 85 Datetime Interval Literals SQL defines the following formats for INTERVAL YEAR TO MONTH literals: INTERVAL 'year-month' YEAR TO MONTH INTERVAL 'year' YEAR INTERVAL 'month' MONTH Oracle9i Database and higher and PostgreSQL 9.0 and higher support these formats and also allow you to specify a precision for the year, which otherwise defaults to two digits: INTERVAL '42-1' YEAR TO MONTH INTERVAL '1042' YEAR(4) Similarly, SQL defines the following formats for INTERVAL DAY TO SECOND literals: INTERVAL 'dd hh:mi:ss.ff' DAY TO SECOND INTERVAL 'hh:mi' HOUR TO MINUTE INTERVAL 'mi' MINUTE ... For an INTERVAL DAY TO SECOND literal, you can specify any contiguous range of time elements from days to seconds. In Oracle9i Database and higher, days (dd) and fractional sec- onds (ff) both default to two digits of precision. Merging Data DB2, Oracle, and SQL Server 2008 support the use of the MERGE statement for updating or inserting rows, depending on whether they already exist in the target table. For example, to merge potentially new waterfall data into the upfall table, specify the following: MERGE INTO upfall u USING (SELECT * FROM new_falls) nf ON (u.id = nf.id) WHEN MATCHED THEN UPDATE SET u.name = nf.name, u.open_to_public = nf.open_to_public 86 | SQL Pocket Guide WHEN NOT MATCHED THEN INSERT (id, name, datum, zone, northing, easting, lat_lon, county_id, open_to_public, owner_id, description, confirmed_date) VALUES (nf.id, nf.name, nf.datum, nf.zone, nf.northing, nf.easting, nf.lat_lon, nf.county_id, nf.open_to_public, nf.owner_id, nf.description, nf.confirmed_date); This statement updates only name and open_to_public for existing waterfalls, although you could choose to update all columns if you wanted to do so. For new falls, all columns are inserted into the upfall table. Oracle allows you to place WHERE conditions on both the UPDATE and INSERT operations. In addition, Oracle allows you to specify rows to be deleted following an UPDATE operation: MERGE INTO upfall u USING (SELECT * FROM new_falls) nf ON (u.id = nf.id) WHEN MATCHED THEN UPDATE SET u.name = nf.name, u.open_to_public = nf.open_to_public WHERE nf.name IS NOT NULL DELETE WHERE u.open_to_public = 'n' WHEN NOT MATCHED THEN INSERT (id, name, datum, zone, northing, easting, lat_lon, county_id, open_to_public, owner_id, description, confirmed_date) VALUES (nf.id, nf.name, nf.datum, nf.zone, nf.northing, nf.easting, nf.lat_lon, nf.county_id, nf.open_to_public, nf.owner_id, nf.description, nf.confirmed_date) WHERE nf.open_to_public = 'y'; This statement uses WHERE nf.name IS NOT NULL to prevent up- dating any name to a null. The subsequent DELETE WHERE clause then deletes any updated rows that no longer represent publicly accessible falls. Merging Data | 87 NOTE DELETE WHERE is a post-update deletion. Rows not updated by the MERGE statement are not candidates for deletion. In DB2, you can specify a DELETE as the statement for a WHEN MATCHED clause. DB2 also allows more than one occurrence of the WHEN MATCHED and WHEN NOT MATCHED clauses. Following is the DB2 version of the pre- ceding MERGE statement: MERGE INTO upfall u USING (SELECT * FROM new_falls) nf ON (u.id = nf.id) WHEN MATCHED AND nf.name IS NOT NULL THEN UPDATE SET u.name = nf.name, u.open_to_public = nf.open_to_public WHEN MATCHED AND u.open_to_public = 'n' THEN DELETE WHEN NOT MATCHED AND nf.open_to_public = 'y' THEN INSERT (id, name, datum, zone, northing, easting, lat_lon, county_id, open_to_public, owner_id, description, confirmed_date) VALUES (nf.id, nf.name, nf.datum, nf.zone, nf.northing, nf.easting, nf.lat_lon, nf.county_id, nf.open_to_public, nf.owner_id, nf.description, nf.confirmed_date); In DB2, a DELETE is something you can specify in place of an UPDATE. In Oracle, a DELETE can happen after an UPDATE. Nulls When writing SQL, it’s critical to understand nulls and three- valued logic. With few exceptions, the result of any expression involving a null will be either null or unknown, and this has ramifications for any expression (comparison or otherwise) that you write. 88 | SQL Pocket Guide Predicates for Nulls You should not compare a null to any other value using the standard comparison operators. For example, the following query will not return all rows from the upfall table: SELECT u.id, u.name, u.datum FROM upfall u WHERE u.datum = 'NAD1927' OR u.datum <> 'NAD1927'; You’d think that any given datum would either be NAD1927 or not be NAD1927, but this is not the case. A null datum is not NAD1927, nor is it not not NAD1927. SQL provides the IS NULL and IS NOT NULL predicates to detect the presence or absence of null. To find all datum values other than NAD1927, including those that are null, specify: SELECT u.id, u.name, u.datum FROM upfall u WHERE u.datum IS NULL OR u.datum <> 'NAD1927'; Similarly, you can use IS NOT NULL to match non-null values explicitly. Using CASE with Nulls CASE expressions can sometimes be helpful when working with potentially null data. For example, you can use CASE to ensure that you always get a non-null datum in your result set: SELECT u.id, u.name, CASE WHEN u.datum IS NULL THEN '*None!*' ELSE u.datum END FROM upfall u; Most databases also provide functions to do this type of thing more succinctly. Nulls | 89 Using the COALESCE Function All platforms support the standard SQL COALESCE function. It takes a series of values and returns the first non-null value encountered. For example, to return a list of waterfall descrip- tions that show name when a description is null and show Unknown! when even the name is null, specify: SELECT id, COALESCE(description, name, '*Unknown!*') FROM upfall; You can provide any number of arguments, but you should ensure that at least one will be non-null. If all arguments are null, COALESCE returns null as well. Functions for Nulls: DB2 DB2 supports CASE and COALESCE. It also supports the NULLIF function, which returns null whenever the two input values are the same. Notice the effect of NULLIF in the first row of these results: SELECT u.name, NULLIF(u.name,'Miners Falls') FROM upfall u WHERE u.name LIKE '%Miners%'; NAME 2 --------------- --------------- Miners Falls Little Miners Little Miners DB2 also now supports the same DECODE and NVL functions as Oracle does. See the section “Functions for Nulls: Ora- cle” on page 91. Functions for Nulls: MySQL MySQL supports CASE and COALESCE. In addition, it sup- ports a function called IFNULL to return an alternate value for a potentially null input value. For example: SELECT id, name, IFNULL(datum, '*None!*') FROM upfall; 90 | SQL Pocket Guide As with DB2, MySQL supports NULLIF to return null when- ever two input values are the same: SELECT u.name, NULLIF(u.name,'Miners Falls') FROM upfall u WHERE u.name LIKE '%Miners%'; You can also use the IF function to return one of two values, depending on whether an expression is TRUE: SELECT id, name, IF(datum IS NULL, '*None!*', datum) FROM upfall; You’d normally use a comparison expression to generate the Boolean TRUE/FALSE value for the first argument. If the ex- pression evaluates to TRUE, the value from the second argu- ment is returned. If the expression evaluates to FALSE or null, the third argument’s value is returned. Functions for Nulls: Oracle Oracle supports CASE and COALESCE. It also provides sev- eral other functions for dealing with nulls. NVL is similar to COALESCE. It is supported for backward compatibility and allows only two arguments: SELECT id, name, NVL(datum, '*None!*') FROM upfall; NVL2 returns one of two values, depending on whether the first is null: SELECT id, name, NVL2(datum, datum, '*None!*') FROM upfall; DECODE is equivalent to an inline IF statement (although you should really use CASE today), and it provides yet another way of dealing with nulls. The following example uses DECODE to replace some county_id values from upfall with their respective names: SELECT id, name, DECODE(county_id, 2, 'Alger County', Nulls | 91 6, 'Baraga County', NULL, 'Unknown', 'Other') FROM upfall; In this example, counties 2 and 6 translate to Alger and Baraga Counties, respectively. Any null county_id results in a value of 'Unknown'. Any other gov_unit_ids are denoted as 'Other'. DECODE supports any number of input/result pairs. Functions for Nulls: PostgreSQL PostgreSQL supports CASE and COALESCE. It also supports NULLIF as described in the section “Functions for Nulls: DB2” on page 90. Functions for Nulls: SQL Server SQL Server supports CASE, COALESCE, and an ISNULL function: ISNULL(possible_null, alternative_value) When possible_null is null, ISNULL will return alternative_value. Otherwise, ISNULL will return possible_null. SQL Server 2008 supports the same NULLIF function that DB2 supports, returning NULL when both input values match. SQL Server also supports a setting known as ANSI_NULLS, which affects the behavior of the = and <> predicates that compare to null: ... WHERE city_name = NULL ... WHERE city_name <> NULL By default, neither of these predicates will ever match any rows. However, if you issue the command SET ANSI_NULLS OFF, you can use = NULL and <> NULL to search for NULL or NOT NULL values, respectively. 92 | SQL Pocket Guide Numeric Conversions: DB2 Use the following functions to convert between different nu- meric types or between numeric and text types: BIGINT(numeric) BIGINT(character) CHAR(integer) CHAR(decimal [,decimal_character]) CHAR(floating [,decimal_character]) DECFLOAT(numeric, 16or34) DECFLOAT(character, 16or34 [, decimal_character]) DECIMAL(numeric [,precision[,scale]]) DECIMAL(character [,precision[,scale [,decimal_character]]]) DOUBLE(numeric) DOUBLE(character) DOUBLE_PRECISION(numeric) FLOAT(numeric) REAL(numeric) SMALLINT(numeric) SMALLINT(character) See “Datetime Conversions: DB2” on page 21 for information on converting between dates and numbers. In the syntax, numeric can be any numeric type or expression; character can be any fixed- or variable-length character type or expression; integer can be any integer type or expression; and decimal can be any decimal type or expression. NOTE DB2 also provides compatibility with TO_CHAR and TO_NUMBER, as implemented by Oracle. See the sec- tion “Numeric Conversions: Oracle” on page 95 for details. Numeric Conversions: DB2 | 93 Each function converts its argument to the type indicated by the function name. The following example shows DECFLOAT being used to convert from a character string: SELECT DECFLOAT('100.123451234512345',16) FROM dual; 100.1234512345123 And following is an example showing CHAR and DECIMAL being used to convert back and forth between numbers and strings: SELECT CHAR(100.12345), CHAR(DECIMAL('100.12345',5,2)) FROM pivot WHERE x=1; 100.12345 100.12 DECIMAL’s default scale is zero when converting from a char- acter string. To preserve digits to the right of the decimal point in that situation, you must specify a scale, which forces you to first specify a precision. No rounding occurs. To round a value being converted, you must first specify a precision and scale sufficient to hold the raw value, and then apply the ROUND function: SELECT DECIMAL('10.999',4,2), DECIMAL('10.999',4), ROUND(DECIMAL('10.999',5,3),2) FROM pivot WHERE x=1; 10.99 10. 11.000 You can use the optional decimal_character parameter to specify the character to use for the decimal point: SELECT DECFLOAT('10/95',16,'/'), CHAR(10.95,'/') FROM pivot WHERE x=1; 10.95 10/95 When converting to an integer type, any decimal portion is truncated. 94 | SQL Pocket Guide Numeric Conversions: MySQL MySQL implements the following numeric conversion functions: FORMAT(number , scale) Provides general-purpose numeric conversions to text. The scale is the number of decimal places that you wish to appear in the result. CONV(number , from_base , to_base) Converts from one base to another. The number may be either an integer or a string, and the base may range from 2 through 36. BIN(number) Returns the binary representation of a base-10 number. OCT(number) Returns the octal representation of a base-10 number. HEX(number) Returns the hexadecimal representation of a base-10 number. For example: SELECT CONV('AF',16,10), HEX(175), FORMAT(123456.789,2); 175 AF 123,456.79 Use CAST to convert a string to a number. Numeric Conversions: Oracle Use the following functions in Oracle to convert to and from the supported numeric types: TO_NUMBER(string, format) TO_BINARY_DOUBLE(string, format) TO_BINARY_FLOAT(string, format) TO_CHAR(number, format) TO_BINARY_DOUBLE(number) TO_BINARY_FLOAT(number) TO_NUMBER(number) Numeric Conversions: Oracle | 95 Use TO_NUMBER and TO_CHAR (the only two functions available prior to Oracle Database 10g) to convert between NUMBER and VARCHAR2 (Table 9 lists the available numeric format elements): SELECT TO_CHAR(1234.56,'C9G999D99') to_char, TO_NUMBER('1,234.56','9G999D99') from_char, TO_CHAR(123,'999V99') v_example FROM dual; TO_CHAR FROM_CHAR V_EXAMPLE ---------------- ---------------------- --------- USD1,234.56 1234.56 12300 Use TO_BINARY_FLOAT and TO_BINARY_DOUBLE to convert to the new 32- and 64-bit IEEE 754 floating-point types added in Oracle Database 10g. Also use these functions to convert values from one numeric type to another. Table 9. Oracle’s numeric format elements Element Description $ Prefix: dollar sign ($). , (comma) Location of comma. Consider G instead. . (period) Location of period. Consider D instead. 0 Significant digit. Leading zeros. 9 Significant digit. Leading blanks. B Prefix: returns zero as blanks. C Location of ISO currency symbol. D Location of decimal point. EEEE Suffix: use scientific notation. FM Prefix: removes leading/trailing blanks. G Location of group separator. L Location of local currency symbol. MI Suffix: trailing minus (–) sign. PR Suffix: angle brackets (< and >) around negative values. RN or rn Roman numerals, upper- or lowercase. Output-only. 96 | SQL Pocket Guide Element Description S Prefix: leading plus (+) or minus (–) sign. TM, TM9, TME Prefix: use minimum number of characters (text-minimum). Output- only. TM9 gives decimal notation. TME gives scientific notation. U Specifies location of Euro symbol (€). V Multiplies the number to the left of the V in the format model by 10 raised to the nth power, where n is the number of 9s found after the V in the format model. See the example earlier in this section. Output-only. X Use hexadecimal notation. Output-only. Precede with 0s for leading zeros. Precede with FM to trim leading/trailing spaces. Numeric Conversions: PostgreSQL You can convert between numeric values and their string rep- resentations using the following functions, where number can be any numeric type: TO_CHAR(number, format) TO_NUMBER(string, format) PostgreSQL’s number format elements closely follow Oracle’s. They are listed in Table 10. Table 10. PostgreSQL’s numeric format elements Element Description $ Prefix: dollar sign ($). , (comma) Location of comma. Consider G instead. . (period) Location of period. Consider D instead. 0 Significant digit. Leading zeros. 9 Significant digit. Leading blanks. B Prefix: returns zero as blanks. C Location of ISO currency symbol. D Location of decimal point. EEEE Suffix: use scientific notation. Numeric Conversions: PostgreSQL | 97 Element Description FM Prefix: removes leading/trailing blanks. G Location of group separator. L Location of local currency symbol. MI Suffix: trailing minus (–) sign. PR Suffix: angle brackets (< and >) around negative values. RN or rn Roman numerals, upper- or lowercase. Output-only. S Prefix: leading plus (+) or minus (–) sign. TM, TM9, TME Prefix: use minimum number of characters (text-minimum). Output- only. TM9 gives decimal notation. TME gives scientific notation. U Specifies location of Euro symbol (€). V Multiplies the number to the left of the V in the format model by 10 raised to the nth power, where n is the number of 9s found after the V in the format model. See the example in “Numeric Conversions: Ora- cle” on page 95. Output-only. X Use hexadecimal notation. Output-only. Precede with 0s for leading zeros. Precede with FM to trim leading/trailing spaces. Numeric Conversions: SQL Server Use the CONVERT function for conversions to and from nu- meric values: CONVERT(datatype[(length)], expression[, style]) Table 11 lists styles for converting FLOAT and REAL values to character strings. Table 12 lists styles for converting MONEY and SMALLMONEY values to character strings. Table 11. SQL Server floating-point styles Style Description 0 Default, 0–6 digits, scientific notation when necessary 1 Eight digits + scientific notation 2 16 digits + scientific notation 98 | SQL Pocket Guide Table 12. SQL Server money styles Style Description 0 Money default, no commas, two decimal digits 1 Commas every three digits, two decimal digits 2 No commas, four decimal digits The following two examples demonstrate numeric conversions using the CONVERT function. The second example combines conversion from text with a monetary conversion: SELECT CONVERT(VARCHAR(10), 1.234567, 2); 1.234567 SELECT CONVERT( VARCHAR, CONVERT(MONEY, '20999.95'), 1); 20,999.95 Numeric/Math Functions Following are some useful numeric and math functions that are fairly universal across database platforms: ABS(number) Returns the absolute value of number. CEIL(number) or CEILING(number) Returns the smallest integer that is greater than or equal to the number that you pass. Use CEILING for SQL Server and CEIL for other platforms. Remember that with neg- ative numbers, the greater value has the lower absolute value: CEIL(5.5) is 6, whereas CEIL(−5.5) is −5. EXP(number) Returns the mathematical constant e (≈2.71828183)— also known as Euler’s constant—raised to the power of number. Numeric/Math Functions | 99 FLOOR(number) Returns the largest integer that is less than or equal to the number you pass. Remember that with negative numbers, the lesser value has the higher absolute value: FLOOR(5.5) is 5, whereas FLOOR(−5.5) is −6. LN(number) Returns the natural logarithm of number. Supported in DB2, Oracle, and PostgreSQL. For other platforms, use LOG instead. LOG(number) Returns the natural logarithm of number (in DB2, SQL Server, and MySQL). In PostgreSQL, it returns the base-10 logarithm of number. LOG(base , number) Returns the logarithm of number in a base that you specify (Oracle and PostgreSQL). LOG10(number) Returns the base-10 logarithm of number (DB2, MySQL, and SQL Server). MOD(top , bottom) Returns the remainder of top divided by bottom (DB2, MySQL, Oracle, and PostgreSQL). NANVL(value , alternate) Returns an alternate value for any floating-point NaN (Not-a-Number) value. If value is NaN, then alternate is returned; otherwise, value is returned (Oracle). REMAINDER(top , bottom) Returns the remainder of top divided by bottom, the same as MOD (Oracle). ROUND(number [, places]) Rounds number to a specified number of decimal places. The default is to round to an integer value. Use a negative value for places to round to the left of the decimal point. SQL Server requires the places argument. 100 | SQL Pocket Guide ROUND(number, places [, option]) SQL Server’s version of ROUND. Use option to specify whether rounding or truncating is performed (see TRUNC below). If option is 0, the function rounds; oth- erwise, the function truncates. SIGN(number) Indicates the sign of a number. SIGN returns −1, 0, or 1, depending on whether number is negative, zero, or positive. TRUNC(number [, precision ]) Truncates number to a specific number of decimal places. The default precision is zero decimal places. Use a nega- tive precision to truncate to the left of the decimal point, forcing those digits to zero. SQL Server implements trun- cation using a special form of ROUND. MySQL imple- ments truncation using TRUNCATE(number, precision), requiring that you specify precision. OLAP Functions Online analytical processing (OLAP) function is the term DB2 uses for what the SQL standard refers to as a window func- tion. See “Window Functions” on page 173 for more on this extremely useful class of functions. Pivoting and Unpivoting Oracle Database (11g Release 1 onward) and SQL Server (2005 onward) both support pivot and unpivot operators. With the PIVOT operation, you can present data in a grid format by turning rows into columns while aggregating some value of interest. The UNPIVOT operation turns columns into rows, allowing you to take multiple columns containing the same type of data and present that data as one column for reporting or analysis. Pivoting and Unpivoting | 101 Pivoting: The Concept Use the pivot operation to create a lookup table presenting values termed measures at the intersection of other values, termed dimensions. For example, say that you wish to report on the number of falls open or closed to the public by county. You might begin with the following query: SELECT county_id, COALESCE(open_to_public, 'n') AS open_to_public, COUNT(id) FROM upfall GROUP BY county_id, open_to_public; And you would get row-by-row results like these: COUNTY_ID O COUNT(ID) ---------- - ---------- *** n 1 11 y 1 2 y 11 6 y 1 7 y 2 10 y 1 9 y 1 Looking carefully at the output, you can see that county #7 has two falls open to the public. The information is all there, but the presentation is cumbersome and not at all compact. A more useful presentation might be the following grid, which allows you to scan down to find the county, and then over to find the number of open and closed falls within that county: COUNTY_ID Open Closed ---------- ---------- ---------- *** 0 1 6 1 0 11 1 0 2 11 0 7 2 0 9 1 0 10 1 0 102 | SQL Pocket Guide You can use the PIVOT operator in both Oracle and SQL Server to generate these results. Pivoting: Oracle You can use the following query to generate results in the tab- ular format at the end of the preceding section. Essentially, the query converts the open/closed counts for each county from two rows into two columns: SELECT * FROM (SELECT id, county_id, COALESCE(open_to_public, 'n') AS open_to_public FROM upfall) PIVOT ( COUNT(id) FOR open_to_public IN ( 'y' AS "Open", 'n' AS "Closed" ) ); Here’s how the query works: 1. The subquery in the FROM clause generates the three values used to create the grid: the county ID for the vertical dimension; the open/closed flag for the horizontal dimen- sion; and the waterfall ID to use as the measure by which to generate values for each combination of vertical and horizontal dimensions. 2. The COALESCE function call in the subquery ensures that the open_to_public flag is never null. (You may or may not need or want such behavior in your own queries.) 3. The query generates one row per county, because county_id is the only column not listed in the PIVOT clause. Pivoting and Unpivoting | 103 4. The query generates one column for each listed value of open_to_public. Each column is given the name specified in the FOR clause: either Open or Closed in this case. 5. All the intersecting points in the grid are filled in by exe- cuting the expression COUNT(id) for each combination of county_id and open_to_public values. Oracle further allows you to pivot on multiple columns. Fol- lowing is an example query that presents the very same information, but this time as one long row: SELECT * FROM (SELECT id, county_id, COALESCE(open_to_public, 'n') AS open_to_public FROM upfall) PIVOT ( COUNT(id) FOR (county_id, open_to_public) IN ( (2, 'y') AS "Alger Open", (2, 'n') AS "Alger Closed", (6, 'y') AS "Baraga Open", (6, 'n') AS "Baraga Closed", (7, 'y') AS "Ontonagon Open", (7, 'n') AS "Ontonagon Closed", (9, 'y') AS "Dickinson Open", (9, 'n') AS "Dickinson Closed", (10, 'y') AS "Gogebic Open", (10, 'n') AS "Gogebic Closed", (11, 'y') AS "Delta Open", (11, 'n') AS "Delta Closed", (NULL, 'y') AS "Unknown Open", (NULL, 'n') AS "Unknown Closed" ) ); Notice that the FOR clause specifies two column names. Fur- ther notice that each entry in the IN list specifies a combination of those same two values. The result will be the following single row: 104 | SQL Pocket Guide Alger Open Alger Closed Baraga Open ... ---------- ------------ ----------- 11 0 1 The reason this second query returns only a single row is that the PIVOT clause consumes all of the columns returned by the subquery. Thus, no column(s) remain to serve as a vertical di- mension. The result is a table having only a horizontal dimen- sion and having one measure for each dimension value. Pivoting: SQL Server SQL Server supports a pivot operator as well, though with a different syntax from Oracle’s implementation. Following is the query to generate the same tabular format as shown at the end of “Pivoting: The Concept” on page 102: SELECT county_id, [1] as 'open', [0] as 'closed' FROM (SELECT id, county_id, CASE open_to_public WHEN 'y' THEN 1 ELSE 0 END AS open_to_public FROM upfall) AS SourceTable PIVOT ( COUNT(id) FOR open_to_public IN ([1], [0]) ) AS PivotTable Here’s how this query operates: 1. The subquery in the FROM clause generates the three values to create the grid. You are required to specify an alias, which in this case is done using the AS clause. 2. The CASE statement in the subquery translates the y and n values into numeric ones and zeros. This is because SQL Server is currently unable to pivot on character columns. 3. The outer query lists county_id as the first column, be- cause that column is the unpivoted column. One row is Pivoting and Unpivoting | 105 ultimately returned for each unpivoted value: in this case for each county_id. 4. The second and third columns listed in the outer SELECT specify the column headings to use for the pivoted data. 5. The FOR…IN clause specifies that the first column is a count of open waterfalls (1=open to public) and that the second column is a count of closed waterfalls. 6. The COUNT(id) expression generates the summary values—in this case a count—for each of the cells. 7. All the intersecting points in the grid are filled in by exe- cuting the expression COUNT(id) for each combination of county_id and open_to_public values. Unlike Oracle, SQL Server does not currently allow pivoting on two columns. Unpivoting: The Concept Sometimes you’ll find yourself working with a table having two or more columns containing the same type of information. For example, you might have two or more phone numbers per row. Or, as in our case, you might choose to treat northing and easting values as two occurrences of a distance, which in fact they are. Following is a simple query to show the northing and easting values as they are represented in the database table: SELECT id, northing, easting FROM upfall WHERE northing IS NOT NULL OR easting IS NOT NULL; The output presents northing and easting each in its own column: ID NORTHING EASTING ---------- ---------- ---------- 1 5141184 528971 2 5140000 528808 3 5137795 527046 106 | SQL Pocket Guide And here is the same data presented with the distance values all in one column: ID LABEL VALUE ---------- -------- ---------- 1 NORTHING 5141184 1 EASTING 528971 2 NORTHING 5140000 2 EASTING 528808 3 NORTHING 5137795 ... This second presentation is the unpivoted form. Unpivoting: Oracle You can generate the unpivoted form shown at the end of the preceding section using the following query: SELECT id, label, value FROM upfall UNPIVOT EXCLUDE NULLS ( value FOR label IN (northing, easting) ); Here is an explanation of the query. It begins from the inside and works outward. 1. The FOR clause specifies that values from the northing and easting columns are to be unpivoted by being turned into rows. 2. The identifier label (following FOR) specifies the name of a new column that Oracle Database creates to identify each unpivoted value in the query results. The SELECT clause lists this column, which receives the name of the original column containing each unpivoted value. 3. The identifier value specified following the first parenthe- sis in the UNPIVOT clause specifies the name for the new column created to hold the unpivoted values. The SELECT clause lists this column. Pivoting and Unpivoting | 107 4. The EXCLUDE NULLS clause throws out any nulls that would otherwise appear in the value column. That is the default behavior. Specify INCLUDE NULLS to retain such null values. 5. The SELECT clause lists the id column, causing the query to return one combination of label and value for each waterfall (i.e., for each distinct id value). Unpivoting: SQL Server Following is the SQL Server UNPIVOT syntax to generate the results shown at the end of the section “Unpivoting: The Con- cept” on page 106: SELECT id, label, value FROM upfall UNPIVOT ( value FOR label IN (northing, easting) ) UnpivotTable Notice the alias name UnpivotTable at the end of the subquery. SQL Server requires an alias for such a subquery. Here is the step-by-step explanation of this unpivot query: 1. The FOR clause specifies that values from the northing and easting columns are to be unpivoted by being turned into rows. 2. The identifier label (following FOR) specifies the name of a new column to identify each unpivoted value in the query results. The SELECT clause lists this column, which receives the name of the original column containing each unpivoted value. 3. The identifier value following the first parenthesis in the UNPIVOT clause specifies the name for the new column created to hold the unpivoted values. The SELECT clause lists this column. 4. SQL Server throws out any nulls that would otherwise appear in the value column. 108 | SQL Pocket Guide 5. The SELECT clause lists the id column, causing the query to return one combination of label and value for each wa- terfall (i.e., for each distinct id value). Predicates Predicates are conditions you write in the WHERE, ON, and HAVING clauses of an SQL statement that determine which rows are affected, or returned, by that statement. For example, use the predicate name = 'Wagner Falls' to return data for only that particular waterfall: SELECT u.zone, u.northing, u.easting FROM upfall u WHERE name = 'Wagner Falls'; Table 13 lists the available comparison operators. Some oper- ators, such as IN and EXISTS, are more fully described in later subsections. Regular-expression operators are described under “Regular Expressions” on page 113. Operators for dealing with nulls are described under “Nulls” on page 88. Table 13. Comparison operators Operator Description !=, <> Tests for inequality < Tests for less than <= Tests for less than or equal to <=> Null-safe test for equality; supported only by MySQL = Tests for equality > Tests for greater than >= Tests for greater than or equal to BETWEEN Tests whether a value lies within a given range EXISTS Tests whether rows exist matching conditions that you specify IN Tests whether a value is contained in a set of values that you specify or that are returned by a subquery IS [NOT] NULL Tests for nullity Predicates | 109 Operator Description LIKE Tests whether a value matches a pattern REGEXP, RLIKE Regular-expression comparison operator; supported only by MySQL REGEXP_LIKE Tests whether a value matches the pattern described by a regular expression; supported only by Oracle EXISTS Predicates Use EXISTS and NOT EXISTS to test for the existence of rows matching a set of conditions that you specify. For example, to return a list of all owners associated with at least one waterfall, specify: SELECT o.id, o.name FROM owner o WHERE EXISTS (SELECT * FROM upfall u WHERE u.owner_id = o.id); Replace EXISTS with NOT EXISTS to find all owners who are not associated with any waterfall. Subqueries used in EXISTS predicates should usually be correlated, which means that a subquery’s WHERE clause compares a column from the subquery with a column from the outer query. IN Predicates Use IN to test whether a value falls within a set of values. You can enumerate that set as a list of literal values, or you can return the set as the result of a subquery. The following exam- ple specifies a set of literal values: SELECT o.id, o.name FROM owner o WHERE o.id IN (1,2,3,4); This next example uses a subquery and restates the EXISTS query from the preceding section, which returns a list of owners associated with at least one waterfall: 110 | SQL Pocket Guide SELECT o.id, o.name FROM owner o WHERE o.id IN (SELECT u.owner_id FROM upfall u); Watch out for nulls! If the subquery you use with a NOT IN predicate returns a null value for even one row in the set, the result of the NOT IN operation will never be true. Rather, it will always be unknown, and your query won’t function as you expect. BETWEEN Predicates Use BETWEEN to see whether a value falls in a given range. For example: SELECT c.name FROM county c WHERE c.population BETWEEN 5000 AND 10000; Any BETWEEN predicate can easily be expressed using the >= and <= operators: SELECT c.name FROM county c WHERE c.population >= 5000 AND c.population <= 10000; When writing BETWEEN predicates, always list the smallest value first. LIKE Predicates The LIKE and NOT LIKE predicates give you rudimentary pattern-matching capabilities. You can use the percent (%) and underscore ( _ ) characters to match any number of characters or any one character, respectively. For example, to find all wa- terfalls containing the word “Miners” in their names, specify: SELECT u.id, u.name FROM upfall u WHERE u.name LIKE '%Miners%'; Predicates | 111 Use NOT LIKE to find all falls without “Miners” in their names. MySQL and PostgreSQL recognize the backslash (\) as an es- cape character by default. Use the escape character to specify pattern-matching characters literally. For example, to find all falls without a percent in their names: SELECT u.id, u.name FROM upfall u WHERE u.name NOT LIKE '%\%%'; You can also use the ESCAPE clause to specify explicitly the escape character. The following example will run in DB2, Ora- cle, and SQL Server: SELECT u.id, u.name FROM upfall u WHERE u.name NOT LIKE '%\%%' ESCAPE '\'; When specifying an escape character in MySQL or Post- greSQL, be aware that the backslash is also the string-literal escape character. Thus, to specify explicitly the backslash as the LIKE escape character, you must escape that backslash in the ESCAPE clause: SELECT u.id, u.name FROM upfall u WHERE u.name NOT LIKE '%\%%' ESCAPE '\\'; Oracle also implements LIKEC, LIKE2, and LIKE4, which work with Unicode characters, code units, and code points, respectively. Recursive Queries See “Hierarchical Queries” on page 62. Also see “CONNECT BY Queries” on page 8 if you are using a release of Oracle Database prior to Oracle Database 11g Release 2. 112 | SQL Pocket Guide Regular Expressions MySQL, Oracle, PostgreSQL, and SQL Server support regular expressions. SQL Server and MySQL support them only for string comparison, whereas PostgreSQL adds support for a regular-expression substring function and Oracle provides support for that and much more. Regular Expressions: MySQL In MySQL, you can perform regular-expression pattern match- ing using the REGEXP predicate in a manner similar to LIKE: string REGEXP pattern REGEXP looks for the specified regular expression anywhere in the target string. For example, to search for variant spellings of Fumee Falls: SELECT u.id, u.name FROM upfall u WHERE u.name REGEXP '(Fumee|Fumie|Fumy)'; MySQL’s regular-expression pattern matching is case- insensitive for nonbinary strings. Because MySQL recognizes the backslash (\) as an escape character in string literals, you must use a double backslash (\\) to represent a single backslash in any pattern that you write as a literal. Table 14 lists the regular-expression operators recognized by MySQL. Table 14. MySQL regular-expression operators Operator Description . Matches any character, including newlines. ^ Matches beginning of string. $ Matches end of string. [ . . . ] Matches any of a set of characters. [^ . . . ] Matches any character not in a set. [[. xx .]] Matches a collation element. Regular Expressions | 113 Operator Description [: class :] Specifies a character class within a bracket expression. For ex- ample, use [[:digit:]] to match all digits. Valid character classes are: [:alnum:], [:alpha:], [:blank:], [:cntrl:], [:digit:], [:graph:], [:lower:], [:print:], [:punct:], [:space:], [:upper:], [:xdigit:]. [= chars =] Specifies an equivalence class. [. charname .] Use within a bracket expression to match a character by name. For example, use [[.tilde.]] or [˜] to match the tilde (˜). You’ll find a list of character names in regexp/cname.h. * Matches zero or more. + Matches one or more. ? Matches zero or one. { x }, { x , y }, { x ,} Matches x times, from x to y times, or at least x times. | Delimits alternatives. ( . . . ) Defines a subexpression. [[:<:]] Matches the beginning of a word. [[:>:]] Matches the end of a word. Regular Expressions: Oracle Oracle Database 10g implements the following regular- expression functions: REGEXP_COUNT(source_string, pattern [, position [, match_parameter]]) REGEXP_INSTR(source_string, pattern [, position [, occurrence [, return_option [, match_parameter [, subexpression]]]]]) REGEXP_LIKE (source_string, pattern [, match_parameter]) 114 | SQL Pocket Guide REGEXP_REPLACE(source_string, pattern [, replace_string [, position [, occurrence [, match_parameter]]]]) REGEXP_SUBSTR(source_string, pattern [, position [, occurrence [, match_parameter]]]) Parameters are as follows: source_string The string you wish to search. pattern A regular expression describing the text pattern you are searching for. This expression cannot exceed 512 bytes in length. replace_string The replacement text. Each occurrence of pattern in source_string is replaced by replace_string, which can use backreferences to refer to values that match subex- pressions in the pattern. position The character position at which to begin the search. This defaults to 1 and must be positive. occurrence The occurrence of pattern you are interested in finding. This defaults to 1. Specify 2 if you want to find the second occurrence of the pattern, 3 for the third occurrence, and so forth. return_option Specify 0 (the default) to return the pattern’s beginning character position. Specify 1 to return the ending charac- ter position. match_parameter A set of options, in the form of a character string, that changes the default manner in which regular-expression pattern matching is performed. You may specify any, all, or none of the following options, in any order: Regular Expressions | 115 'i' Specifies case-insensitive matching. 'c' Specifies case-sensitive matching. 'n' Allows the period (.) to match the newline character. (Normally, that is not the case.) 'm' Causes the caret (^) and dollar sign ($) to match the beginning and ending, respectively, of lines within the source string. Normally, the caret and dollar sign match only the very beginning and the very end of the source string, regardless of any newline characters within the string. 'x' Ignores whitespace, preventing whitespace charac- ters from matching themselves. subexpression Specify 0 to return the position at which the entire pattern matches (INSTR) or to return the substring matching the entire pattern (SUBSTR). Specify 1 through 9 to return the position corresponding to that subexpression of the pat- tern (INSTR) or to return the string corresponding to that subexpression (SUBSTR). Defaults to 0. The NLS_SORT parameter setting determines whether case- sensitive or case-insensitive matching is done by default. NOTE For detailed information and examples of Oracle’s regular-expression support, see the Oracle Regular Expressions Pocket Reference by Jonathan Gennick and Peter Linsley (O’Reilly). Table 15 lists the regular-expression operators supported by these functions. 116 | SQL Pocket Guide Table 15. Oracle regular-expression operators Operator Description \ Escapes a metacharacter \1 . . . \9 Backreferences an earlier subexpression; the replace_string parameter supports from \1 to \500 . Matches any character ^ Matches beginning of line $ Matches end of line [ . . . ] Matches any of a set of characters [^ . . . ] Matches any character not in a set [.xx.] Encloses a collation element [:class:] Specifies a character class such as [:digit:], [:alpha:], or [:upper:] within a bracket expression [=chars=] Specifies an equivalence class * Matches zero or more + Matches one or more ? Matches zero or one {x}, {x,y}, {x,} Matches x times, from x to y times, or at least x times | Delimits alternatives ( . . . ) Defines a subexpression Regular Expressions | 117 Table 16 lists additional Perl-influenced operators added in Oracle Database 10g Release 2. Table 16. Perl-influenced regular-expression operators in Oracle Operator Description \d Matches any digit \D Matches any nondigit \w Matches a word character, which is defined to include alphabetic characters, numeric characters, and the underscore \W Matches any nonword character \s Matches any whitespace character \S Matches any nonwhitespace character \A Anchors an expression to the beginning of a string \Z Anchors an expression to the end of a string *? Nongreedy “zero or more” quantifier +? Nongreedy “one or more” quantifier ?? Nongreedy “zero or one” quantifier {x}?, {x,y}?, {x,}? Nongreedy versions of {x}, {x, y}, {x,} Regular Expressions: PostgreSQL PostgreSQL implements regular expressions in two ways. First, it provides support in the form of the SQL standard’s SIMILAR TO predicate. For example, to find variant spellings of Fumee Falls, specify: SELECT u.id, u.name FROM upfall u WHERE u.name SIMILAR TO '(Fumee|Fumie|Fumy) Falls'; Table 17 lists the regular-expression operators that you can use with SIMILAR TO. Use a backslash (\) to embed any of the operators as a literal character. Use the ESCAPE clause to spec- ify an alternate escape character: 118 | SQL Pocket Guide WHERE u.name SIMILAR TO '(Fumee|Fumie|Fumy) Falls' ESCAPE '@' Table 17. PostgreSQL regular-expression operators Operator Description _ Matches any single character % Matches any string of characters ( . . . ) Defines a subexpression | Denotes alternation * Matches zero or more + Matches one or more ? Matches zero or one {x}, {x,y}, {x,} Matches x times, from x to y times, and at least x times [ . . . ] Matches any of a set of characters [^ . . . ] Matches any character not in a set The following form of the SUBSTRING function supports the operators in Table 17: SUBSTRING(string FROM pattern FOR escape) For example: SELECT u.name FROM upfall u WHERE SUBSTRING(u.name FROM '(Fumee|Fumie|Fumy) Falls' FOR '\\') IS NOT NULL; As with the other queries in this section, this query searches for alternate spellings of Fumee Falls. Second, PostgreSQL implements Posix-style regular expres- sions. For example, to find waterfalls that are described by Michigan state highway names in the form M-28, M-1, and so forth, up to three digits, you can write: SELECT u.name, u.description FROM upfall u WHERE u.description ~ '.*M-[[:digit:]]{1,3}'; Regular Expressions | 119 The ˜ operator returns TRUE when the text on the left matches the expression on the right. The match is case-sensitive. Use ˜* for a case-insensitive match. Similarly, you can use !˜ and !˜* to return TRUE when the text to the left does not match the pattern. The following two functions provide additional support for Posix-style regular expressions: SUBSTRING(string FROM pattern) REGEXP_REPLACE(source, pattern, replacement [,flags ]) For example, to change waterfall names from “Fumee Falls” to “Falls, Fumee,” specify: SELECT REGEXP_REPLACE( u.name, '(.+?) (Falls)', '\\2, \\1') FROM upfall u; The flags argument to REGEXP_REPLACE is optional. Spec- ify 'i' for a case-insensitive match, 'g' to replace all matching substrings, or both (as in 'ig' or 'gi'). Flags must be lowercase. Use backreferences \1 through \9 in the replacement string to insert subexpressions from the matched text (denoted by ( . . . )). Use \& in the replacement string to insert the entire matched text. Use \\ to place a single backslash in the replacement string. \1 through \9 are always backreferences. When multiple digits are involved, the construct is assumed to be a backreference if it is within the valid range of currently existing subexpressions. If the construct is outside the valid range, it is treated as an octal character escape. However, if the first digit is a zero, the construct is always treated as an octal character escape, re- gardless of where it falls in relation to the range. Table 18 lists the Posix-style regular-expression operators available in PostgreSQL. 120 | SQL Pocket Guide Table 18. PostgreSQL Posix-style regular-expression operators Operator Description \ Escapes a metacharacter \1 . . . \9 . . . Backreferences an earlier subexpression . Matches any character ^ Matches beginning of line $ Matches end of line [ . . . ] Matches any of a set of characters [^ . . . ] Matches any character not in a set [: class :] Specifies a character class within a bracket expression; valid classes are: [:alnum:], [:alpha:], [:blank:], [:cntrl;], [:digit:], [:graph:], [:lower:], [:print:], [:punct:], [:space:], [:upper:], [:xdigit:] [. xx .] Encloses a collation element within a bracket expression [= chars =] Specifies an equivalence class within a bracket expression [[:<:]] Matches beginning of word [[:>:]] Matches ending of word * Matches zero or more + Matches one or more ? Matches zero or one {x}, {x,y}, {x,} Matches x times, from x to y times, or at least x times | Delimits alternatives ( . . . ) Defines a subexpression (?: . . . ) Defines a noncapturing subexpression (?= . . . ) Anchors to the beginning of a subexpression match (?! . . . ) Anchors to the point of a subexpression mismatch \a Matches the alert bell \b Matches backspace Regular Expressions | 121 Operator Description \B Matches a backslash (\); synonym for \\ \cX Matches a character in which the low-order five bits are the same as in the character X \d Matches any digit \D Matches any nondigit \e Matches the escape character \f Matches the form feed \m Anchors to the beginning of a word \M Anchors to the end of a word \n Matches newline \r Matches carriage return \t Matches horizontal tab \uXXXX Matches the UTF-16 codepoint specified by the four-digit hexadecimal number XXXX \Uxxxxxxxx Reserved for an eventual UTF-32 extension \v Matches vertical tab \w Matches a word character, which is defined to include alphabetic characters, numeric characters, and the underscore \W Matches any nonword character \xHEX_DIGITS Matches the character at code point HEX_DIGITS \0 Matches the null character (hex 0) \xx Matches the character at octal code point xx when xx is not a backreference \xxx Same as \xx, but for three octal digits \y Anchors to either the beginning or ending of a word \Y Anchors to a point that is not the beginning or ending of a word \s Matches any whitespace character \S Matches any nonwhitespace character \A Anchors an expression to the beginning of a string \Z Anchors an expression to the end of a string 122 | SQL Pocket Guide Operator Description *? Nongreedy “zero or more” quantifier +? Nongreedy “one or more” quantifier ?? Nongreedy “zero or one” quantifier {x}?, {x,y}?, {x,}? Nongreedy versions of {x}, {x, y}, {x,} Table 19 lists embedded regular-expression option letters that you can use to control overall matching behavior. To embed options in an expression, use the syntax (?xxx . . . ), where each x is an option letter from the table. Specify as many xs as you need. Table 19. PostgreSQL Posix-style option letters Option Description a Makes the rest of the expression a basic regular expression (BRE) c Specifies case-sensitive matching e Makes the rest of the expression an extended regular expression (ERE) i Specifies case-insensitive matching m Same as n n Specifies newline-sensitive matching p Specifies partial newline-sensitive matching q Makes the rest of the regular expression a literal (no more operators are recognized) s Specifies non-newline-sensitive matching (the default) t Specifies tight syntax w Specifies inverse partial newline-sensitive matching (a.k.a. weird matching) x Switches to expanded syntax Regular Expressions | 123 Regular Expressions: SQL Server SQL Server supports a very limited regular-expression syntax for its version of the LIKE predicate. For example, to find Fumee Falls even if it is misspelled as “Fumie Falls,” write: SELECT * FROM upfall WHERE name LIKE 'Fum[ie]e Falls'; SQL Server does not support quantifiers, alternation, subex- pressions, or backreferences. Table 20 lists the few operators that SQL Server does support. Table 20. SQL Server regular-expression operators Operator Description % Matches any number of characters _ Matches any character, including newlines [ . . . ] Matches any of a set of characters [^ . . . ] Matches any character not in a set Selecting Data Use a SELECT statement, or query, to retrieve data from a database—typically from a table or view or from a combination of tables and views: SELECT expression_list FROM data_source WHERE predicates GROUP BY expression_list HAVING predicates ORDER BY expression_list DB2, Oracle, PostgreSQL, and SQL Server support factoring out subqueries using a WITH clause. See “Hierarchical Quer- ies” on page 62 and “Subqueries” on page 139 for some ex- amples of this technique. 124 | SQL Pocket Guide The SELECT Clause Each expression in the SELECT clause becomes a column in the result set returned by the query. Expressions may be simple column names, may generate a new value using a column value as input, or may have nothing to do with any columns at all. Listing the columns to retrieve The SELECT clause specifies the individual data elements you want the statement to return. The simple case is to specify a comma-delimited list of one or more column names from the tables listed in the FROM clause: SELECT id, name FROM owner; The result set for this query will contain the following columns: ID NAME ------------ --------------- 1 Pictured Rocks 2 Michigan Nature 3 AF LLC 4 MI DNR 5 Horseshoe Falls Taking shortcuts with the asterisk To return all columns from a table, you can specify a single asterisk rather than write out each column name: SELECT * FROM owner; ID NAME PHONE TYPE ------------ --------------- ------------ ------- 1 Pictured Rocks 906.387.2607 public 2 Michigan Nature 517.655.5655 private 3 AF LLC private 4 MI DNR 906-228-6561 public 5 Horseshoe Falls 906.387.2635 private The asterisk is a helpful shortcut when executing queries in- teractively because it can save you a fair bit of typing. However, it’s a risky proposition to use the asterisk in program code Selecting Data | 125 because the columns in a table may change over time, causing your program to fail when more or fewer columns than expec- ted are returned. Writing expressions You can use column names in expressions. The following statement predicts the effect of a 10 percent drop in population (rounded to zero decimal places): SELECT name, ROUND(population * 0.90, 0) FROM county; It is not necessary for an expression in a SELECT list to refer to any column at all in the table or view from which you are selecting. In Oracle, it’s very common to issue queries against a special table known as dual, as in the following query, which returns the current date and time: SELECT SYSDATE FROM dual; In DB2, you can query sysibm.sysdummy1: SELECT CURRENT_DATE FROM sysibm.sysdummy1; Your database system will evaluate such expressions for each row returned by the query. Oracle’s dual table is special in that it holds only one row. Thus, the preceding query from dual will return only one value. In SQL Server and MySQL, you can return the result of an expression without selecting from a table at all. For example, use the following to get the current time (SQL Server): SELECT getdate(); A SELECT such as this one, in which no table is specified, is the SQL Server/MySQL equivalent of Oracle’s SELECT...FROM dual. Specifying result-set column names SQL enables you to specify a name, or alias, for each expression in your SELECT list. To specify a column alias, place the alias 126 | SQL Pocket Guide name immediately after the column name or expression, sep- arating the two by at least one space: SELECT id, name, ROUND(population * 0.90, 0) est_pop FROM county; ID NAME EST_POP ---------- ---------- ------------------- 2 Alger 8876 6 Baraga 7871 7 Ontonagon 7036 ... Alternatively, you can introduce a column alias using the AS keyword: SELECT id, name, ROUND(population * 0.90, 0) AS est_pop FROM county; PostgreSQL 8.1 and earlier require the use of AS to introduce a column alias. In a given situation, it may not be important to provide an alias for a simple column name such as id. However, it’s very im- portant to use aliases when working with expressions to give sensible names to the resulting columns. Dealing with case and punctuation in names By default, SQL is case-insensitive and converts keywords and identifiers (such as table and column names) to uppercase. In MySQL, case sensitivity depends on whether the underlying operating system is case-sensitive (with respect to filenames). Oddly, in PostgreSQL, the default is to convert to lowercase. If you must specify an identifier in a case-sensitive manner, you can enclose it in double quotes. The following example uses double quotes to generate mixed-case column aliases. Note that the double quotes also allow for spaces to be included in the alias names: SELECT id AS "Fall #", name AS "Fall Name" FROM upfall; Selecting Data | 127 Fall # Fall Name ------ --------------- 1 Munising Falls 2 Tannery Falls 3 Alger Falls ... The ability to quote identifiers also enables you to work with column and table names containing mixed cases, spaces, and other unusual characters. Using subqueries in a SELECT list Current versions of all of the platforms allow you to embed a subquery in a SELECT list. Ensure that the embedded sub- query is scalar: it must return zero or one rows and one column. When no row is returned, you get a null. You should also spec- ify a column alias so that the corresponding result-set column has a simple name to which you can easily refer in your code. For example, the following query returns the number of wa- terfalls for each owner: SELECT o.id, o.name, (SELECT COUNT(*) FROM upfall u WHERE u.owner_id = o.id) AS fall_count FROM owner o; Subqueries can be correlated or uncorrelated. The subquery in this example is correlated, meaning that it refers to the enclos- ing table. Qualifying column names You can qualify a column name by its table name. This is es- pecially important when writing queries that involve multiple tables, because sometimes two tables will have columns with the same name. To qualify a column name, use dot notation, as in table_name.column_name. For example: SELECT owner.id, owner.name FROM owner; 128 | SQL Pocket Guide If you qualify a column name by its table name, you can also qualify that table name by its schema or database name: SELECT sqlpocket.owner.id FROM sqlpocket.owner; To make it easier to qualify column names, you can provide table aliases. The following example gives the alias o to the table owner: SELECT o.id, o.name FROM owner o; or: SELECT o.id FROM sqlpocket.owner o; Qualifying column names is often necessary to remove ambi- guity in a query. ALL and DISTINCT Use the ALL and DISTINCT keywords to specify whether you want the SELECT operation to eliminate duplicate rows from the result set. Duplicate elimination typically involves a partial sorting of the data, though other approaches are possible and the approach taken depends upon the implementation. Following are two examples showing the difference between ALL and DISTINCT: SELECT ALL o.type, u.open_to_public FROM owner o JOIN upfall u ON o.id = u.owner_id; TYPE OPEN_TO_PUBLIC ------- -------------- public y private y private y public y public y public y Selecting Data | 129 SELECT DISTINCT o.type, u.open_to_public FROM owner o JOIN upfall u ON o.id = u.owner_id; TYPE OPEN_TO_PUBLIC ------- -------------- private y public y The first query simply returns one row for each owner. (The ALL keyword is optional and is assumed by default.) The sec- ond uses DISTINCT to return a list of different type/ open_to_public combinations. Use DISTINCT when you need each combination of column values to be returned only one time. The FROM Clause Use the FROM clause to specify the source of the data you want to retrieve. The simplest case is to specify a single table or view in the FROM clause of a SELECT statement: SELECT name FROM upfall WHERE id = 2; You can also qualify a table or view name with either a schema or database name, depending on your platform. Use dot nota- tion for that purpose: SELECT name FROM sqlpocket.upfall WHERE id = 2; This query retrieves specifically from the upfall table or view owned by the user sqlpocket. Table aliases in the FROM clause You can specify a name, or table alias, for any table or view expression in a FROM clause. Aliases are useful for queries having ambiguous column names resulting from a join or the use of a subquery. For example, the following query returns a 130 | SQL Pocket Guide list of waterfalls, and for each fall, it shows the number of other falls in the same county: SELECT u.name, (SELECT COUNT(*) FROM upfall u2 WHERE u2.county_id = u.county_id) FROM upfall u; You couldn’t write this query without using aliases because the table names are identical. You also lose the ability to reference the outer query from the inner query. For example, without aliases, this query’s WHERE clause would be: WHERE county_id = county_id Using aliases is the only way to differentiate between the two references to the upfall table. Subqueries in the FROM clause Subqueries can sometimes be used to good effect in the FROM clause, where they are also known as inline views. Such sub- queries must be noncorrelated; in other words, they must not reference columns from the main query. For example, the fol- lowing query lists all publicly owned falls: SELECT u.name AS fall_name, o.name AS owner_name FROM (SELECT * FROM owner WHERE type = 'public') o JOIN upfall u ON o.id = u.owner_id; The subquery conceptually materializes a temporary table of falls that are publicly owned. That temporary table is then joined to the upfall table. Generating tables through the VALUES clause DB2 and SQL Server 2008 allow the use of the VALUES clause to generate tables on the fly: SELECT id, name FROM (VALUES (1, 'Munising Falls'), (2, 'Tannery Falls')) AS falls(id, name); Selecting Data | 131 ID NAME ----------- -------------- 1 Munising Falls 2 Tannery Falls Be sure to place parentheses around the entire VALUES clause and to separate value lists using commas. The WHERE Clause Use the WHERE clause to restrict query results to only those rows of interest. Rarely will you want all rows from a table. More often, you’ll want rows that match specific criteria. The following example retrieves only those waterfalls located in Alger County that are publicly accessible: SELECT u.name FROM upfall u WHERE u.open_to_public = 'y' AND u.county_id IN ( SELECT c.id FROM county c WHERE c.name = 'Alger'); The query uses an equality predicate (=) to identify publicly accessible waterfalls and an IN predicate (IN) to identify falls in Alger County. See the section “Predicates” on page 109 for more examples and a list of predicates that you can use in the WHERE clause. NOTE Join conditions are also used to restrict data returned by a query. See “Joining Tables” on page 72. The GROUP BY Clause See the section “Grouping and Summarizing” on page 52. 132 | SQL Pocket Guide The HAVING Clause See the section “Grouping and Summarizing” on page 52. The ORDER BY Clause Use ORDER BY to specify how you want results to be sorted. For example, the following returns a list of waterfalls sorted by owner name, and then sorted within each owner by fall name: SELECT COALESCE(o.name, 'Unknown') AS owner, u.name AS fall FROM upfall u LEFT OUTER JOIN owner o ON u.owner_id = o.id ORDER BY o.name, u.name; The default sort is an ascending sort. You can use the keywords ASCENDING and DESCENDING (which you can abbreviate ASC and DESC) to control the sort on each column. The fol- lowing is a modification of the previous sort, but this time, it sorts owner names in reverse order: SELECT COALESCE(o.name, 'Unknown') AS owner, u.name AS fall FROM upfall u LEFT OUTER JOIN owner o ON u.owner_id = o.id ORDER BY o.name DESC, u.name ASC; You can sort by columns and expressions that are not in your SELECT list: ORDER BY o.id DESC, u.id You can also sort by numeric column position: SELECT COALESCE(o.name, 'Unknown') AS owner, u.name AS fall ... ORDER BY 1 DESC, 2 ASC; Selecting Data | 133 And, in Oracle, PostgreSQL, and SQL Server you can even sort by the results of a correlated subquery (i.e., one that references a column from the main query): SELECT COALESCE(o.name, 'Unknown') AS owner, u.name AS fall FROM upfall u LEFT OUTER JOIN owner o ON u.owner_id = o.id ORDER BY (SELECT COUNT(*) FROM upfall u2 WHERE u2.owner_id = o.id) DESC, u.name; The subquery returns the number of waterfalls owned by each owner, so the result of this query is to first list those falls whose owners own the greatest number of falls. A second sort is then performed on falls’ names. String Functions The following sections show how to use functions to perform common string operations. Searching a String In DB2 and Oracle, use the following version of INSTR to find the location of a substring within a string: INSTR(string, substring[, position[, occurrence]]) You can specify a starting position for the search, and you can request that a specific occurrence be found. If position is neg- ative, the search begins from the end of the string. NOTE In Oracle Database 10g and higher, you can also use REGEXP_INSTR, as described in the section “Regular Expressions” on page 113. 134 | SQL Pocket Guide Oracle implements INSTR, INSTRB, INSTR2, and INSTR4, which work in terms of the input character set, bytes, Unicode code units, and Unicode code points, respectively. DB2 imple- ments INSTR (and also INSTRB in version 9.7). DB2 also supports the LOCATE and POSSTR functions: LOCATE(substring, string[, position]) POSSTR(substring, string) Both functions return the first occurrence of substring within string. Zero is returned if no match is found. The default is to search string beginning from character position 1. In SQL Server, use the CHARINDEX function: CHARINDEX(substring, string[, position]) The arguments are the same as they are for DB2’s LOCATE. In MySQL, use either INSTR or LOCATE: INSTR(string, substring) LOCATE(substring, string[, position]) Use position to specify a starting character position other than 1. Zero is returned if substring is not found within string. In PostgreSQL, use either POSITION or STRPOS: POSITION(substring IN string) STRPOS(string, substring) Notice that the two functions use opposite argument orders. Replacing Text in a String Use the REPLACE function to perform a search-and-replace operation on a string: REPLACE(string, search, replace) You can delete occurrences of search by specifying an empty string ('') as the replace text. Also, Oracle allows you to omit the replacement string, which has the same effect as specifying the empty string (''). String Functions | 135 NOTE Oracle Database 10g and higher and PostgreSQL sup- port regular-expression search and replace through the REGEXP_REPLACE function described in the section “Regular Expressions” on page 113. Extracting a Substring In DB2, Oracle, and PostgreSQL, you can use the SUBSTR function to extract length characters from a string, beginning at position start: SUBSTR(string, start[, length]) Strings begin with position 1. Oracle treats a start of 0 as though you had specified 1. If start is negative, Oracle counts backward from the end of the string. Omit length to get all characters from start to the end of the string. DB2 pads any result with spaces, if necessary, to ensure that the result is always length characters long. NOTE In Oracle Database 10g and higher, you can also use REGEXP_SUBSTR, as described in the section “Regular Expressions” on page 113. Oracle implements SUBSTR, SUBSTRB, SUBSTR2, and SUBSTR4, which work in terms of the input character set, bytes, Unicode code units, and Unicode code points, respec- tively. DB2 implements SUBSTR (and also SUBSTRB in version 9.7). PostgreSQL implements only SUBSTR. PostgreSQL also supports: SUBSTRING(string FROM start) SUBSTRING(string FROM start FOR length) 136 | SQL Pocket Guide In SQL Server, use SUBSTRING. All three arguments are required: SUBSTRING(string, start, length) MySQL implements the following substring functions: SUBSTRING(string, start) SUBSTRING(string FROM start) SUBSTRING(string, start, length) SUBSTRING(string FROM start FOR length) The arguments to these SUBSTRING functions are the same as they are for SUBSTR. MySQL supports a negative start po- sition, which counts from the right. Finding the Length of a String Use the LENGTH function (LEN in SQL Server) to determine the length of a string: LENGTH(string) Oracle implements LENGTH, LENGTHB, LENGTH2, and LENGTH4, which count characters in the input character set, bytes, Unicode code units, and Unicode code points, respectively. Concatenating Strings The easiest way to concatenate strings is to use the SQL stand- ard string concatenation operator (||): string1 || string2 SQL Server does not support the ISO SQL string concatenation operator. Use a + instead: string1 + string2 MySQL supports neither || nor + for concatenating strings, but it does support an unlimited number of string arguments to CONCAT: CONCAT(string[, string ...]) String Functions | 137 PostgreSQL supports a TEXTCAT function: TEXTCAT(string, string) Trimming Unwanted Characters LTRIM, RTRIM, and TRIM remove unwanted characters from a string. TRIM is part of the SQL standard; the others are not. TRIM’s syntax is: TRIM(string) TRIM(character FROM string) TRIM(option [character] FROM string) option ::= {LEADING|TRAILING|BOTH} Using TRIM, you can trim leading characters, trailing charac ters, or both from a string. The character to trim defaults to a single space. The default option is BOTH. NOTE SQL Server began to support TRIM in SQL Server 2005. LTRIM removes unwanted characters from the beginning (left edge) of a string, whereas RTRIM removes from the end (right edge). The implementation for Oracle and PostgreSQL is: LTRIM(string[, unwanted]) RTRIM(string[, unwanted]) The unwanted argument is a string containing the characters you want trimmed, and it defaults to a single space. For ex- ample, to remove various punctuation from both ends of a string, specify: RTRIM(LTRIM(string,'.,! '),'.,! ') DB2, MySQL, and SQL Server do not support the unwanted argument; you can trim only spaces. 138 | SQL Pocket Guide Changing the Case of a String Use the UPPER and LOWER functions to upper- or lowercase all letters in a string: UPPER(string) LOWER(string) In DB2 9.7 and higher, Oracle, and PostgreSQL, you can also use INITCAP(string) to uppercase the first letter of each word in a string and lowercase the other letters. DB2 supports UCASE and LCASE as synonyms for UPPER and LOWER. Subqueries Subject to various platform restrictions, subqueries can be used in most SQL statements as follows: In the SELECT list of a SELECT statement See “The SELECT Clause” on page 125. In the FROM clause of a SELECT statement See “The FROM Clause” on page 130. In the WHERE clause of a SELECT statement See “Predicates” on page 109, and also “The WHERE Clause” on page 132. In the ORDER BY clause of a SELECT statement See “The ORDER BY Clause” on page 133. In an INSERT…SELECT…FROM statement See “Subquery Inserts” on page 69. In the SET clause of an UPDATE statement See “New Values from a Subquery” on page 169. Subqueries | 139 Subqueries in the FROM Clause A subquery in the FROM clause of a SELECT statement func- tions like a view and replaces a table as a data source. You can use subqueries—just as you can use views—as targets of IN- SERT, DELETE, and UPDATE statements. For example, for all platforms except MySQL and SQL Server, you can specify: DELETE FROM (SELECT * FROM upfall u WHERE u.open_to_public = 'n') u2 WHERE u2.owner_id IS NOT NULL; This statement deletes waterfalls that are not open to the public and for which an owner is known. Subqueries in the WITH Clause The SQL standard defines a WITH clause that you can use to factor out a subquery so that you don’t need to repeat it in your SELECT statement. DB2, Oracle, PostgreSQL, and SQL Server support WITH. NOTE See “Hierarchical Queries” on page 62 to learn how WITH is used to write recursive queries. The following SELECT repeats two subqueries twice to gen- erate a list of counties containing more than the average number of waterfalls per county: SELECT c.name, (SELECT COUNT(*) FROM upfall u2 WHERE u2.county_id = c.id) fall_count, (SELECT AVG(fall_count) FROM (SELECT COUNT(*) fall_count FROM upfall u3 GROUP BY u3.county_id) x1) avg_count FROM county c WHERE (SELECT COUNT(*) FROM upfall u2 WHERE u2.county_id = c.id) 140 | SQL Pocket Guide > (SELECT AVG(fall_count) FROM (SELECT COUNT(*) fall_count FROM upfall u3 GROUP BY u3.county_id) x2); Aside from being difficult to read and comprehend, this query is a potential maintenance disaster because any change to either subquery must be made twice. Using WITH, you can rewrite the query in a way that specifies each subquery only once. For example, in all but PostgreSQL: WITH fall_count_query AS (SELECT u2.county_id id, COUNT(*) fall_count FROM upfall u2 GROUP BY u2.county_id), avg_count_query AS (SELECT AVG(fall_count) avg_count FROM (SELECT COUNT(*) fall_count FROM upfall u3 GROUP BY u3.county_id)) SELECT c.name, (SELECT fall_count FROM fall_count_query WHERE fall_count_query.id = c.id) fall_count, (SELECT avg_count FROM avg_count_query) avg_count FROM county c WHERE (SELECT fall_count FROM fall_count_query WHERE fall_count_query.id = c.id) > (SELECT avg_count FROM avg_count_query); PostgreSQL requires an alias for any subquery in the FROM clause of a subquery in the WITH clause. For example, notice the alias fc at the end of the following snippet: ... avg_count_query AS (SELECT AVG(fall_count) avg_count FROM (SELECT COUNT(*) fall_count FROM upfall u3 GROUP BY u3.county_id) fc) ... The correlated subquery changes to a noncorrelated version as it moves into the WITH clause. The original query used the Subqueries | 141 following subquery to retrieve the number of waterfalls for a given county: (SELECT COUNT(*) FROM upfall u2 WHERE u2.county_id = c.id) fall_count, In the WITH version of the query, the correlated subquery that counts waterfalls for a single county is replaced with a GROUP BY subquery that counts falls for all counties: (SELECT u2.county_id id, COUNT(*) fall_count FROM upfall u2 GROUP BY u2.county_id), The “correlation” becomes a WHERE clause when selecting from the factored-out query: (SELECT fall_count FROM fall_count_query WHERE fall_count_query.id = c.id) When moving a correlated subquery into the WITH clause, you’ll need to uncorrelate it. Determining how best to accom- plish this sometimes requires a bit of thought and experimentation. Although the preceding query using the WITH clause is some- what more complex than the one it replaces, the logic for com- puting the fall count and average fall count is now encapsulated in the WITH clause. The other subqueries do nothing more than select specific columns from the result sets of the WITH- clause queries. As the size of the subqueries increases, so does the apparent simplification. Further refactoring is possible. This time, the avg_count_query references the previously defined fall_count_query, consolidating the logic for counting water- falls by county into only one subquery: WITH fall_count_query AS (SELECT u2.county_id id, COUNT(*) fall_count FROM upfall u2 GROUP BY u2.county_id), avg_count_query AS (SELECT AVG(fall_count) avg_count 142 | SQL Pocket Guide FROM (SELECT * FROM fall_count_query)) SELECT c.name, (SELECT fall_count FROM fall_count_query WHERE fall_count_query.id = c.id) fall_count, (SELECT avg_count FROM avg_count_query) avg_count FROM county c WHERE (SELECT fall_count FROM fall_count_query WHERE fall_count_query.id = c.id) > (SELECT avg_count FROM avg_count_query); DB2 and SQL Server support an alternative method for naming the result columns from a WITH-clause query. Instead of pro- viding column names as aliases, you can provide them in parentheses following the query name: WITH fall_count_query (id, fall_count) AS The WITH clause doesn’t eliminate multiple subqueries en- tirely. It does allow you to locate all of the complex logic in one place, leaving only simple SELECTs for the subqueries in the main statement. Tables, Creating You create a new table in a database by issuing a CREATE TABLE statement. The syntax varies widely among vendors, but the following subsections show reasonable examples for each platform. Bear in mind the following points: • At a minimum, all you need is a list of column names and their data types: CREATE TABLE simple_example ( id NUMERIC, name VARCHAR(15), last_changed DATE ); • The examples give explicit names for many of the constraints, which I consider a best practice, but the CONSTRAINT constraint_name syntax is optional and is often omitted (especially on column constraints such as the NOT NULL constraint). Tables, Creating | 143 • You can usually declare constraints that involve a single column as part of that column’s definition. Multicolumn constraints must be declared as table-level elements. The examples demonstrate both approaches. See the platform-specific sections on “Data Types” for lists of valid data types by platform. Creating a Table: DB2 The following is a typical CREATE TABLE statement for DB2: CREATE TABLE db2_example ( id DECIMAL(6) NOT NULL GENERATED ALWAYS AS IDENTITY ( START WITH 1 INCREMENT BY 1 MAXVALUE 999999 CACHE 20 NO ORDER), name VARCHAR(15) NOT NULL, country VARCHAR(2) DEFAULT 'CA' NOT NULL CONSTRAINT country_check CHECK (country IN ('CA','US')), indexed_name VARCHAR(15), CONSTRAINT db2_example_pk PRIMARY KEY (id), CONSTRAINT db2_example_fk01 FOREIGN KEY (name, country) REFERENCES parent_example (name, country), CONSTRAINT db2_example_u01 UNIQUE (name, country), CONSTRAINT db2_example_c01 CHECK (indexed_name = UPPER(name)) ) IN userspace1; In DB2, you must specify NOT NULL explicitly for all primary key columns. Other vendors generally infer NOT NULL from your primary key specification. Likewise, DB2 requires NOT NULL on columns involved in UNIQUE constraints. The id column in this table is automatically generated from a sequence of values from 1 to 9999999. Sequence values are cached in memory for faster access and are not necessarily as- signed in order (which also improves performance). 144 | SQL Pocket Guide Creating a Table: MySQL The following is a typical CREATE TABLE statement for MySQL. The id column is autogenerated: CREATE TABLE mysql_example ( id INTEGER AUTO_INCREMENT, name VARCHAR(15) NOT NULL, country VARCHAR(2) DEFAULT 'CA' NOT NULL CHECK (country IN ('CA','US')), indexed_name VARCHAR(15), CONSTRAINT mysql_example_pk PRIMARY KEY (id), CONSTRAINT mysql_example_fk01 FOREIGN KEY (name, country) REFERENCES parent_example (name, country), CONSTRAINT mysql_example_u01 UNIQUE (name, country), CONSTRAINT mysql_example_index_upper CHECK (indexed_name = UPPER(name)) ) ENGINE = INNODB; MySQL does not support the CONSTRAINT constraint_name syntax or the definition of foreign key and check constraints at the column level. WARNING MySQL silently ignores foreign key constraints, except between InnoDB tables. It will even silently ignore the declaration of such constraints to tables that do not exist, unless you are creating an InnoDB table. MySQL supports different storage engines, which are physical mechanisms for storing table rows. Use the ENGINE keyword to specify an engine type. The following are valid engines in MySQL 5.1: ARCHIVE, BLACKHOLE, CSV, EXAMPLE, FEDERATED, INNODB, MEMORY, MERGE, MYISAM (called ISAM prior to 5.0), and NDBCLUSTER. MYISAM is the default, although that can be changed when starting the MySQL daemon. Tables, Creating | 145 NOTE Earlier versions of MySQL require you to use the key- word TYPE rather than ENGINE. Creating a Table: Oracle The following is a typical CREATE TABLE statement for Oracle: CREATE TABLE oracle_example ( id NUMBER(6), name VARCHAR2(15) NOT NULL, country VARCHAR2(2) DEFAULT 'CA' CONSTRAINT country_not_null NOT NULL CONSTRAINT country_check CHECK (country IN ('CA','US')), indexed_name VARCHAR2(15), CONSTRAINT oracle_example_pk PRIMARY KEY (id), CONSTRAINT oracle_example_fk01 FOREIGN KEY (name, country) REFERENCES parent_example (name, country), CONSTRAINT oracle_example_u01 UNIQUE (name, country), CONSTRAINT oracle_example_index_upper CHECK (indexed_name = UPPER(name)) ) TABLESPACE users; This statement assigns the table to the users tablespace. The TABLESPACE clause is optional. If you aren’t certain which tablespace to specify, you can either omit the clause to accept your default tablespace assignment or ask your database administrator’s advice. If you want the ID column to be an automatically generated sequential ID number, you can begin by creating an Oracle sequence: CREATE SEQUENCE oracle_example_pk NOCYCLE MAXVALUE 999999 START WITH 1; Then, create a trigger to derive a new id value from the sequence whenever a new row is inserted: 146 | SQL Pocket Guide CREATE OR REPLACE TRIGGER oracle_example_pk BEFORE INSERT ON oracle_example FOR EACH ROW DECLARE next_id NUMBER; BEGIN SELECT oracle_example_pk.NEXTVAL INTO next_id FROM dual; :NEW.id := next_id; END; / Oracle sequences generate values up to 1027. Use the MAX- VALUE clause to constrain the value range to something that is appropriate to your application and does not exceed the range of your primary key column. Creating a Table: PostgreSQL The following is a typical CREATE TABLE statement for PostgreSQL: CREATE TABLE postgre_example ( id SERIAL, name VARCHAR(15) NOT NULL, country VARCHAR(2) DEFAULT 'CA' CONSTRAINT country_not_null NOT NULL CONSTRAINT country_check CHECK (country IN ('CA','US')), indexed_name VARCHAR(15), CONSTRAINT postgre_example_pk PRIMARY KEY (id), CONSTRAINT postgre_example_fk01 FOREIGN KEY (name, country) REFERENCES parent_example (name, country), CONSTRAINT postgre_example_u01 UNIQUE (name, country), CONSTRAINT postgre_example_index_upper CHECK (indexed_name = UPPER(name)) )TABLESPACE pg_default; The id column’s type is SERIAL, which results in an auto- incrementing four-byte integer. Support for tablespaces came Tables, Creating | 147 about in PostgreSQL 8.0. The TABLESPACE clause is optional. Creating a Table: SQL Server The following is a typical CREATE TABLE statement for SQL Server, with an auto-incrementing primary key column that begins at 1 and increments by 1: CREATE TABLE msss_example ( id INT IDENTITY (1,1), name VARCHAR(15) NOT NULL, country VARCHAR(2) DEFAULT 'CA' CONSTRAINT country_not_null NOT NULL CONSTRAINT country_check CHECK (country IN ('CA','US')), indexed_name VARCHAR(15), CONSTRAINT msss_example_pk PRIMARY KEY (id), CONSTRAINT msss_example_fk01 FOREIGN KEY (name, country) REFERENCES parent_example (name, country), CONSTRAINT msss_example_u01 UNIQUE (name, country), CONSTRAINT msss_example_index_upper CHECK (indexed_name = UPPER(name)) ); Tables, Dropping When you no longer need a table, you can drop it from your schema: DROP TABLE table_name; In Oracle, you can drop a table that is referenced by foreign key constraints using the following syntax: DROP TABLE table_name CASCADE CONSTRAINTS; In PostgreSQL, you can do the same thing using: DROP TABLE table_name CASCADE; 148 | SQL Pocket Guide Foreign key constraints that reference the table being dropped will be dropped themselves. NOTE In DB2, referencing foreign key constraints are always dropped; no CASCADE clause is needed. In all other cases, you must drop any referencing foreign key constraints manually before dropping the referenced table. Tables, Modifying You can change the columns and other attributes of a table using the ALTER TABLE statement. The syntax varies signif- icantly among vendors. The following subsections show the same sequence of common table alterations. Many other types of changes are possible; consult your vendor documentation for details. Modifying a Table: DB2 Use ALTER TABLE’s ADD clause to add a column or table constraint. You may add more than one item at a time: ALTER TABLE db2_example ADD COLUMN lower_name VARCHAR(15) ADD CONSTRAINT lower_name CHECK(lower_name = LOWER(name)); Use the ALTER clause to change a column’s default value or data type. For example: ALTER TABLE db2_example ALTER COLUMN name SET DEFAULT 'Missing!' ALTER COLUMN indexed_name SET DATA TYPE VARCHAR(30); You can change only one item at a time for a given column. If you need to change both name and data type for a given Tables, Modifying | 149 column, you will need to issue separate ALTER TABLE state- ments for each of those two changes. You can add table constraints but not column constraints, so the no_leading_space constraint added at the column level on other platforms must be added at the table level for DB2: ALTER TABLE db2_example ADD CONSTRAINT no_leading_space CHECK (indexed_name = LTRIM(indexed_name)); DB2 9.7 and higher support changing the nullability of a col- umn. For example: ALTER TABLE db2_example ALTER COLUMN name SET NOT NULL; To change the nullability of a column prior to DB2 9.7, you must drop and recreate the table. (Remember that columns participating in unique and primary key constraints cannot be nullable.) Unlike on the other platforms, there is no easy way to make name nullable. However, you can achieve the desired effect of adding NOT NULL to a column by creating a CHECK constraint: ALTER TABLE db2_example ADD CONSTRAINT indexed_name_not CHECK (indexed_name IS NOT NULL); To remove a constraint, use the DROP clause: ALTER TABLE db2_example DROP CONSTRAINT lower_name DROP CONSTRAINT no_leading_space DROP CONSTRAINT indexed_name_not; You cannot drop a column from a table in DB2. If avoiding the use of an unwanted column is not sufficient, you must drop and recreate the table. Modifying a Table: MySQL Use the ADD clause to add columns and constraints. Be sure to avoid using the same name for both a column and a constraint: 150 | SQL Pocket Guide ALTER TABLE mysql_example ADD lower_name VARCHAR(15), ADD CONSTRAINT lower_name_chk CHECK (lower_name = LOWER(name)); To create new definitions for a column, use MODIFY. You must specify at least a data type for each column, and you may also specify a default value and nullability (e.g., NOT NULL). New definitions completely overwrite the old. Thus, in the fol- lowing example, country will lose its existing default value because it was not respecified in the MODIFY clause: ALTER TABLE mysql_example MODIFY name VARCHAR(30) DEFAULT 'Missing!' NULL, MODIFY country VARCHAR(2) NOT NULL, MODIFY indexed_name VARCHAR(30) NOT NULL; Constraints—even those referencing a single column—must be added via the ADD clause: ALTER TABLE mysql_example ADD CONSTRAINT no_leading_space CHECK (indexed_name = LTRIM(indexed_name)); MySQL does not allow you to drop CHECK constraints. You can drop primary key and foreign key constraints as follows: ALTER TABLE table_name DROP PRIMARY KEY, DROP FOREIGN KEY constraint_name; To drop a UNIQUE constraint, you must drop the index used to enforce it: ALTER TABLE table_name DROP INDEX index_name; Use DROP to remove a column: ALTER TABLE mysql_example DROP COLUMN lower_name; Tables, Modifying | 151 Modifying a Table: Oracle Use ALTER TABLE…ADD to add columns and table constraints: ALTER TABLE oracle_example ADD ( lower_name VARCHAR2(15), CONSTRAINT lower_name CHECK (lower_name = LOWER(name)) ); Use MODIFY to change a column’s data type, default value, or nullability. You can also add new constraints to a column. Anything you do not specify is left unchanged: ALTER TABLE oracle_example MODIFY ( name VARCHAR2(30) DEFAULT 'Missing!' CONSTRAINT name_canbe_null NULL, country DEFAULT NULL, indexed_name varchar2(30) NOT NULL CONSTRAINT no_leading_space CHECK (indexed_name = LTRIM(indexed_name)) ); Use DROP to remove a column or constraint. Each drop must be specified separately, and no parentheses are used: ALTER TABLE oracle_example DROP CONSTRAINT lower_name; ALTER TABLE oracle_example DROP COLUMN lower_name; Modifying a Table: PostgreSQL Use the ADD clause to add columns and constraints: ALTER TABLE postgre_example ADD lower_name VARCHAR(15), ADD CONSTRAINT lower_name CHECK (lower_name = LOWER(name)); Use various ALTER clauses to change a column’s data types, default values, and nullability: ALTER TABLE postgre_example ALTER name TYPE VARCHAR(30), 152 | SQL Pocket Guide ALTER name SET DEFAULT 'Missing!', ALTER name DROP NOT NULL, ALTER country DROP DEFAULT, ALTER indexed_name TYPE VARCHAR(30), ALTER indexed_name SET NOT NULL; Constraints—even those referencing a single column—must be added via the ADD clause: ALTER TABLE postgre_example ADD CONSTRAINT no_leading_space CHECK (indexed_name = LTRIM(indexed_name)); Use DROP to remove a column or constraint: ALTER TABLE postgre_example DROP CONSTRAINT lower_name, DROP COLUMN lower_name; Modifying a Table: SQL Server Use ALTER TABLE…ADD to add columns and table constraints: ALTER TABLE msss_example ADD lower_name VARCHAR(15), CONSTRAINT lower_name CHECK (lower_name = LOWER(name)); Issue ALTER TABLE…ALTER COLUMN to modify a col- umn’s data type or nullability. You can make only one alteration per statement: ALTER TABLE msss_example ALTER COLUMN name VARCHAR(15) NULL; ALTER TABLE msss_example ALTER COLUMN country VARCHAR(2) NULL; ALTER TABLE msss_example ALTER COLUMN indexed_name VARCHAR(30) NOT NULL; Column name’s data type cannot be changed in this example because SQL Server does not allow data type changes for Tables, Modifying | 153 columns involved in certain types of constraints (e.g., foreign key constraints). You cannot add column-level constraints—only table-level constraints. The following example adds a constraint that tests the value in a column, but the constraint is associated with the table, not with the column in question: ALTER TABLE msss_example ADD CONSTRAINT no_leading_space CHECK (indexed_name = LTRIM(indexed_name)); Add or remove default values by adding or removing so-called default constraints. For example: ALTER TABLE msss_example ADD CONSTRAINT name_default DEFAULT 'Missing!' FOR name; ALTER TABLE msss_example DROP CONSTRAINT name_default; To remove the default value from the country column, you must first look up the automatically generated constraint name (e.g., via the GUI), and then drop that constraint by specifying its name. Use ALTER TABLE…DROP to remove a column or constraint: ALTER TABLE msss_example DROP CONSTRAINT no_leading_space; Transaction Management A transaction is a collection of operations treated as a unit. Either all operations in the unit are completed or none of them are. All commonly used databases make provisions for transactions. When working in a transactional environment, you need to know how to begin and end a transaction. You also need to know how to specify various characteristics of a transaction— for example, whether it will update any data. 154 | SQL Pocket Guide Autocommit Mode MySQL, PostgreSQL, and SQL Server default to an autocommit mode in which each statement you execute is treated as a transaction in and of itself. (Thus, you cannot roll back a statement when the result isn’t what you expected). You can disable autocommit in SQL Server with the following statement: SET IMPLICIT_TRANSACTIONS ON You can enable autocommit again using: SET IMPLICIT_TRANSACTIONS OFF You leave SQL Server’s and PostgreSQL’s autocommit mode whenever you issue an explicit BEGIN TRANSACTION (SQL Server) or BEGIN (PostgreSQL) statement. See “Starting a Transaction” below for details. In MySQL, you can disable autocommit with: SET AUTOCOMMIT=0 And you can enable it again with: SET AUTOCOMMIT=1 You automatically leave autocommit mode whenever you issue a BEGIN or BEGIN WORK statement. Starting a Transaction: DB2 DB2 does not implement an SQL statement to explicitly begin a transaction. When you connect and issue an SQL statement, you begin a transaction. You also begin a transaction with the first SQL statement following a COMMIT. Starting a Transaction: MySQL Use START TRANSACTION to begin a MySQL transaction explicitly (when using any version prior to MySQL 4.0.11, use BEGIN or BEGIN WORK). When not in autocommit mode, Transaction Management | 155 any SQL statement you issue will begin a new transaction implicitly. WARNING Only certain types of MySQL tables (InnoDB tables, for example) support transactions. Changes to data in non- transactional tables take place immediately and perma- nently, regardless of whether you are in a transaction. Before beginning a transaction, you can use SET TRANSAC- TION to change the transaction isolation level. A reasonable sequence of statements might then be: SET [GLOBAL|SESSION] TRANSACTION ISOLATION LEVEL {READ UNCOMMITTED|READ COMMITTED |REPEATABLE READ|SERIALIZABLE}; START TRANSACTION; By default, SET TRANSACTION sets the isolation level only for your next transaction. Use SET SESSION TRANSACTION to set the default isolation level for your entire session. Starting a Transaction: Oracle Within Oracle, for all practical purposes, you’re always in a transaction. The first SQL statement you execute after you connect begins an implicit transaction, as does the first SQL statement you execute following the end of a transaction. Ora- cle’s default transaction type is read/write with statement-level read consistency. You can begin a transaction using SET TRANSACTION explicitly: SET TRANSACTION options [NAME 'tran_name'] options ::= {READ {ONLY|WRITE} |ISOLATION LEVEL {SERIALIZABLE|READ COMMITTED} |USE ROLLBACK SEGMENT segment_name 156 | SQL Pocket Guide The options and parameters are as follows: NAME 'tran_name ' Specifies a name of up to 255 bytes for the transaction. Upon COMMIT, the name will be saved as the transaction comment, overriding any COMMIT comment. It’s espe- cially helpful to name distributed transactions. READ ONLY Gives you a read-only transaction that does not “see” any changes committed after the transaction begins. READ WRITE Gives you the default transaction type: a read/write trans- action with statement-level read consistency. ISOLATION LEVEL SERIALIZABLE Gives you a read/write serializable transaction, as defined in the SQL standard. ISOLATION LEVEL READ COMMITTED Gives you the default Oracle transaction behavior, but using ANSI/ISO SQL syntax. USE ROLLBACK SEGMENT segment_name Creates a default transaction and assigns it to the specified rollback segment. (Obsolete; use automatic undo man- agement instead.) Here are some example SET TRANSACTION statements: SET TRANSACTION READ ONLY; SET TRANSACTION ISOLATION LEVEL SERIALIZABLE; SET TRANSACTION ISOLATION LEVEL READ COMMITTED; NAME 'Delete all attractions'; If you name a distributed transaction and that transaction fails, its name will appear in the DBA_2PC_PENDING table’s TRAN_COMMENT column. Transaction Management | 157 Starting a Transaction: PostgreSQL To start a PostgreSQL transaction, issue a BEGIN command, which takes PostgreSQL out of autocommit mode. The syntax for PostgreSQL 8.0 and higher is: BEGIN [WORK|TRANSACTION] [iso_mode|mode_iso] iso_mode ::= isolation [[,] mode] mode_iso ::= mode [[,] isolation] isolation ::= ISOLATION LEVEL {SERIALIZABLE|REPEATABLE READ |READ COMMITTED|READ UNCOMMITTED} mode ::= {READ WRITE|READ ONLY} For example: BEGIN ISOLATION LEVEL READ COMMITTED, READ WRITE; In this syntax, you can use the keyword START instead of BEGIN. Also, you can separate isolation and mode using either whitespace or a comma. The default isolation and mode are READ COMMITTED and READ WRITE, respectively. In PostgreSQL 7.4 and earlier, BEGIN syntax was simply: BEGIN [WORK|TRANSACTION] Next, to set transaction characteristics, you can follow the BEGIN statement immediately with a SET TRANSACTION statement: SET TRANSACTION [iso_mode|mode_iso] For example: BEGIN; SET TRANSACTION READ WRITE; Finally, you can set the default isolation and mode for a session by using the following command: SET SESSION CHARACTERISTICS AS TRANSACTION [iso_mode|mode_iso] 158 | SQL Pocket Guide Starting a Transaction: SQL Server Use the following statement to begin an SQL Server transaction explicitly: BEGIN TRAN[SACTION] [[transaction_name] [WITH MARK ['description']]] Transaction names are limited to 32 characters. You can spec- ify a name by means of a variable, as in @variable. Use the WITH MARK clause to note a transaction in the da- tabase log; you can also specify a description for it if you wish. To begin a distributed transaction, use: BEGIN DISTRIBUTED TRAN[SACTION] [transaction_name] As with BEGIN TRANSACTION, you can specify the trans- action name by means of a variable in the form @variable. SQL Server’s default isolation level is READ COMMITTED. Before beginning a transaction, use the following statement to specify the isolation level of your choice: SET TRANSACTION ISOLATION LEVEL {READ COMMITTED|READ UNCOMMITTED |REPEATABLE READ|SERIALIZABLE} This statement sets the isolation level to be used for all subse- quent transactions in your session. Ending a Transaction To end a transaction and make the transaction’s changes per- manent, issue a COMMIT statement: COMMIT [WORK] Oracle supports an optional COMMENT clause: COMMIT [WORK] [COMMENT 'text'] WORK is an optional word allowed by the ISO SQL standard (but not supported by MySQL), and it is commonly omitted. Transaction Management | 159 In Oracle, any name you specify using SET TRANSACTION when you begin a transaction overrides any comment you specify when you commit that transaction. SQL Server also supports a COMMIT TRANSACTION state- ment, which enables you to identify the transaction you want to commit: COMMIT TRAN[SACTION] [transaction_name] SQL Server actually ignores any transaction_name that you specify. It allows a name only to make it easier for you to as- sociate nested COMMITs with their corresponding BEGIN TRANSACTION statements. Oracle supports the following syntax to force a distributed transaction to commit: COMMIT [WORK] FORCE {'local_tran_id'|'global_tran_id'} [system_change_number] You identify a distributed transaction using either its local or global transaction ID, which you can obtain from the DBA_2PC_PENDING view. You have the option of assigning a system change number (SCN) or defaulting to the current SCN. NOTE DDL statements such as TRUNCATE also typically end transactions. However, issuing a DDL statement and depending on an implicit COMMIT is not as clean and tidy as issuing the COMMIT explicitly. Aborting a Transaction To abort a transaction, use the ROLLBACK statement: ROLLBACK [WORK] 160 | SQL Pocket Guide As with COMMIT, the word WORK (which is not supported by MySQL) is commonly omitted. When you ROLLBACK a transaction, you undo all of that transaction’s changes. SQL Server also supports a ROLLBACK TRANSACTION statement, which enables you to specify the name of the trans- action to roll back: ROLLBACK TRAN[SACTION] [transaction_name] By default, ROLLBACK TRANSACTION rolls back the cur- rent transaction. In a nested transaction, that means the in- nermost transaction. If you specify a transaction name, you must specify the outermost transaction. That transaction and all nested transactions are then undone. Oracle supports the following syntax to force a distributed transaction to roll back: ROLLBACK [WORK] FORCE {'local_tran_id'|'global_tran_id'} You identify a distributed transaction using either its local or global transaction ID, which you can obtain from the DBA_2PC_PENDING view. NOTE In SQL Server, use SET XACT_ABORT {ON|OFF} to deter- mine whether an error in an SQL statement aborts the current transaction automatically. Aborting to a Savepoint Rather than rolling back an entire transaction, you can roll back only part of one. To do this, you must have marked points in the transaction, known as savepoints, which are specified using the following syntax for MySQL, Oracle, and PostgreSQL: SAVEPOINT savepoint_name Transaction Management | 161 For DB2, you can specify: SAVEPOINT savepoint_name [UNIQUE] ON ROLLBACK RETAIN CURSORS [ON ROLLBACK RETAIN LOCKS] For SQL Server, you can specify: SAVE TRAN[SACTION] savepoint_name You can then ROLLBACK to any of those savepoints using: ROLLBACK [WORK] TO savepoint_name Except in DB2, you must use: ROLLBACK TO SAVEPOINT savepoint_name The following is an example from Oracle: SET TRANSACTION ISOLATION LEVEL READ COMMITTED; UPDATE township SET name = UPPER(name); SAVEPOINT name_upper_cased; DELETE FROM trip; ROLLBACK TO name_upper_cased; COMMIT; The net effect of this transaction is to set all township names to uppercase. The DELETE against the trip table is undone by the ROLLBACK TO the savepoint that was established fol- lowing the UPDATE statement. Union Queries Union queries use keywords such as UNION, EXCEPT (MINUS in Oracle), and INTERSECT to “combine” results from two or more queries in useful ways. UNION and UNION ALL Use the UNION keyword to combine results from two SELECT statements into one result set. (Think of stacking the rows from two result sets.) Any duplicate rows are eliminated from the final results, unless you specify UNION ALL to pre- serve them. 162 | SQL Pocket Guide NOTE Some would argue that you should use UNION ALL when you know for a certainty that no duplicates are possible, thus improving performance by avoiding the sort. UNION The UNION operator conforms closely to SQL’s origin in set theory. It is used to combine two rowsets and remove any du- plicates from the results. For example: SELECT u.id, u.name FROM upfall u WHERE open_to_public = 'y' UNION SELECT u.id, u.name FROM upfall u JOIN owner o ON u.owner_id = o.id WHERE o.type = 'public'; This query lists waterfalls that are either open to the public or that are owned by a public entity (such as a national park). Duplicate elimination ensures that even if a waterfall fits into both categories (is both open to the public and owned by a public entity), it is returned only once in the query’s result set. NOTE Duplicate elimination requires overhead, generally in the form of a limited sort operation. If you don’t need duplicate elimination, you’ll get better performance with UNION ALL. UNION ALL UNION ALL is UNION without the duplicate elimination. The following UNION ALL query simulates an outer join, with upfall as the required table and owner as the optional table. The first SELECT picks up waterfalls that can join to owner, Union Queries | 163 whereas the second SELECT picks up those falls with no known owner: SELECT u1.name AS fall, o.name AS owner FROM upfall u1 JOIN owner o ON u1.owner_id = o.id UNION ALL SELECT u2.name AS fall, 'Unknown' AS owner FROM upfall u2 WHERE u2.owner_id IS NULL; Getting correct results from this query depends on a foreign key integrity constraint to ensure that any non-null value in upfall.owner_id references an existing row in owner. Without such a constraint, the second SELECT must be written to in- clude rows with invalid owner_id values. ORDER BY in Union Queries SQL allows only one ORDER BY clause per query. In a union query, the ORDER BY clause belongs at the very end: SELECT u.id, u.name FROM upfall u WHERE open_to_public = 'y' UNION SELECT u.id, u.name FROM upfall u JOIN owner o ON u.owner_id = o.id WHERE o.type = 'public' ORDER BY name; The sorting operation then applies to the collective results from all SELECT expressions involved in the union. Names and Data Types in a Union The column names used for the first SELECT in a union query become the names of their respective result columns. In the following query, the columns will be named col_a and col_b: SELECT 'One' col_a, 'Two' col_b FROM dual UNION SELECT 'Three', 'Four' FROM dual; 164 | SQL Pocket Guide Also be aware that column data types must correspond. For example, don’t try to perform a union of a numeric column to a text column without first writing an explicit conversion to synchronize the data types. NOTE Remember, PostgreSQL requires the AS keyword when specifying column aliases (e.g., 'One' AS col_a). Order of Evaluation When writing a statement with multiple UNION operations, you can use parentheses to specify the order in which the union operations occur in all platforms except MySQL. The follow- ing is a contrived example (to run it in Oracle, replace EXCEPT with MINUS): SELECT * FROM upfall EXCEPT (SELECT * FROM upfall UNION SELECT * FROM upfall); The first SELECT returns all rows from upfall. The UNION of the two SELECTs in parentheses also returns all rows from upfall. When you subtract all rows from all rows, you have none left. Thus, as written, the query returns no rows. If you remove the parentheses, however, you’ll get all upfall rows in the result set. NOTE Unless you specify otherwise, union operations are per- formed in top-down order, except that INTERSECT takes precedence over UNION and EXCEPT. Union Queries | 165 EXCEPT (or MINUS) and EXCEPT ALL Use the EXCEPT union operation (MINUS in Oracle) to “sub- tract” the results of one query from another. If you do not need duplicate elimination, use EXCEPT ALL. Note that MySQL does not support EXCEPT. EXCEPT (MINUS in Oracle) To find all owners without waterfalls, you can subtract the list of owners with falls from the total list of owners: SELECT o.id FROM owner o EXCEPT SELECT u.owner_id FROM upfall u; Remember that you must use the MINUS keyword rather than EXCEPT to run this query in Oracle. EXCEPT ALL DB2 and PostgreSQL support EXCEPT ALL. The following query uses that operation to return a list of owners who have at least two waterfalls: SELECT u.owner_id FROM upfall u EXCEPT ALL SELECT o.id FROM owner o; The first SELECT will potentially return many IDs per owner (one from each fall that is owned), whereas the second SELECT will return exactly one ID for each owner. Owners with two or more falls will have their IDs listed two or more times in the results of the first query. In the case of Pictured Rocks, the resulting subtraction looks like this: 1 1 1 EXCEPT ALL 1 Because EXCEPT ALL is used, the single owner ID from the owner table is subtracted from the three from the upfall table, which leaves two occurrences in the final result set: 166 | SQL Pocket Guide 1 1 Because the EXCEPT operation still requires some sorting of the results to perform the subtraction operation, using EXCEPT ALL may not produce the same kind of performance boost you get from using UNION ALL instead of UNION. INTERSECT and INTERSECT ALL The INTERSECT operation is used to find rows in common between the result sets of two SELECTs. Use INTERSECT ALL when you do not want duplicate elimination. Note that MySQL does not support INTERSECT. INTERSECT The following INTERSECT query is similar to the query used to illustrate UNION. In this example, rather than being an “either/or” query, the query returns falls that are both open to the public and owned by a private (not public this time) organization: SELECT u.id, u.name FROM upfall u WHERE open_to_public = 'y' INTERSECT SELECT u.id, u.name FROM upfall u JOIN owner o ON u.owner_id = o.id WHERE o.type = 'private'; Some kind of sorting or hashing operation will be executed to find rows in common between the two result sets. Duplicate elimination ensures that each fall is returned only once. INTERSECT ALL Use INTERSECT ALL when you want to consider duplicates. Only DB2 and PostgreSQL support this. For example, given the following data: MARQUETTE MARQUETTE Union Queries | 167 BARAGA BARAGA MUNISING INTERSECT ALL MARQUETTE MARQUETTE BARAGA INTERSECT will yield: MARQUETTE BARAGA whereas INTERSECT ALL will yield: MARQUETTE MARQUETTE BARAGA Because Marquette appears twice in both result sets, it appears twice in the final results. Baraga, on the other hand, appears only once in the second result set, so it appears just once in the final result set. Updating Data To modify existing data in a table, use the UPDATE statement. You can update one row or many rows, you can specify a single set of new values in the statement, or you can generate new values through a subquery. Simple Updates A simple UPDATE takes the following form: UPDATE table SET column = value, column = value ... WHERE predicates In this form, predicates identifies one or more rows that you want to update. You can specify as many column = value pairs as you like—one for each column you want to modify: 168 | SQL Pocket Guide UPDATE upfall SET owner_id = 1 WHERE name = 'Munising Falls'; When you specify only one new value, you will usually want to update only one row, and your WHERE-clause predicates should reference primary or unique key values to identify that row. Using expressions, you can write sensible UPDATEs that modify many rows. The following example works in DB2: UPDATE upfall SET datum = UPPER(datum), lat_lon = TRIM(UPPER(lat_lon)); This example also demonstrates the use of the comma to sep- arate multiple-column updates in a SET clause. NOTE In MySQL, if you are updating a self-referential foreign key or its related primary key, you should include an ORDER BY clause at the end of your update to control the order in which rows are updated. For more on this issue, see “Deleting in Order” on page 47 and “Subquery Inserts” on page 69. New Values from a Subquery You can also generate new values from a subquery. One way to do this is to write separate subqueries for each column that you are updating: UPDATE table SET column = (subquery), column = (subquery), ... For example (note that PostgreSQL and SQL Server do not allow the table alias u): UPDATE upfall u SET owner_id = (SELECT o.id FROM owner o WHERE o.name = 'Pictured Rocks') WHERE u.name = 'Miners Falls'; Updating Data | 169 Such subqueries must always return zero or one row and one column. If zero rows are returned, then the value is set to null. In DB2 and Oracle, you can also write a subquery that returns more than one column value, in which case the number of val- ues returned must correspond to the columns you are updating: UPDATE table SET (column, column, ...) = (subquery) For example, to update names and descriptions with any new information in the new_falls table (see the section “Merging Data” on page 86 for a better way to do this), specify: UPDATE upfall u SET (u.name, u.description) = (SELECT nf.name, nf.description FROM new_falls nf WHERE u.id = nf.id) WHERE u.id IN (SELECT nf2.id FROM new_falls nf2); Be careful with this kind of update. If you omit the WHERE clause in this query, all rows in upfall will be updated, regard- less of whether corresponding rows exist in new_falls. Worse, upfall’s name and description columns will be set to null in cases where no corresponding new_falls rows exist. Updating Views and Subqueries All platforms allow UPDATEs to run against views. DB2 and Oracle also allow updates to run against subqueries (i.e., inline views): UPDATE (SELECT * FROM upfall WHERE owner_id IS NULL) SET open_to_public = 'n'; PostgreSQL does not support updates to inline views. Post- greSQL requires any view that is the target of an UPDATE statement to be associated with an ON UPDATE DO INSTEAD rule. 170 | SQL Pocket Guide Database systems place various restrictions on the updating of views, but in general, you must be able to access unambigu- ously a single table row from a given view row in order to issue an update against that view (or subquery). UPDATE FROM Clause PostgreSQL and SQL Server let you write a FROM clause in an UPDATE statement in order to gather columns from multiple tables to use in your SET expressions. For example, the fol- lowing statement works in SQL Server and appends the owner type from the owner table to each waterfall’s description: UPDATE upfall SET description = u.description + ' (' + o.type + ')' FROM upfall u JOIN owner o ON u.owner_id = o.id; When using this syntax, you must ensure that the UPDATE is deterministic, meaning that there is only one possible value for any column you reference in a SET expression. The join con- dition in this query accomplishes this—there will always be only one owner per waterfall. Notice the use of aliases in the SET clause. The first reference to description is unqualified. The column name you specify following the SET keyword must be in the table that is the target of the UPDATE statement. The second occurrence of description, however, is qualified. All values feeding into the update must come from the tables listed in the FROM clause. Returning Updated Data: DB2 DB2 allows you to query the before and after values from rows affected by an UPDATE statement. Simply SELECT from that UPDATE statement. For example: SELECT * FROM NEW TABLE ( UPDATE gov_unit SET name = UPPER(name) Updating Data | 171 WHERE MOD(id,2) = 0 ); Specify FROM NEW TABLE to see the newly updated values. Specify FROM OLD TABLE to see the original values. Be sure to wrap your UPDATE in parentheses. Returning Updated Data: Oracle You can use Oracle’s RETURNING clause to return values that you update. Here is the syntax: UPDATE ... SET ... WHERE ... RETURNING expression [,expression ...] INTO variable [,variable ...] If you update a single row, Oracle expects to return values into bind variables; if you update more than one row, it expects to return values into bind arrays. See “Returning Deleted Data: Oracle” on page 50 for an example involving arrays and “Re- turning Inserted Values: Oracle” on page 70 for a single-row example. Returning Updated Data: SQL Server You can use SQL Server’s OUTPUT clause to return values from newly inserted rows. For example: UPDATE gov_unit SET type = UPPER(type) OUTPUT INSERTED.id, INSERTED.type AS new_type, DELETED.type AS old_type; Specify INSERTED to reference post-update values. Specify DELETED to reference pre-update values. The preceding query displays both old and new type values as follows: id new_type old_type ----------- -------- -------- 1 CITY City 2 COUNTY County 172 | SQL Pocket Guide 3 STATE State ... You can use the syntax INSERTED.* or DELETED.* to return all post- and pre-update values respectively. You can specify ex- pressions such as LOWER(INSERTED.type). You can specify col- umn aliases as in any query, with or without the optional AS keyword. Window Functions Window functions enable you to look at different levels of ag- gregation in the same result row. They make it easy to specify cumulative sum, moving average, share-of, and many other important calculations. Window functions are supported in Oracle (where they are known as analytic functions), DB2 (where they are called OLAP functions), SQL Server, and PostgreSQL. Defining a Summary Window The defining role of a window function is to specify a window, or partition of rows, over which the function operates. You specify a window using the OVER ( . . . ) clause, which you can apply to any of the aggregate functions listed in Ta- ble 6 (under “Grouping and Summarizing” on page 52). For example: SELECT u.id, u.county_id, u.northing n1, MIN(u.northing) OVER (PARTITION BY u.county_id) n2, AVG(u.northing) OVER () n3, MAX(u.northing) OVER (PARTITION BY u.open_to_public) n4 FROM upfall u; Window Functions | 173 Each row returned by this query will have the following four northing values: n1 The northing value for the waterfall described by the cur- rent row n2 The lowest northing value of any waterfall in the same county n3 The average northing value of all waterfalls n4 The highest northing value of any waterfall having the same “open to public” status An OVER() clause with nothing between the parentheses sim- ply denotes an aggregate function’s use as a window function. No GROUP BY clause is necessary, and the specified summary value is returned in each detail row. The summary comprises all rows in the result set. To gain the effect of a GROUP BY, add a PARTITION BY clause within the OVER() clause. The PARTITION BY u.county_id in the example partitions detail rows into groups based on their county_id values. The MIN(u.northing) function is applied to each group. Each detail row is then returned with a copy of the MIN(u.northing) function’s result for that row’s group. The example query in this section returns data at a variety of summary levels. First, you have the detail—one row for each fall. Each of those rows then contains data summarized by county, by the entire result set, and by “open to public” status. You can partition by more than one column; for example: MAX(u.northing) OVER ( PARTITION BY u.county_id, u.open_to_public) 174 | SQL Pocket Guide Ordering and Ranking Within a Window You can sort the rows within each partition by placing an ORDER BY clause within the OVER() clause. After you’ve sor- ted the rows, you can rank them in various ways. For example, the following query ranks the northernness of each waterfall in three different ways with respect to other falls in the same county that have the same “open to public” status: SELECT u.id, u.county_id, u.open_to_public, u.name, ROW_NUMBER() OVER ( PARTITION BY u.county_id, u.open_to_public ORDER BY northing DESC) r1, RANK() OVER ( PARTITION BY u.county_id, u.open_to_public ORDER BY northing DESC) r2, DENSE_RANK() OVER ( PARTITION BY u.county_id, u.open_to_public ORDER BY northing DESC) r3, u.northing FROM upfall u WHERE u.northing IS NOT NULL; The following output showing the ranking of publicly accessi- ble waterfalls in Alger County illustrates the three different ranking approaches: ...NAME R1 R2 R3 NORTHING ...--------------- ---- ---- ---- --------- ...Munising Falls 1 1 1 5141184 ...Twin Falls #1 2 2 2 5140500 ...Twin Falls #2 3 2 2 5140500 ...Tannery Falls 4 4 3 5140000 ... The ORDER BY northing DESC clause sorts the rows within each window in descending order by northing. (This sorting is con- ceptual and may be optimized away by your database plat- form.) The three functions, ROW_NUMBER(), RANK(), and DENSE_RANK(), then apply their ranking logic to the rows of each window as follows: Window Functions | 175 ROW_NUMBER() Applies a sequentially increasing number to each row in a window. This is evident in column R1 of the result set. The northernmost row will be number 1, the next northern- most row will be number 2, and so forth. RANK() Returns the same result as ROW_NUMBER(), except that when two rows have the same northing value, they will be given the same rank. This is why Twin Falls #1 and #2 are both ranked in the number 2 position (column R2 in the result set). RANK() will then skip values to ensure that the rank assigned to a given row is always one greater than the number of rows that are ranked lower. For this reason, Tannery Falls is ranked at number 4—it occupies the fourth position. DENSE_RANK() Does not skip values. Compare the results from RANK() in column R2 with those of DENSE_RANK() in column R3. In both cases, Twin Falls #1 and #2 tie for the number 2 position. In the case of DENSE_RANK(), however, Tan- nery Falls is treated as though it occupies the third posi- tion, not the fourth. There is no particular “right way” to rank. Choose the method that delivers the results that work best in your application. NOTE You don’t need to specify a PARTITION BY clause in front of an ORDER BY. For example, RANK() OVER (ORDER BY northing DESC) will rank the current row’s northing against all rows in the query’s result set. Comparing Values Across Rows DB2, Oracle, PostgreSQL, and SQL Server implement some functions that are quite useful for comparing values across row boundaries. LAG and LEAD allow you to look ahead and 176 | SQL Pocket Guide behind a specified number of rows. FIRST_VALUE and LAST_VALUE return values from the first and last rows in a window, respectively. NTH_VALUE (Oracle only) lets you re- turn a value from a specific row in a window by specifying that row’s number. The following query runs in Oracle. Remove the NTH_VALUE invocation to run on the other platforms. SELECT u.id, u.county_id, u.northing n1, FIRST_VALUE(u.northing) OVER ( PARTITION BY u.county_id ORDER BY northing DESC) n2, LAG(u.northing, 1, 9999999) OVER ( PARTITION BY u.county_id ORDER BY northing DESC) n3, LEAD(u.northing, 1, 0) OVER ( PARTITION BY u.county_id ORDER BY northing DESC) n4, LAST_VALUE(u.northing) OVER ( PARTITION BY u.county_id ORDER BY northing DESC) n5, NTH_VALUE(u.northing, 2) IGNORE NULLS OVER ( PARTITION BY u.county_id ORDER BY northing DESC) N6 FROM upfall u; The LAG and LEAD functions each take three arguments. The first argument is a value in which you are interested. The sec- ond argument is a numeric row offset. The third argument is a default value to return when there is no row at the specified offset. The LAG function in this example returns u.northing from the preceding row in the same window (i.e., for the same county) when it is sorted in descending order by northing. Oracle’s NTH_VALUE function takes two arguments: the value to return and the row number from which to retrieve that value. The IGNORE NULLS clause is optional. RESPECT NULLS is the default. You can also specify FROM FIRST or Window Functions | 177 FROM LAST preceding the IGNORE NULLS clause, to specify whether to count from the beginning or the end of the window. NOTE It is not necessary for all window functions in a given query to PARTITION BY or ORDER BY the same set of columns as in this section’s example. The window and sorting criteria can be different in each function. The six northing values returned by the preceding query are as follows: n1 The northing for the waterfall described by the current row. n2 The highest northing of all waterfalls in the same county. n3 The next highest northing for the county, or 9,999,999 if no higher value exists in the same window. LAG sounds as though it should return the next lower northing, but the descending sort turns that around. n4 The next lower northing for the county, or 0 if no lower value exists in the same window. See the entry for n3 for the effect of the descending sort. n5 The lowest northing of all waterfalls in the same county. n6 The northing for the waterfall described by the second row in the window. LAG and LEAD can be very useful, but only when the same offset applies consistently to each row, and only when you have sorted your windows (i.e., the rows in each window) in some meaningful order. 178 | SQL Pocket Guide Summarizing over a Moving Window Within a partition (keeping in mind that the entire rowset can be considered a partition), you can choose to summarize over a moving window of rows. DB2 and Oracle implement this functionality. For example, you might choose to return the MIN and MAX northing values within 1000 meters of each waterfall. Here’s how you might accomplish that: SELECT u.id, u.county_id, u.northing n1, MIN(u.northing) OVER ( ORDER BY u.northing RANGE BETWEEN 1000 PRECEDING AND 1000 FOLLOWING) n2, MAX(u.northing) OVER ( ORDER BY u.northing RANGE BETWEEN 1000 PRECEDING AND 1000 FOLLOWING) n3 FROM upfall u WHERE u.northing IS NOT NULL; In this particular example, there is only one partition—the en- tire rowset. You could easily add PARTITION BY u.county_id to restrict the MIN and MAX computations to each current waterfall’s county. The RANGE BETWEEN clause in this example is considered a framing clause. Framing clause syntax is complex, and it var- ies between vendors. Window Function Evaluation and Placement In the scheme of SQL processing, window functions are among the last elements to be evaluated. They follow any WHERE, GROUP BY, and HAVING clause processing, and they precede ORDER BY. Thus, you can only place window functions in the SELECT list and ORDER BY clauses of a query. Window Functions | 179 Index Symbols " " (quotation marks, double) in case-sensitive identifiers, 127 ' ' (quotation marks, single) denoting empty string, 135 in text literals, 83 * (asterisk) using with SELECT clause, 125 + (plus sign), string concatenation operator in SQL Server, 137 , (comma), separating multicolumn updates in SET clause, 169 \ (backslash) in backreferences, 120 escape character in LIKE predicates, 112 escaping regular expression metacharacters in Oracle, 117 escaping string literals, 84 | | (pipe symbols), string concatenation operator, 137 A ADD clauses in ALTER TABLE statements DB2, 149 MySQL, 150 Oracle, 152 PostgreSQL, 152 SQL Server, 153 aggregate functions, 52 applying to groups of rows with GROUP BY, 53 listing of common functions, 53 aliases for column names in SELECT result sets, 127 ALL keyword EXCEPT ALL queries, 166 versus FIRST in conditional multitable INSERTs, 72 INSERT ALL statements, 71 INTERSECT ALL queries, 167 UNION ALL statements, 164 using in SELECT operations, 129 We’d like to hear your suggestions for improving our indexes. Send email to index@oreilly.com. 181 ALTER clauses in ALTER TABLE statements DB2, 149 PostgreSQL, 152 ALTER COLUMN clauses in ALTER TABLE statements (SQL Server), 153 ALTER SESSION statements in Oracle, 40 ALTER TABLE statements, 149 in DB2, 149 in MySQL, 150 in Oracle, 152 in PostgreSQL, 152 in SQL Server, 153 analytic functions (see window functions) ANSI_NULLS setting (SQL Server), 92 ASCENDING (ASC) and DESCENDING (DESC) sorts, 133 AT TIME ZONE (PostgreSQL), 45 autocommit mode (transactions), 155 B backreferences, 120 BEGIN DISTRIBUTED TRAN[SACTION] statements (SQL Server), 159 BEGIN statements in PostgreSQL, 158 BEGIN TRAN[SACTION] statements (SQL Server), 159 BETWEEN predicates, 111 BIGINT function (DB2), 23 BIGINT type, 15 binary integer types, 15 breadth-first sorting of hierarchical query results, 64 BULK COLLECT keywords in multi-row DELETEs (Oracle), 50 C Cartesian products, 73 cross joins, 74 CASCADE clauses in DROP TABLE statements, 148 case changing for strings, 139 in column and table names, 127 CASE expressions in ORDER BY clause of DELETE statement, 48 searched, 7 simple, 7 using with nulls, 89 CAST function, 8 datetime conversions in SQL Server, 34 CHAR function (DB2), 23 converting between numbers and strings, 94 CHAR option (Oracle), 16 character string data types, 15 CHARINDEX function (SQL Server), 135 CHECK constraints adding and removing in DB2, 150 in MySQL, 151 COALESCE function, 90 using with SELECT statements, 133 code examples from this book, 3 column names case and punctuation in, 127 and data types in union queries, 164 182 | Index qualifying, 128 specifying name or alias for expressions in SELECT list, 127 COMMENT clauses, COMMIT statements (Oracle), 159 COMMIT statements, 159 COMMIT TRANSACTION statements (SQL Server), 160 comparison operators, 109 nulls and, 89 CONCAT function (MySQL), 137 concatenating strings + operator in SQL Server, 137 | | operator, 137 CONNECT BY queries, 8–14 core syntax, 9 joins with, 11 loops in hierarchical data, 13 sorting results, 12 supporting functions and operators, 14 using on data other than hierarchical, 10 WHERE clauses in, 11 CONNECT_BY_ISCYCLE function, 13, 14 CONNECT_BY_ISLEAF function, 14 CONSTRAINT constraint_name syntax, 143 constraints adding and removing in MySQL, 151 adding and removing in Oracle, 152 adding and removing in PostgreSQL, 153 adding and removing in SQL Server, 154 adding in DB2, 149, 150 when creating table in MySQL, 145 naming and declaring, 143 conversions, data type, 15 CAST function, 8 datetime types DATE_FORMAT and TIME_FORMAT functions, 26 in DB2, 21 in Oracle, 28 in MySQL, 24 in PostgreSQL, 31 seconds in day (MySQL), 26 in SQL Server, 34 TO_DAYS and FROM_DAYS functions, 25 Unix timestamps, 25 numeric types in DB2, 93 in MySQL, 95 in Oracle, 95 in PostgreSQL, 97 in SQL Server, 98 CONVERT function datetime conversions in SQL Server, 35 numeric conversions in SQL Server, 98 Coordinated Universal Time (UTC), 32 correlated subqueries, 134, 141 COUNT function, 52 asterisk (*) with, 53 CREATE INDEX statements, 66 CREATE TABLE statements in DB2, 144 in MySQL, 145 in Oracle, 146 in PostgreSQL, 147 in SQL Server, 148 cross joins, 74 CUBE operation, 59 related functions, 60 CYCLE clauses, 65 Index | 183 D data types binary integer, 15 casting, using CAST function, 8 character string, 15 column, in union queries, 164 for columns, 143 converting (see conversions, data type) datetime, 16 decimal, 19 DATE function (DB2), 23 DATENAME and DATEPART functions (SQL Server), 36 datetime data types, 16 conversions in DB2, 21 conversions in MySQL, 24 conversions in Oracle, 28 conversions in PostgreSQL, 31 conversions in SQL Server, 34 conversions of Unix timestamps, 25 conversions using DATE_FORMAT and TIME_FORMAT functions, 26 conversions using TO_DAYS and FROM_DAYS functions, 25 DB2, 16 functions for (see datetime functions) MySQL, 16 Oracle, 17 PostgreSQL, 17 retrieving datetime value elements with EXTRACT function, 51 SQL Server, 18 datetime functions in DB2, 38 in MySQL, 39 in Oracle, 40 in PostgreSQL, 43 getting current date and time, 43 rounding and truncating values, 44 in SQL Server, 45 datetime literals, 85 interval, 86 DATE_FORMAT function, 26 DATE_PART function (PostgreSQL), 45 DAY, MONTH, and YEAR functions (SQL Server), 37 DB2 creating tables, 144 cross joins with LATERAL clause, 74 datetime conversions, 21 datetime functions, 38 datetime types, 16 DECFLOAT type, 19 functions for nulls, 90 modifying tables, 149 numeric type conversions, 93 returning deleted data, 49 returning inserted values, 70 returning updated data, 171 searching strings, LOCATE and POSSTR functions, 135 SELECT…FROM sysibm.sysdummy1 statement, 126 starting transactions, 155 union queries, EXCEPT ALL clause, 166 WITH-clause queries, naming result columns, 143 DECFLOAT function (DB2), 94 DECFLOAT type (DB2), 19 decimal data types, 19 184 | Index DB2, DECFLOAT type, 19 DECIMAL or NUMBER type, 20 DECIMAL function (DB2), 23 converting between numbers and strings, 94 DECIMAL type, 20 DECODE function, 90, 91 DEFAULT keyword in VALUES clause, 68 DELETE statements, 47 deleting all rows, 48 deleting from views and subqueries, 49 deleting in order, 47 double FROM clauses in SQL Server, 51 subqueries in FROM clauses, 140 WHEN NOT MATCHED clause, 88 DELETE WHERE clauses after UPDATEs, 87 DELETED and INSERTED values in SQL Server, 172 deleted data returning in DB2, 49 returning in Oracle, 50 returning in SQL Server, 50 depth, tracking in hierarchy, 63 depth-first sorting of hierarchical query results, 64 DESCENDING (DESC) and ASCENDING (ASC) sorts, 133 deterministic UPDATEs, 171 DISTINCT command, using with COUNT function, 52 DISTINCT keyword using in SELECT operations, 129 dollar-quoted string constants (PostgreSQL), 83 DROP clauses in ALTER TABLE statements DB2, 150 MySQL, 151 Oracle, 152 PostgreSQL, 153 SQL Server, 154 DROP INDEX statements, 67 DROP TABLE statements, 148 durations, 21 labeled, in DB2, 38 E empty strings (' '), 135 ENGINE keyword (MySQL), 145 equi-joins, 76 defined, 78 natural joins, 77 USING clause in, 76 escape characters in SIMILAR TO predicates, PostgreSQL, 118 specifying pattern-matching characters in LIKE predicates, 112 ESCAPE clauses, 112 specifying escape characters in PostgreSQL SIMILAR TO, 118 escape sequences for string literals, 84 example data for this book, 2 EXCEPT (or MINUS) in union queries, 166 EXCEPT ALL clauses in union queries, 166 EXISTS and NOT EXISTS predicates, 110 expressions indexes on, 66 writing for SELECT clauses, 126 EXTRACT function, 45 retrieving specific elements from datetime values, 51 Index | 185 F Finding Flight Legs (case study), 11 FIRST versus ALL in conditional multitable INSERTs, 72 FIRST_VALUE and LAST_VALUE functions, 177 floating-point styles (SQL Server), 98 FORCE clauses (Oracle) in COMMIT statements, 160 in ROLLBACK statements, 161 foreign key constraints dropping in MySQL, 151 in MySQL, 145 for tables being dropped, 148 format elements date, in MySQL, 26 datetime, in Oracle, 29 rounding and truncating DATE values to, 41 datetime, in PostgreSQL, 32 format masks (Oracle), 29 framing clauses, 179 FROM clauses double, in DELETE statements (SQL Server), 51 subqueries in, 140 in UPDATE statements, 171 using in SELECT statements, 130 generating tables with VALUES clause, 131 subqueries in FROM clause, 131 table aliases in FROM clause, 131 WHERE clause, 132 FROM_DAYS function (MySQL), 25 full outer joins, 81 functions datetime conversions in DB2, 21 leading parenthesis in invocation, 1 supporting CONNECT BY queries, 14 G GREATEST function, 52 Gregorian dates, 25 grouping and summarizing, 52– 61 aggregate functions, 52 CUBE operation, 59 GROUP BY clause, 53 grouping before joins, 57 GROUPING SETS function, 60 HAVING clause in GROUP BY queries, 57 listing detail values, 55 reducing GROUP BY list, 56 related functions, 60 ROLLUP operation, 58 GROUPING function, using SQL Server WITH CUBE syntax, 61 H HAVING clauses in GROUP BY queries, 57 hierarchical queries, 62–66 breadth-first versus depth- first sorting, 64 executing using CONNECT BY, 8 recursive WITH, 62 tracking depth in hierarchy, 63 186 | Index I IF function, returning null values in MySQL, 91 IFNULL function (MySQL), 90 IGNORE NULLS clauses, 178 implicit conversions, 15 IN predicates, 110 indexes creating, 66 removing, 67 INITCAP function, 139 inline views (see subqueries) INNER JOIN keywords, 75 inner joins, 75 INSERT statements, 67 multirow inserts, 68 multitable inserts, 71 conditional, 72 unconditional, with INSERT ALL, 71 single-row inserts, 67 subqueries in FROM clauses, 140 subquery inserts, 69 targets of, 68 INSERTED and DELETED values in SQL Server, 172 inserted values returning in DB2, 70 returning in Oracle, 70 returning in SQL Server, 71 inserting data with MERGE statements, 86 INSTR function, 134 INTEGER type, 15 INTERSECT ALL queries, 167 INTERSECT queries, 167 intervals adding and subtracting in MySQL, 39 datetime interval literals, 86 string-based formats in MySQL, 39 IS NOT NULL predicate, 89 IS NULL predicate, 89 IS NULL test, 7 ISNULL function (SQL Server), 92 J join conditions, 75 joins, 72–82 with CONNECT BY queries, 11 cross, 74 explanation of concept, 73 grouping before, 57 inner, 75 natural, 77 non-equi, 78 outer, 79 full outer joins, 81 left outer joins, 79 right outer joins, 80 vendor-specific syntax, 81 USING clause in, 76 JULIAN_DAY function (DB2), 24 L labeled durations (DB2), 38 LAG and LEAD functions, 177 LAST_VALUE and FIRST_VALUE functions, 177 LATERAL clauses (DB2), cross joins with, 74 LCASE and UCASE functions (DB2), 139 LEAST function, 82 left outer joins, 79 interpreting nulls in results, 80 LEN function, 137 LENGTH function, 137 LEVEL operator, 14 LIKE and NOT LIKE predicates, 111 LISTAGG function, 55 literals, 82–86 Index | 187 datetime, 85 datetime interval, 86 numeric, 85 text, 83 LOCAL functions (PostgreSQL), 44 LOCATE function in DB2, 135 in MySQL, 135 loops avoiding with CONNECT BY queries, 10 detecting in recursive data, 65 in hierarchical data, 13 LOWER and UPPER functions, 139 LTRIM and RTRIM functions, 138 M math functions, 99 MAX function, 179 MAXVALUE clauses, 147 MERGE statements, 86–88 WHERE conditions on UPDATE and INSERT operations, 87 MIN function, 174, 179 MINUS clause in union queries (Oracle), 166 MODIFY clauses in ALTER TABLE statements MySQL, 151 Oracle, 152 money styles (SQL Server), 99 MONTH function (SQL Server), 37 MySQL constraints checking and subquery inserts, 69 constraints in, 145 creating tables, 145 datetime conversions, 24 date and time elements, 24 seconds in day, 26 using TO_DAYS and FROM_DAYS functions, 25 datetime functions, 39 datetime types, 16 DATE_FORMAT function, date format elements, 26 deleting data, ORDER BY clause in DELETE statement, 47 escaping an escape character in LIKE predicates, 112 functions for nulls, 90 modifying tables, 150 numeric conversions, 95 regular expressions, 113 regular-expression operators, 113 searching strings, using INSTR and LOCATE functions, 135 starting transactions, 156 string concatenation with CONCAT function, 137 string literal escape sequences, 84 SUBSTRING functions, 137 TEXT type, 15 N natural joins, 77 NOCYCLE, using with CONNECT BY, 13 non-equi-joins, 78 NOT EXISTS and EXISTS predicates, 110 NOT LIKE AND LIKE predicates, 111 188 | Index NOT NULL constraints in DB2, 144 NOW function (PostgreSQL), 44 NTH_VALUE function (Oracle), 177 nullability of a column (DB2), 150 NULLIF function in DB2, 90 in MySQL, 91 in PostgreSQL, 92 in SQL Server, 92 nulls, 88 functions for, in DB2, 90 functions for, in MySQL, 90 functions for, in Oracle, 91 functions for, in PostgreSQL, 92 functions for, in SQL Server, 92 interpreting in outer joins, 80 IS NULL test, 7 predicates for, 89 preventing updating of names to null, 87 using CASE expressions with, 89 using COALESCE function with, 90 NUMBER type, 20 converting to and from VARCHAR2 in Oracle, 96 numeric and math functions, 99 numeric data types conversions in DB2, 93 conversions in MySQL, 95 conversions in Oracle, 95 conversions in PostgreSQL, 97 conversions in SQL Server, 98 numeric format elements in Oracle, 96 in PostgreSQL, 97 numeric literals, 85 NVL function, 90, 91 O OLAP (online analytical processing) functions (see window functions) ON clauses in inner joins, 75 ON DELETE DO INSTEAD rule (PostgreSQL), 49 operators comparison operators, 109 regular expression in MySQL, 113 in Oracle, 116 Perl-influenced, in Oracle, 118 Posix-style, in PostgreSQL, 120 PostgreSQL SIMILAR TO, 118 in SQL Server, 124 supporting CONNECT BY queries, 14 option letters (PostgreSQL Posix- style regular expressions), 123 Oracle alternative quoting delimiters in strings, 83 creating tables, 146 datetime conversions, 28 datetime format elements, 29 datetime elements, 51 datetime functions, 40 getting current date and time, 40 rounding and truncating values, 41 datetime types, 17 depth- or breadth-first sorting, specifying with SEARCH clause, 64 detecting recursive loops with CYCLE clause, 65 Index | 189 DROP TABLE statement, CASCADE CONSTRAINTS clause, 148 functions for nulls, 91 LENGTH functions, 137 LIKE predicates, 112 LISTAGG function, 55 merging data, WHERE conditions on UPDATEs and INSERTs, 87 modifying tables, 152 NTH_VALUE function, 177 numeric conversions, 95 numeric format elements, 96 outer join syntax, 81 pivoting in, 103 regular expressions, 114–118 functions, 114 operators, 116 Perl-influenced operators, 118 regular-expression functions parameters, 115 returning deleted data, 50 returning inserted values, 70 returning updated data, 172 searching strings, INSTR functions, 135 SELECT…FROM dual statement, 126 starting transactions, 156 SUBSTR functions, 136 TRUNCATE TABLE statement, 49 union queries, MINUS clause, 166 unpivoting in, 107 ORDER BY clauses CONNECT BY queries and, 12 in DELETE statements, 47 within OVER clauses, 175 in SELECT statements, 133 in union queries, 164 using with LISTAGG function, 56 window functions in, 179 ORDER SIBLINGS BY clauses sorting CONNECT BY query results, 12 outer joins, 79 full, 81 left, 79 nulls in results, 80 right, 80 vendor-specific syntax, 81 OUTPUT clauses returning deleted data in SQL Server, 50 returning inserted data in SQL Server, 71 returning updated data in SQL Server, 172 OVER clauses, 173 ORDER BY clause within, 175 P PARTITION BY clause within OVER clause, 174 pattern-matching in LIKE and NOT LIKE predicates, 111 Perl-influenced regular expression operators (Oracle), 118 pivot tables, 3 pivoting, 101 explanation of the concept, 102 in Oracle, 103 in SQL Server, 105 POSITION function (PostgreSQL), 135 Posix-style regular expressions (PostgreSQL), 119 operators, 120 option letters, 123 POSSTR function (DB2), 135 190 | Index PostgreSQL alternative quoting delimiters for string literals, 83 column aliases, 127 creating tables, 147 datetime conversions, 31 datetime elements, 51 datetime functions, 43 getting current date and time, 43 rounding and truncating values, 44 datetime types, 17 DELETE statement with view as target, 49 escaping an escape character in LIKE predicates, 112 functions for nulls, 92 modifying tables, 152 numeric type conversions, 97 regular expressions, 118–124 Posix-style, 119 searching strings, using POSITION and STRPOS functions, 135 starting transactions, 158 string concatenation with TEXTCAT function, 138 string literal escape sequences, 84 subquery in FROM clause of a subquery in WITH clause, 141 SUBSTRING functions, 136 union queries, EXCEPT ALL clause, 166 WITH RECURSIVE clause, 62 predicates, 109 BETWEEN, 111 comparison operators available for, 109 EXISTS and NOT EXISTS, 110 for nulls, 89 identifying rows for updates, 168 IN, 110 LIKE and NOT LIKE, 111 REGEXP, 113 SIMILAR TO, 118 primary key constraints dropping in MySQL, 151 PRIOR keyword (in CONNECT BY queries), 10 PRIOR operator, 14 punctuation in column and table names, 127 Q qualifying column names, 128 R RANGE BETWEEN clauses, 179 ranking within a window, 175 recursive loops avoiding with NOCYCLE in Oracle, 13 detecting, 65 recursive queries (see CONNECT BY queries; hierarchical queries) REGEXP predicates, 113 REGEXP_REPLACE function supporting Posix-style regular expressions in PostgreSQL, 120 regular expressions, 113–124 in MySQL, 113 operators, 113 in Oracle, 114–118 in PostgreSQL Posix-style, 119 SIMILAR TO predicate, 118–124 in SQL Server, 124 operators supported, 124 Index | 191 REPLACE function, 135 RESPECT NULLS clauses, 178 RETURNING clauses (Oracle) returning inserted values, 70 returning updated data, 172 right outer joins, 80 ROLLBACK statements, 160 to a savepoint, 162 ROLLBACK TRANSACTION statements (SQL Server), 161 ROLLUP operation related functions, 60 using with GROUP BY, 58 rounding modes, DECFLOAT type, 20 rounding numbers DATE values to datetime elements in Oracle, 41 functions for, 100 ROUND function, 94 row generator, using CONNECT BY as, 10 RTRIM and LTRIM functions, 138 S savepoints (transactions), 161 schema (example) for this book, 2 SEARCH clauses (Oracle), 64 searched CASE expressions, 7 seconds in day, MySQL functions for, 26 SELECT expressions in union queries, 164 EXCEPT (or MINUS) and EXCEPT ALL, 165, 166 INTERSECT and INTERSECT ALL, 167 SELECT statements, 124–134 ALL and DISTINCT keywords, 129 FROM clause, 130 generating tables with VALUES clause, 131 subqueries in, 131, 140 table aliases in, 131 WHERE clause in, 132 FROM OLD TABLE, 49 ORDER BY clause, 133 SELECT clause, 125 case and punctuation in names, 127 listing columns to retrieve, 125 qualifying column names, 128 specifying result set column names, 127 subqueries in, 128 taking shortcuts with asterisk (*), 125 writing expressions, 126 subqueries in WITH clause, 140 in subquery INSERTs, 69 using with UPDATE, 171 window functions in, 179 sequences (Oracle), 146 SET clauses in UPDATE statements comma separating multi- column updates, 169 use of aliases, 171 SET DATEFORMAT command (SQL Server), 34 SET TRANSACTION statements MySQL, 156 Oracle, 156 PostgreSQL, 158 SIMILAR TO predicate (PostgreSQL), 118 regular-expression operators, 119 SMALLINT type, 15 192 | Index special registers returning datetime information in DB2, 38 SQL Server COMMIT TRANSACTION statement, 160 creating tables, 148 datetime conversions, 34 using CAST and SET DATEFORMAT, 34 using CONVERT function, 35 using DATENAME and DATEPART functions, 36 using DAY, MONTH, and YEAR functions, 37 datetime functions, 45 datetime styles, 35 datetime types, 18 DELETE statements, double- FROM clauses in, 51 functions for nulls, 92 modifying tables, 153 numeric type conversions, 98 outer join syntax, 82 pivoting in, 105 regular expressions, 124 returning deleted data, 50 returning inserted data, 71 returning updated data, 172 ROLLBACK TRANSACTION statement, 161 searching strings with CHARINDEX function, 135 starting transactions, 159 string concatenation with + operator, 137 string length, finding with LEN function, 137 SUBSTRING function, 137 unpivoting in, 108 WITH-clause queries, naming result columns, 143 START statements in PostgreSQL, 158 START TRANSACTION statements (MySQL), 156 START WITH clause, CONNECT BY queries, 9 storage engines in MySQL, 145 string functions, 134–139 changing case of a string, 139 concatenating strings, 137 extracting a substring, 136 finding length of a string, 137 replacing text in a string, 135 searching a string, 134 trimming unwanted characters, 138 strings conversions (see conversions, data type) delimiting with single quotation marks, 83 embedding single quote in, 83 STRPOS function (PostgreSQL), 135 subqueries, 139 correlated, sorting results of, 134 deleting from, 49 in FROM clauses, 140 inserts into, 69 new values from, in UPDATE operations, 169 in SELECT lists, 128, 131 updating, 170 in WITH clauses, 140 SUBSTR function, 136 SUBSTRING function in PostgreSQL, 136 PostgreSQL SIMILAR TO regular-expression operators in, 119 Index | 193 supporting Posix-style regular expressions in PostgreSQL, 120 summarizing (see grouping and summarizing) SYSDATE function (Oracle), 40 SYS_CONNECT_BY_PATH function, 14 T table names case and punctuation in, 127 qualifying column names by, 128 specifying name or alias in SELECT FROM clause, 131 tables creating, 143–149 in DB2, 144 in MySQL, 145 in Oracle, 146 in PostgreSQL, 147 in SQL Server, 148 dropping, 148 joining (see joins) modifying, 149–148 in DB2, 149 in MySQL, 150 in Oracle, 152 in PostgreSQL, 152 in SQL Server, 153 tablespaces assigning tables to, in Oracle, 146 indexes on, 66 support in PostgreSQL, 148 text literals, 83 TEXT type (MySQL), 15 TEXTCAT function (PostgreSQL), 138 time zone offsets (SQL Server functions), 46 TIMEOFDAY function (PostgreSQL), 44 timestamps converting to and from, 21 converting to and from character representation in PostgreSQL, 32 Unix timestamp support, 25 TIMEZONE function (PostgreSQL), 45 TO_DAYS function (MySQL), 25 transactions, 154–162 aborting, 160 aborting to a savepoint, 161 autocommit mode, 155 date and time for, in PostgreSQL functions, 44 ending, 159 starting in DB2, 155 starting in MySQL, 156 starting in Oracle, 156 starting in PostgreSQL, 158 starting in SQL Server, 159 TRIM function, 138 TRUNCATE TABLE statements, 48 truncation DATE values in Oracle, 42 datetime values in PostgreSQL, 44 U UCASE and LCASE functions (DB2), 139 union queries, 162 EXCEPT (or MINUS) and EXCEPT ALL, 166 INTERSECT and INTERSECT ALL, 167 names and data types in, 164 ORDER BY clauses, 164 194 | Index order of evaluation with multiple UNION operations, 165 UNION ALL operator, 164 UNION operator, 162 UNIQUE constraints in DB2, 144 dropping in MySQL, 151 Universal Transverse Mercator (UTM) grid coordinates, 2 Unix timestamps conversion functions for, 25 conversions in MySQL, 24 conversions in PostgreSQL, 32 TIMESTAMP type in MySQL, 17 unpivoting explanation of the concept, 106 in Oracle, 107 in SQL Server, 108 UPDATE statements, 168–173 FROM clause, 171 new values generated from subquery, 169 returning updated data in DB2, 171 returning updated data in Oracle, 172 returning updated data in SQL Server, 172 running against views and subqueries, 170 simple updates, 168 subqueries in FROM clauses, 140 updates, using MERGE statements, 86 UPPER and LOWER functions, 139 USING clauses in joins, 76 UTC (Coordinated Universal Time), 32 V VALUES clauses in INSERT statements, 67 multirow inserts, 68 in SELECT FROM statements, generating tables with, 131 VARCHAR type, 15 VARCHAR2 type (Oracle), 16 converting between NUMBER and, 96 VARIABLE command (SQL*Plus), 70 views deleting from, 49 inserts into, 68 updating, 170 W web page for this book, 5 WHEN clauses in CASE expressions, 7 in conditional multitable inserts, 72 WHERE clauses conditions on UPDATEs and INSERTs in MERGE statements, 87 in CONNECT BY queries, 11 in join queries, 75 in SELECT FROM statements, 132 in UPDATE statements, predicates identifying rows to update, 169 window functions, 7, 173–179 comparing values across rows, 177 defining summary window, 173 evaluation and placement, 179 ordering and ranking within a window, 175 Index | 195 summarizing over moving window of rows, 179 WITH clauses in CONNECT BY queries, 8 recursive, 62 further information on, 63 subqueries in, 140 WITH CUBE clause (SQL Server), 59, 61 WITH MARK clauses, BEGIN TRAN statements (SQL Server), 159 WITH ROLLUP clause, using with GROUP BY, 58 WITHIN GROUP keywords using with LISTAGG function, 56 WORK in COMMIT statements, 160 in ROLLBACK statements, 161 Y YEAR function (SQL Server), 37 196 | Index
还剩204页未读

继续阅读

下载pdf到电脑,查找使用更方便

pdf的实际排版效果,会与网站的显示效果略有不同!!

需要 15 金币 [ 分享pdf获得金币 ] 0 人已下载

下载pdf

pdf贡献者

vonezzz

贡献于2011-12-15

下载需要 15 金币 [金币充值 ]
亲,您也可以通过 分享原创pdf 来获得金币奖励!
下载pdf