You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@subversion.apache.org by Paul Burba <pb...@collab.net> on 2007/09/27 19:35:22 UTC

RE: svn commit: r26803 - in trunk/subversion: libsvn_client tests/cmdline

> -----Original Message-----
> From: kameshj@tigris.org [mailto:kameshj@tigris.org] 
> Sent: Thursday, September 27, 2007 10:25 AM
> To: svn@subversion.tigris.org
> Subject: svn commit: r26803 - in trunk/subversion: 
> libsvn_client tests/cmdline
> 
> Author: kameshj
> Date: Thu Sep 27 07:25:29 2007
> New Revision: 26803
> 
> Log:
> Fix issue 2821.
> Improve merge algorithm used for target trees with child 
> paths which have 
> mergeinfo that differs from their parent.

<snip>

> --- trunk/subversion/libsvn_client/merge.c	(original)
> +++ trunk/subversion/libsvn_client/merge.c	Thu Sep 27 07:25:29 2007

<snip>

> @@ -3566,9 +3732,9 @@
>                                        &initial_rev2,
>                                        ctx, pool));
>  
> -  same_urls = (strcmp(URL1, URL2) == 0);
> +  notify_b.same_urls = (strcmp(URL1, URL2) == 0);
>  
> -  if (!same_urls && record_only)
> +  if (!notify_b.same_urls && record_only)
>      return svn_error_create(SVN_ERR_INCORRECT_PARAMS, NULL,
>                              _("Use of two URLs is not 
> compatible with "
>                                "mergeinfo modification"));
> @@ -3612,7 +3778,7 @@
>       recursive diff-editor thing. */
>    else if (entry->kind == svn_node_dir)
>      {
> -      if (same_urls)
> +      if (notify_b.same_urls)
>          {
>            /* Merge children with differing mergeinfo. */
>            
> SVN_ERR(discover_and_merge_children(&children_with_mergeinfo,
> @@ -3625,26 +3791,27 @@
>                                                depth,
>                                                ignore_ancestry,
>                                                adm_access,
> +                                              &notify_b,
>                                                &merge_cmd_baton,
>                                                pool));
>          }
> -
> -      /* Merge of the actual target.*/
> -      SVN_ERR(do_merge(URL1,
> -                       range.start,
> -                       URL2,
> -                       range.end,
> -                       is_rollback,
> -                       merge_cmd_baton.target_missing_child,
> -                       target_wcpath,
> -                       adm_access,
> -                       depth,
> -                       ignore_ancestry,
> -                       &merge_callbacks,
> -                       &merge_cmd_baton,
> -                       children_with_mergeinfo,
> -                       merge_cmd_baton.existing_mergeinfo ? 0 : -1,
> -                       pool));
> +      else
> +        {
> +          SVN_ERR(do_merge(URL1,
> +                           range.start,
> +                           URL2,
> +                           range.end,
> +                           merge_cmd_baton.target_missing_child,
> +                           target_wcpath,
> +                           adm_access,
> +                           depth,
> +                           ignore_ancestry,
> +                           &merge_callbacks,
> +                           &notify_b,
> +                           &merge_cmd_baton,
> +                           children_with_mergeinfo,
> +                           pool));
> +        }

Hi Kamesh,

Your change to call either discover_and_merge_children or do_merge
allowed my compiler to finally catch this: 

c:\svn\src-trunk\subversion\libsvn_client\merge.c(3813) : warning C4700:
uninitialized local variable 'children_with_mergeinfo' used

This will lead to a segfault in drive_merge_report_editor() if
do_merge() is ever called.  Our test suite never provokes a situation
like this, but it's easy enough to replicate:

Copy some directory DIR to DIR_COPY and commit it at rN
Make a modification to DIR in rN+1
Move DIR to DIR_MOVED in rN+2
Make a modification to DIR_MOVED in rN+3
Merge -rN:N+3 DIR_MOVED DIR_COPY -- and boom!

Anyhow, in r26813 I added an intialization and also tweaked
drive_merge_report_editor() to check for a possible NULL
children_with_mergeinfo (not really necessary right now, but you check
it elsewhere in that function, so in the name of safety and
consistency...

Paul

>        SVN_ERR(cleanup_noop_merge(&merge_cmd_baton, 
> children_with_mergeinfo,
>                                   adm_access, pool));

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org


Re: svn commit: r26803 - in trunk/subversion: libsvn_client tests/cmdline

Posted by Kamesh Jayachandran <ka...@collab.net>.
>> Lieven,
>> Yes I *love* tests that shows detects all the problems in my code well
>> in advance(This is what exactly I feel when I change something in
>> merge-tracking code). But writing a testcase for short lived bugs of
>> this nature would be an overkill.
>>     
> Kamesh,
>
> I understand you're not taking two hours to write a full regression test
> to point out a small issue that can be fixed in 2 minutes, that wasn't
> what I asked for btw.
>
> If you're doing TDD and you'd find such an issue, ideally you'd write a
> small unit test (in C), fix the issue and commit both. This would make
> sure that you pointed out the exact issue and fixed it correctly, with
> only a small overhead in time in the short run.
>
> In my reply I was in fact pointing to the scenario that Paul described,
> which looks like a way a user can trigger a path in our code that isn't
> covered by any regression test for now. In my ideal of having at least
> 100% code coverage this scenario seems like a worthy addition for the
> test suite. No? I just want to make sure we don't forget it. Anyway,
> don't worry about it, I'll add a place holder or maybe a full test to
> the test suite this weekend.
>
>   

Thanks for the explanation Lieven. Yes the case is not covered by our 
testsuite. I stand corrected.

With regards
Kamesh Jayachandran

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org

Re: svn commit: r26803 - in trunk/subversion: libsvn_client tests/cmdline

Posted by Lieven Govaerts <sv...@mobsol.be>.
Kamesh Jayachandran wrote:     
>> So our test suite lacks this test?
>>  
>>> Anyhow, in r26813 I added an intialization and also tweaked
>>> drive_merge_report_editor() to check for a possible NULL
>>> children_with_mergeinfo (not really necessary right now, but you check
>>> it elsewhere in that function, so in the name of safety and
>>> consistency...
>>>       
>> Just wondering why you didn't add that test in r26813? If you don't have
>> time, why not add an XFail-ing place holder explaining in comments the
>> above scenario, then we can still add it later.  
> Lieven,
> Yes I *love* tests that shows detects all the problems in my code well
> in advance(This is what exactly I feel when I change something in
> merge-tracking code). But writing a testcase for short lived bugs of
> this nature would be an overkill.
Kamesh,

I understand you're not taking two hours to write a full regression test
to point out a small issue that can be fixed in 2 minutes, that wasn't
what I asked for btw.

If you're doing TDD and you'd find such an issue, ideally you'd write a
small unit test (in C), fix the issue and commit both. This would make
sure that you pointed out the exact issue and fixed it correctly, with
only a small overhead in time in the short run.

In my reply I was in fact pointing to the scenario that Paul described,
which looks like a way a user can trigger a path in our code that isn't
covered by any regression test for now. In my ideal of having at least
100% code coverage this scenario seems like a worthy addition for the
test suite. No? I just want to make sure we don't forget it. Anyway,
don't worry about it, I'll add a place holder or maybe a full test to
the test suite this weekend.

BTW, thanks for doing a great job to get the merge tracking code
finished on time for 1.5.

Lieven

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org

Re: svn commit: r26803 - in trunk/subversion: libsvn_client tests/cmdline

Posted by Kamesh Jayachandran <ka...@collab.net>.
>>   
>>     
> So our test suite lacks this test?
>   
>> Anyhow, in r26813 I added an intialization and also tweaked
>> drive_merge_report_editor() to check for a possible NULL
>> children_with_mergeinfo (not really necessary right now, but you check
>> it elsewhere in that function, so in the name of safety and
>> consistency...
>>   
>>     
> Just wondering why you didn't add that test in r26813? If you don't have
> time, why not add an XFail-ing place holder explaining in comments the
> above scenario, then we can still add it later.
>
> Lieven
>
>   
Lieven,
Yes I *love* tests that shows detects all the problems in my code well 
in advance(This is what exactly I feel when I change something in 
merge-tracking code). But writing a testcase for short lived bugs of 
this nature would be an overkill.

With regards
Kamesh Jayachandran

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org

Re: svn commit: r26803 - in trunk/subversion: libsvn_client tests/cmdline

Posted by Lieven Govaerts <sv...@mobsol.be>.
Paul Burba wrote:
> ..
> This will lead to a segfault in drive_merge_report_editor() if
> do_merge() is ever called.  Our test suite never provokes a situation
> like this, but it's easy enough to replicate:
>
> Copy some directory DIR to DIR_COPY and commit it at rN
> Make a modification to DIR in rN+1
> Move DIR to DIR_MOVED in rN+2
> Make a modification to DIR_MOVED in rN+3
> Merge -rN:N+3 DIR_MOVED DIR_COPY -- and boom!
>   
So our test suite lacks this test?
> Anyhow, in r26813 I added an intialization and also tweaked
> drive_merge_report_editor() to check for a possible NULL
> children_with_mergeinfo (not really necessary right now, but you check
> it elsewhere in that function, so in the name of safety and
> consistency...
>   
Just wondering why you didn't add that test in r26813? If you don't have
time, why not add an XFail-ing place holder explaining in comments the
above scenario, then we can still add it later.

Lieven

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org

Re: svn commit: r26803 - in trunk/subversion: libsvn_client tests/cmdline

Posted by Kamesh Jayachandran <ka...@collab.net>.
> Hi Kamesh,
>
> Your change to call either discover_and_merge_children or do_merge
> allowed my compiler to finally catch this: 
>
> c:\svn\src-trunk\subversion\libsvn_client\merge.c(3813) : warning C4700:
> uninitialized local variable 'children_with_mergeinfo' used
>
> This will lead to a segfault in drive_merge_report_editor() if
> do_merge() is ever called.  Our test suite never provokes a situation
> like this, but it's easy enough to replicate:
>
> Copy some directory DIR to DIR_COPY and commit it at rN
> Make a modification to DIR in rN+1
> Move DIR to DIR_MOVED in rN+2
> Make a modification to DIR_MOVED in rN+3
> Merge -rN:N+3 DIR_MOVED DIR_COPY -- and boom!
>
> Anyhow, in r26813 I added an intialization and also tweaked
> drive_merge_report_editor() to check for a possible NULL
> children_with_mergeinfo (not really necessary right now, but you check
> it elsewhere in that function, so in the name of safety and
> consistency...
>
> Paul
>
>   

Thanks Paul. Via r26822 addressed Dan's comments and few of my thoughts 
on r26813.

With regards
Kamesh Jayachandran

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org