bug#12832: 24.3.50; Emacs lockup when idle

[Lennart Borgman]

I have no time to build Emacs now and have not had it for a very long
time. However I have seen Emacs freeze a lot of times when it is idle.
This started to get more common when I moved to a 64-bit windows 7
(from windows xp, 32-bit).

I have always thought that this must be bad handling of the messages
from the operating systems, but I have had no chance to pin this down.
Reading here I wonder if data are accessed somewhere in a system time
thread without critical section handling.

On Wed, Nov 14, 2012 at 5:29 PM, Andy Moreton <andrewjmoreton@gmail.com> wrote:
> On 14/11/2012 12:44, Andy Moreton wrote:
>>
>> On 13/11/2012 17:04, Eli Zaretskii wrote:
>>>
>>> Can you please try the patch below and see if it prevents the
>>> lock-ups?
>>>
>>> === modified file 'src/w32proc.c'
>>> --- src/w32proc.c       2012-11-05 03:18:32 +0000
>>> +++ src/w32proc.c       2012-11-13 16:59:53 +0000
>>> @@ -431,13 +431,24 @@ timer_loop (LPVOID arg)
>>>           /* Simulate a signal delivered to the thread which installed
>>>              the timer, by suspending that thread while the handler
>>>              runs.  */
>>> -         DWORD result = SuspendThread (itimer->caller_thread);
>>> +         DWORD result;
>>> +
>>> +         if (dwMainThreadId)
>>> +           enter_crit ();
>>> +         result = SuspendThread (itimer->caller_thread);
>>>
>>>           if (result == (DWORD)-1)
>>> -           return 2;
>>> +           {
>>> +             if (dwMainThreadId)
>>> +               leave_crit ();
>>> +             return 2;
>>> +           }
>>>
>>>           handler (sig);
>>>           ResumeThread (itimer->caller_thread);
>>> +
>>> +         if (dwMainThreadId)
>>> +           leave_crit ();
>>>         }
>>>
>>>         /* Update expiration time and loop.  */
>>>
>>
>> I applied this to emacs-24 branch (r110866) this morning. So far I've not
>> seen
>> a lockup, but I'll to run it for a day or two to be sure.
>
>
> After longer uptime, it seems this patch is not successful. I haven't had a
> complete lockup, but I have seen a couple of glitches where it froze but
> then recovered a short while later.
>
> The unfreeze may have been due to capturing a stack trace with Process
> Explorer (I have upgraded to the latest version which is less buggy).
>
> The patched emacs-24 does seem to leak handles: at the moment Process
> Explorer report that emacs has 50805 handles in all, most of which are
> thread handles. The number of handles seems to increase at a rate of 2 to 4
> per second.
>
>     AndyM
>
>
>
>
>
>
>