[cairo] Strange Win32 font failure (was: Access violation in SSE2 pixman code?)

[Damian Frank]

I reported a few days ago on a strange crash I was seeing on win32 in
some pixman compositing code.  I think that was a misleading stack
trace, as I am now seeing different problems with SSE2 turned off.  I
pretty reliably see failures in
_win32_scaled_font_set_world_transform(), but only after a bunch of
rendering and creating and destroying several surfaces.  The failure I
see is that the GDI function SetWorldTransform() is returning failure,
but GetLastError() says everything is okay.  The docs for that
function don't mention why it might fail, unless the HDC is not in
"advanced" graphics mode.  I checked using GetGraphicsMode() and that
does not appear to be the problem.  The transformation looks fine --
just a uniform scale by 0.03125.

So I guess my question is, does this sound familiar to anyone?  Does
anyone have any clues why this might be happening?

I guess my best guess right now is stack smashing somewhere inside
cairo, but that's a very preliminary guess.  (Those are always fun to
find.)

Damian