On Sun, Dec 13, 2015 at 7:27 AM Eli Zaretskii <eliz@gnu.org> wrote:

Looks like another reincarnation of bug#21667.

http://debbugs.gnu.org/cgi/bugreport.cgi?bug=21667 appears to be unrelated to this one. I suspect you made a typo.

Can you modify lui-adjust-undo-list so that GC is inhibited (by
binding gc-cons-threshold to most-positive-fixnum around the whole
function)?

OK, I've done that. If that works, it'll take days of use before I'm confident that it did work (because the bug only appears once every few days); can you think of some sort of debug message I can stick in there that can tell us that you've correctly guessed the cause?