Re: [Valgrind-developers] Patch: Track allocation times

[Philippe W. <phi...@sk...>]

On Sat, 2013-05-18 at 12:05 +0200, Martin Richtarsky wrote:
> Hi all,
> 
> we use valgrind on some long-running tests consisting of many testcases.
> Currently it is hard to find out what testcase caused a memory leak: A
> developer who fixes a leak basically has to run the complete test to make
> sure he fixed it. It would be helpful to see the allocation time for the
> memory chunk that leaked so it can be corelated to the testcase that ran
> at that time.
At work, we have the same case : a big application, made of several
processes, running several thousands of tests.

To associate bugs and/or leaks to tests in each process,
the test driver instructs each process to do a leak search at
the end of each test. In our case,
this is done using a shell script launching 
   vgdb --pid=<processpid> leak_check full increased
to see the delta of definite and possibleleaks caused by this test.
(this is also doable from your program, using e.g. the client
requests VALGRIND_DO_ADDED_LEAK_CHECK).


> 
> Also, if this patch is welcome, how should I contribute it?
A bug report in bugzilla, with an explanation about why this is needed
(which I am not sure of, cfr approach above, which e.g. does not
have a limitation of 5 allocations :)

Philippe