Re: GTK+ 2.2.1 crashes

[Owen Taylor <otaylor redhat com>]

On Wed, 2003-05-14 at 20:23, Hans Petter Jansson wrote:
> On Wed, 2003-05-14 at 16:38, Owen Taylor wrote:
> > On Wed, 2003-05-14 at 16:06, Hans Petter Jansson wrote:
> 
> > It wasn't clear to me what you mean by "apparently been fixed on the 2.2
> > branch in CVS".  I guess you mean "I tested it with 2.2 and it didn't
> > happen".
> > 
> > It's also not clear what you mean to me by "not in HEAD". Do you mean
> > you tested it with HEAD and it was broken or do you mean that you
> > didn't test it with HEAD?
> > 
> > We have a policy that all bug fixes in 2.2 go *immediately* into HEAD.
> 
> What I meant was "not fixed in HEAD". Here's how it looks:
> 
> GTK_2_2_1: Crash.
> HEAD:      Crash.
> gtk-2-2:   No crash.
> 
> With checkouts made today and the test case I attached earlier.
> GTK_2_2_1 and HEAD yield the same stack traces. gtk-2-2 doesn't exhibit
> any misbehaviour.

Weird. There are *no* code differences between gtk-2-2 and HEAD in
gdk/x11 and basically none in gdk/.

Maybe the bug is actually in GtkTreeView; there are changes to the 
rendering code, related, I think related to the adding of expansion
settings in HEAD.

> > It could be serious, since it's memory corrupting and appears to happen
> > > in a memcpy () somewhere under gdk_draw_pixbuf () [the rest of the stack
> > > trace seems imprecise]. Most of the time it doesn't crash, it just
> > > writes outside its buffer. That code has undergone a revamp on the 2.2
> > > branch, and I guess that's why the bad behaviour went away, although I
> > > don't see a reference to the problem in the ChangeLog.
> 
> > I don't remember any changes to gdk_draw_pixbuf() in 2.2 that aren't in
> > HEAD; can you give a particular reference to what you are talking about?
> 
> I just saw there were a bunch of changes in gdk/x11/ (where the only
> likely memcpy()s were) from GTK_2_2_1 to gtk-2-2. I have a better stack
> trace now, which shows that the memory violation happens in
> convert_to_format() at gdkdrawable-x11.c:1167 (today's HEAD). That file
> is identical in gtk-2-2 and HEAD, so I was probably wrong and the
> problem lies somewhere else (see attached full trace).
> 
> Although 2.2 might have the bug too; it might just not show under the
> same circumstances.
> 
> Sorry for not coming up with better information, or a patch. I'm not
> familiar with this code, and the bug doesn't look trivial. I'm still
> looking at it, though.

Well, file a bug, and hopefully the treeview whizzes will be able to
figure it out.

Regards,
                                               Owen