Need to add a signed-off-by line to the commit message as outlined in https://github.com/ceph/ceph/blob/master/SubmittingPatches

SUCCESS: http://jenkins.ceph.dachary.org/job/ceph/6662/

• SUCCESS http://jenkins.ceph.dachary.org/job/ceph/LABELS=centos-7&&x86_64/6662/
• SUCCESS http://jenkins.ceph.dachary.org/job/ceph/LABELS=ubuntu-14.04&&x86_64/6662/

@Aran85 I understand that there's an issue. I just think that it seems to me that the issue needs to be fixed in the formatter. I don't think that avoiding a call to flush is the way to go as we'll eventually hit this issue again.

dear @yehudasa
i think you are right.i add a condition check when flush.the format is more pretty now :)

FAILURE: http://jenkins.ceph.dachary.org/job/ceph/6718/

• FAILURE http://jenkins.ceph.dachary.org/job/ceph/LABELS=centos-7&&x86_64/6718/

dear @yehudasa
as you said,i think there is no need for flush to output "\n" .but when close section,we need to output the final "\n" when all closeed.i have checked many command's output.hope Ceph & formatter be more pretty!

FAILURE: http://jenkins.ceph.dachary.org/job/ceph/6749/

• FAILURE http://jenkins.ceph.dachary.org/job/ceph/LABELS=centos-7&&x86_64/6749/

FAILURE: http://jenkins.ceph.dachary.org/job/ceph/6776/

• FAILURE http://jenkins.ceph.dachary.org/job/ceph/LABELS=centos-7&&x86_64/6776/

FAILURE: http://jenkins.ceph.dachary.org/job/ceph/6806/

• FAILURE http://jenkins.ceph.dachary.org/job/ceph/LABELS=centos-7&&x86_64/6806/

FAILURE: http://jenkins.ceph.dachary.org/job/ceph/6813/

• FAILURE http://jenkins.ceph.dachary.org/job/ceph/LABELS=centos-7&&x86_64/6813/

SUCCESS: http://jenkins.ceph.dachary.org/job/ceph/6818/

• SUCCESS http://jenkins.ceph.dachary.org/job/ceph/LABELS=centos-7&&x86_64/6818/
• SUCCESS http://jenkins.ceph.dachary.org/job/ceph/LABELS=ubuntu-14.04&&x86_64/6818/

dear @yehudasa
need you to review.

@Aran85 you should probably also squash the commits together

dear @yehudasa
just see the TEST modify which i commited,if no eol when m_stack empty,the "[]" will be connect with the next shell PS1.

SUCCESS: http://jenkins.ceph.dachary.org/job/ceph/6891/

• SUCCESS http://jenkins.ceph.dachary.org/job/ceph/LABELS=centos-7&&x86_64/6891/
• SUCCESS http://jenkins.ceph.dachary.org/job/ceph/LABELS=ubuntu-14.04&&x86_64/6891/

SUCCESS: http://jenkins.ceph.dachary.org/job/ceph/6901/

• SUCCESS http://jenkins.ceph.dachary.org/job/ceph/LABELS=centos-7&&x86_64/6901/
• SUCCESS http://jenkins.ceph.dachary.org/job/ceph/LABELS=ubuntu-14.04&&x86_64/6901/

dear @yehudasa
the eol when section close is just as before,only the useless eol in sections removed .plz check.

@Aran85 I see why it fixes the original issue, but it seems to me that we break other things in the process. From what I understand is that we now switched to appending eol just in the case where the stack is empty, which means that it will only happen to the top level (or am I misreading?). Also, having to update a test to get this change passing doesn't seem quite right.

dear @yehudasa
I think we have three problems here right now.
1.about the eol dumping in json flush method
like the c++ std::flush,the eol in flush is not not necessary.
eol is controlled by print_comma and close section,and it works well.
if somewhere readlly need to split two lines by eol,we could add a endl method also like std::endl.
removing the eol in flush will compact the json format,and i had test it.
2.about the eol appending when stack empty.
i think you are misreaded.the appending eol where the stack is empty means that it will only happen when a sigle whole json had dumped,the next is a new command input line.
3.about the test modify
I had check crush dump_rules usage in OSDMonitor.cc (line.3335) and in crush/CrushWrapper.cc (line.1404),dump_rules is always between open_section and close_section,so the test should use it like which in the process.

dear @yehudasa
example:

1.  {
2.      "data": {                  
3.          "user_id": "aaa",
4.          "display_name": "aaa"
5.      }
6.  } 
7.                              

we could see that the eols at line1 to line5 are common condition.they are dumped by the next line.There are only tow methods which are print_name() and close_section() in the common condithon.
the eol at line 1 is dumped by print_comma which is called by print_name("data",...)
the eol at line 2 is dumped by print_comma which is called by print_name("user_id") .
the eol at line 3 is dumped by print_comma which is called by print_name("display_name")
the eol at line 4 is dumped by close_section (section opened at line 2)
the eol at line 5 is dumped by close_section (section opened at line 1)

the eol at line 6 is dumped by final flush

So most of the time the eols in json are not dumped by flush.only the eol at last line is dumped by flush.

The reason why i append eol in the case where the stack is empty is to replace the eol which is dumped by final flush.

I think there are two options:
1.As @Aran85 's change, flush()'s function is relatively pure, i.e. flush() doesn't have to deal with eol.
2.Keep formatter unchanged. To avoid outputing extra blank lines, we should never use flush() when the buffer is empty.

@yehudasa

@Aran85 sorry for not getting into it earlier. The problem I'm having here is that I'm afraid this will break tests unintentionally, ones that go beyond the rgw use case. Maybe @liewegas can add this one to his set of tests?

hello @liewegas
how about the test?