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

[Pip Cet]

I think the problem is the call to do_pending_window_change in xg_frame_resized in gtkutil.c: the commit message (commit 3477e27021db) says:

Author:     Martin Rudalics <rudalics@gmx.at>
AuthorDate: Sun Jul 27 15:21:30 2014 +0200
Commit:     Martin Rudalics <rudalics@gmx.at>
CommitDate: Sun Jul 27 15:21:30 2014 +0200

    Complete pixelwise frame/window resizing, add horizontal scrollbar support.
    [...]
    * gtkutil.c (xg_frame_resized): Don't call
    do_pending_window_change.

but the diff is:

@@ -883,6 +884,8 @@ xg_frame_resized (struct frame *f, int pixelwidth, int pixelheight)
       change_frame_size (f, width, height, 0, 1, 0, 1);
       SET_FRAME_GARBAGED (f);
       cancel_mouse_face (f);
+
+      do_pending_window_change (0);
     }
 }

And my current understanding is this bug would not occur if that call were removed. The same issue applies to the change to x_set_window_size, but I'm not certain about removing either call.

On Sun, Aug 30, 2015 at 3:27 PM, Pip Cet <pipcet@gmail.com> wrote:

I forgot to make clear that I verified with gdb that args[0] == Vtimer_list. And if there's anything else you would like me to debug, please let me know. It's very unfortunate I can't reproduce it with emacs -Q and I realize that makes it impossible for you to deal with this bug except through information I provide.

Thanks for trying anyway,

Pip

On Sun, Aug 30, 2015 at 3:24 PM, Pip Cet <pipcet@gmail.com> wrote:

On Sun, Aug 30, 2015 at 3:01 PM, Eli Zaretskii <eliz@gnu.org> wrote:

> Date: Sun, 30 Aug 2015 12:51:26 +0000
> From: Pip Cet <pipcet@gmail.com>
> Somehow, the argument to Fcopy_sequence was changed while concat was
> underway.

How do you see that?

I originally concluded it was the only way to trigger the bug, but I just managed to trigger it again and have it open in a GDB session:

#1  0x00000000005efdb3 in concat (nargs=1, args=0x7fffffff76e8, target_type=Lisp_Cons, last_special=false) at fns.c:747
747                     XSETCAR (tail, elt);
(gdb) p result_len
$22 = 4
(gdb) p debug_print(Flength(args[0]))
5
$23 = void
(gdb)

 

> Further investigation indicates that
> window-configuration-change-hook was called in the middle of concat:

Did you understand how this fact is related to the segfault?

I _think_ I do.

1. concat called with args[0] == Vtimer_list

2. concat stores result_len (=4)

3. concat calls make_list (4)

4. make_list interrupted by QUIT

5. see stack trace

6. window-configuration-change-hook modifies Vtimer_list, which now has length 5

7. control returns to concat

8. concat tries to write 5 elements into a 4-element list, which causes the segfault because `tail' is unexpectedly NULL.

Does that make sense to you?

Thanks,

Pip