Re: [Valgrind-developers] [PATCH] Add support for GLib memory management functions

[Bart V. A. <bar...@gm...>]

On Thu, Jul 10, 2008 at 8:25 AM, Julian Seward <js...@ac...> wrote:
>
> On Thursday 10 July 2008 08:25, Bart Van Assche wrote:
>> By the way, I noticed that GLib library uses several global variables
>> which are shared over threads and which are not protected by a mutex,
>> e.g. g_threads_got_initialized and g_mem_initialized. While this is
>> correct, it is annoying for tools like Helgrind or DRD. Suppression
>> patterns have to be added for every call stack and every access to any
>> of these global variables to prevent that these appear in race
>> reports.
>
> Can you show an example of how they are used across multiple threads?
> I would have thought that drd should not show any race in such a case,
> if Glib is really correct.

You should know that most but not all data races detected by the
happens-before algorithm are dangerous. E.g. if you analyze the
function g_mem_init_nomessage() in source file glib/gmem.c, you will
see that it is possible that the last eight lines of this function are
executed twice but that this is harmless. I have filed a bug report
for this issue anyway -- see also
http://bugzilla.gnome.org/show_bug.cgi?id=542316.

Bart.