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

[Pip Cet]

I think that's a good idea, but there's one corner case that, I think, we should think about: a timer deleting the next timer from the list. If we keep the code in timer.el that says:

(defun cancel-timer (timer)
  "Remove TIMER from the list of active timers."
  (timer--check timer)
  (setq timer-list (delq timer timer-list))
  (setq timer-idle-list (delq timer timer-idle-list))
  nil)

then that delq might, in some circumstances, modify the list we're working on. I'm not sure whether it's okay, performance-wise, to replace delq by remq here. Is cancelling one timer from another timer's code something that should ever have well-defined effects (I strongly suspect the answer is "Don't do that")? I confess I do not know whether delq is guaranteed only to modify the list in ways that "make sense" (the current implementation does, but it also doesn't appear to QUIT at all, so maybe it should be modified anyway...).

On Thu, Sep 3, 2015 at 3:36 PM, Stefan Monnier <monnier@iro.umontreal.ca> wrote:

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 think delq also side-effects them. And, of course, "(setq timer-list nil)", which is what I tend to do when I've added a silly timer that does something bizarre and makes Emacs unusable :-)

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 haven't tested this, but I think it would. I still think the underlying problem here is not having a well-defined rule for when QUIT is allowed to call Lisp code. (Eli correctly objected to my initial idea that the rule should be "never", but I still think it should be "much less often than we currently do". I've performed some very boring and somewhat tedious semi-automated call chain analysis to get a better idea of what the current state of things is, and so far the results appear consistent with my idea that QUIT shouldn't call hooks, but should be able to call isolated functions implemented in Lisp, such as those used for relative font sizing).