It's starting to look like "lui-adjust-undo-list" might be culpable; I
think I've seen that function on the lisp stack each time.