bug#25279: 26.0.50; Slowdown/crash on certain characters

[Eli Zaretskii <eliz@gnu.org>]

> From: Richard Copley <rcopley@gmail.com>
> Date: Mon, 26 Dec 2016 21:21:57 +0000
> Cc: 25279@debbugs.gnu.org
> 
> > A crash shouldn't happen in any case, so if you can show a recipe and
> > a backtrace, maybe this could be fixed.
> >
> > Thanks.
> 
> Reproducing the slowness and high CPU consumption is easy.
> 
> The "crash" I referred to is probably better termed a "hang" (as
> far as I can see an indefinite one), sorry. I don't know a simple
> recipe, but if you use Emacs in this state for long enough, it
> usually hangs eventually.

Is it a hang, or just some long and very busy loop?  E.g., if you set
garbage-collection-messages to a non-nil value, do you see periodic
announcements of GC during the "hang"?  And if you go for a coffee,
does Emacs eventually recover and starts responding again?  What if
you type M-< or M-> to force the problematic characters out of the
displayed portion of the buffer -- does Emacs recover then?

> Recipe:
> Uninstall Symbola (but perhaps other font changes are required).

It isn't easy to uninstall fonts from a running system, because they
are usually in use by some application.  So reproducing this will be
hard for me, as I have Symbola installed everywhere.  I will try to
find such a system, but no promises.

> Below are C call stacks obtained from a hung emacs process
> by attaching gdb to it.

Is it possible for you to try to figure out what is the looping stack
frame, by following the advice in etc/DEBUG, under "If the symptom of
the bug is that Emacs fails to respond"?

> Possibly a clue:
> While I had the hung emacs process stopped in GDB I started a
> new Emacs process to edit this email and the new Emacs was
> also horribly slow! When I killed the process in the debugger,
> the new Emacs recovered to normal speed.

Attaching a debugger to Emacs built from the master branch causes such
problems due to the low-level keyboard hook in Emacs.  (We avoid that
problem when Emacs is started from GDB to begin with.)  So I don't
think this is relevant to the issue at hand.

Thanks.