Andy Piper <andy(a)xemacs.org&gt; writes: Well I now put the hard core check I wrote about in my previous messages in mark_window_as_deleted and sure the check hits there. (The Fdelete_window is call is part of delete_other window). The image_instance is involved is an IMAGE_WIDGET with xpos = ypos = 0 and which is very wide and not so high. To me that seems like the tab widget. I had a look at image_instantiate if (UNBOUNDP (instance)) { Lisp_Object locative = noseeum_cons (Qnil, noseeum_cons (pointerp ? ls3 : instantiator, DEVICEP (governing_domain) ? subtable : XWINDOW (governing_domain) ->subwindow_instance_cache)); int speccount = specpdl_depth (); record_unwind_protect (image_instantiate_cache_result, locative); instance = instantiate_image_instantiator (governing_domain,domain,...) cache_subwindow_instance_in_frame_maybe (instance); Fsetcar (locative, instance); unbind_to (speccount, Qnil); } Two remarks 1. Shouldn't locative and/or governingdomain be GCRPO'ed?, instantiate_image_instantiator gcpro's its variables. 2. Isn't it possible that DEVICEP(governing_domain) is true, but domain is a window?. In that case cache_subwindow_instance_in_frame_maybe might add it to the frame cache but it won't be in the window cache, so mark_window_as_deleted does not see it? Jan