Re: Valgrind and GTK libraries

[Dan Kegel <dank kegel com>]

On Sat, Jan 2, 2010 at 4:20 AM,  <jcupitt gmail com> wrote:
> 2010/1/1 Erik de Castro Lopo <mle+gtk mega-nerd com>:
> >  - How does one go about teasing out the difference between my memory
> >   leaks and GTK stuff I have no control over?
>
> I usually run my app twice: once with just startup/shutdown and once
> with startup/run for a while/shutdown. The diff between the two runs
> (with a little trimming) shows interesting memleaks.
>
> Also, with experience it becomes easy to spot the gtk things that
> aren't leaks. Anything related to type init is not really a leak, for
> example.

Yeah.   For comparison, you can see what suppressions are firing on
Chromium's valgrind runs.  See e.g.
http://build.chromium.org/buildbot/waterfall/builders/Linux%20Tests%20(valgrind)(2)/builds/2588/steps/valgrind%20test:%20ui/logs/stdio
which says how often the following six suppressions fire during our ui tests:

     47 http://sources.redhat.com/bugzilla/show_bug.cgi?id=5171
     59 glibc leak.  See also
http://sources.redhat.com/bugzilla/show_bug.cgi?id=2451
   1176 pango_font_leak_todo
  12161 Fontconfig leak?
  12338 gtk developers don't like cleaning up one-time leaks.  See
http://mail.gnome.org/archives/gtk-devel-list/2004-April/msg00230.html
  19086 dl-hack3-cond-1

Our suppressions file is at
http://src.chromium.org/cgi-bin/gitweb.cgi?p=chromium.git;a=blob_plain;f=tools/valgrind/memcheck/suppressions.txt

I imagine some of those would go away if we were
able to update to the latest versions of things like glibc,
fontconfig, gtk, and pango, but we have to live with
whatever distro the test bot admins picked (which
is probably hardy or something like that).
- Dan