bug#25875: 26.0.50; Hang logging out of MS-Windows

[Richard Copley <rcopley@gmail.com>]

On 27 February 2017 at 20:27, Eli Zaretskii <eliz@gnu.org> wrote:
>> From: Richard Copley <rcopley@gmail.com>
>> Date: Mon, 27 Feb 2017 19:46:21 +0000
>> Cc: Eli Zaretskii <eliz@gnu.org>, 25875@debbugs.gnu.org
>>
>> > Bug#23483.
>>
>> That's not a real issue, in my opinion. It's already covered,
>> by autosave.
>
> I don't think it is, because when WM_ENDSESSION comes in, Emacs will
> be terminated without giving it a chance to auto-save.
>
> Ken's change was meant to delay the shutdown long enough for Emacs to
> exit in an orderly fashion.  The idea of the design is correct, IMO,
> it's just that we should avoid the hang.

OK. I don't mean to be difficult, I just don't see what testing I can do
that would be of any use.

Eli, you said:

> As I understand it, this happens because when the input thread gets
> the WM_ENDSESSION message, it posts it to the main thread and goes on
> to sleep for 1000 sec, to avoid ending the Emacs process before it
> finishes orderly shutdown.  But if the main thread happens to be
> inside redisplay, it could invoke one of the function that send
> messages to the input thread via SendMessage, which waits for the
> input thread to respond.  So we do have a kind of deadlock.

Posting a message and then sleeping while it's processed is odd,
isn't it? If the input thread /sent/ its message to the main thread,
then while waiting for SendMessage to return, the input thread would
automatically continue to process sent messages