GNU bug report logs - #21380
25.0.50; GTK-induced segfault when scheduling timer from window-configuration-change-hook

Previous Next

Full log

Message #107 received at 21380 <at> debbugs.gnu.org (full text, mbox):

From: Eli Zaretskii <eliz <at> gnu.org>
To: Stefan Monnier <monnier <at> IRO.UMontreal.CA>
Cc: rudalics <at> gmx.at, pipcet <at> gmail.com, 21380 <at> debbugs.gnu.org
Subject: Re: bug#21380: 25.0.50;
 GTK-induced segfault when scheduling timer from
 window-configuration-change-hook
Date: Sat, 05 Sep 2015 10:38:53 +0300

> From: Stefan Monnier <monnier <at> IRO.UMontreal.CA>
> Cc: Eli Zaretskii <eliz <at> gnu.org>, Pip Cet <pipcet <at> gmail.com>,
>         21380 <at> debbugs.gnu.org
> Date: Thu, 03 Sep 2015 11:36:44 -0400
> 
> >> So maybe we should introduce a special copy_sequence_no_quit function
> >> that never calls QUIT, and then use it for copying the timer lists.
> 
> That'd be OK, yes.

I believe the currently preferred solution is to block input and
atimers while copying the timer lists to local copies.  Are you okay
with that?

> This said, maybe an even better solution would be to avoid the copy
> altogether.
> 
> AFAICT these lists are only ever side-effected by timer.el's
> timer--activate, which has a special `reuse-cell' argument just to be
> able to do that.
> 
> I'm not completely sure why we do it this way, but my naive
> understanding is the following:
> - For historical reasons of limited resources, timer.el tries hard to
>   avoid allocating cons cells.
> - Then many years later we found a problem with this cell-reuse and
>   circumvented it by copying the whole list all the time.
> - So we end up working hard to avoid allocating a couple cells on one
>   side, only to end up allocating many more on the other.
> 
> Maybe we should go back to bugs #12447 and #12326 and see if just
> removing the "reuse-cell" code (and the Fcopy_sequence(s)) fixes the
> problem as well.

I'm not sure I understand this plan.  Are you saying that consing a
new list in timer--activate, instead of reusing an existing cell, will
avoid the need to wok on a copy of the timer's list when invoking the
timer callbacks?  If so, I'm probably missing something here, because
timer--activate will update the timer list variable anyway, and we
have the same problem, whereby the list changes under our feet, back
again.

IOW, if some Lisp run by a timer callback ends up doing (directly or
indirectly) something like

  (setq timer-list (cons my-new-timer timer-list))

doesn't it mean the value of Vtimer_list in C seen by timer_check
changes as well that very moment?

Not to mention the fact that with timers firing every several tens of
ms, something we've seen while discussing these bugs, allocating a
couple of cells each time might cause a lot of consing per second,
which in turn causes GC, which slows down everything.

Previous Next

GNU bug tracking system
Copyright (C) 1999 Darren O. Benham, 1997,2003 nCipher Corporation Ltd, 1994-97 Ian Jackson.