[pygtk] Problem with the GC starting in the middle of _wrap_gtk_window_list_toplevels

[Hrvoje Niksic]

Margarita Manterola wrote:
> I probably screwd up, because it didn't work.  This is what I was left with:

My code was not meant to be run verbatim, sorry I didn't make that 
clearer.  For example, PyImport_Import takes a PyObject * argument, not 
a C string, use PyImport_ImportModule instead.  This is why you got a 
crash when you tried to use the returned object (likely NULL or garbage).

>     // error checking elided for brevity

Please don't "elide error checking for brevity" in the code you're using 
for actual testing, include it!  If you're not sufficiently familiar 
with Python/C, at least read up sections on reference counting and error 
checking of "Embedding and extending" manual.

>> Maybe you're seeing a weird corner case of
>> http://bugzilla.gnome.org/show_bug.cgi?id=546802 which is known to cause GC
>> to prematurely break cycles which it shouldn't.  I'd try applying the fix
>> for that bug, maybe it helps.
> 
> I applied that patch and it seems to have worked.  No more segfaults,
> no calls to the tooltips_destroy.
> 
> However, I have no idea what the patch does and if the bug is really
> fixed by this, or it's just a coincidence (being that it's a bug
> related to what might happen if GC runs, it's very difficult to
> check).

The bug causes objects reachable only from GTK be erroneously trashed by 
the GC.  The details are hard to understand because they involve 
interaction between two reference-counting systems (those of GObject and 
Python), but that would be the gist of it.  It has bitten me in the past 
and I've reported a duplicate of it.  What you saw could well be caused 
by that bug, but it'd be a good idea to verify it, for starters by 
confirming that the bug is caused in the GC, and then that the fix for 
546802 fixes it.