I'll try to do a reproducer, but I have a steam game here making gnome-shell abort on app startup most of the times, but not always. It looked like the fullscreening effect had some interaction with a real size change (?), which results on gnome-shell triggering effect completion twice. The gnome-shell side is still pending investigation, I however think mutter should avoid abort()s here.

In this case, I see meta_window_actor_effect_completed() checks for unpaired calls, and warn on the ongoing effect, I see this warning being hit:

g_warning ("Error in size change accounting.");

But right afterwards, the effect_completed() call will still try to thaw the window actor, whose accounting will be also off, and that leads to the more disruptive:

g_error ("Error in freeze/thaw accounting");

I think we should avoid at least the thaw() call if the effect_completed() call itself turned out inconsistent, a warning will already be issued hinting there's something off, so IMO there's no need to go head first to something we know will probably terminate the process.

I'm attaching a patch adding a boolean that is checked before the thaw() call, although perhaps we should just return and avoid the after_effects() call as well?