Stefan, any chance this may have been introduced by this commit of yours?

commit 7b13aeee329834a29642360f958f27a6ef23bcf7
Author: Stefan Kost <ensonic@users.sf.net>
Date:   Tue Feb 16 12:06:08 2010 +0200

    x(v)imagesink: take new size from event thread and do not poll for every frame
    
    We can update the geometry in ConfigureNotify (unless we disable event-
    handling). If event handling is disabled, one should use _expose() to trigger a
    redraw and update the geometry.

A backtrace would be nice. gst_ximagesink_xwindow_update_geometry is called from two places:

gst_ximagesink_expose()

gst_ximagesink_handle_xevents()
here it is called within
  while (XCheckWindowEvent (ximagesink->xcontext->disp,
          ximagesink->xwindow->win, ExposureMask | StructureNotifyMask, &e)) {

so the assert would trigger earlier. But then the check is a bit stupid nowadays. From the two places where we call with we pass ximagesink->xwindow as xwindow and inside gst_ximagesink_xwindow_update_geometry() deref ximagesink->xwindow again.

So the assert should either just check ximagesink->xwindow and we remove the extra parameter or we use the passed parameter.

The former change is already in xvimagesink. I attach a patch in a sec.

Created attachment 158789 [details] [review]
refactor _update_goemetry()

This could come from gst_ximagesink_expose(). We call this also from the event thread (both ximagesink and xvimagesink). The expose impl look like this:

  gst_ximagesink_xwindow_update_geometry (ximagesink);
  gst_ximagesink_ximage_put (ximagesink, NULL);

where _put is handling ximagesink>xwindow == NULL.

Simmilar we could drop the assert from _update_geometry() and add this check.

  if (G_UNLIKELY (ximagesink->xwindow == NULL)) {
    g_mutex_unlock (ximagesink->flow_lock);
    return;
  }

Stack trace:

(gdb) bt

+ Trace 221382

It seems fairly easy to produce if you start totem and press the 'F' key for fullscreen as soon as it comes up.

Created attachment 158798 [details] [review]
gracefully handle ximagesink>xwindow == NULL

Comment on attachment 158798 [details] [review]
gracefully handle ximagesink>xwindow == NULL

Thanks, these patches fix the problem for me and look pretty harmless to me.