[PATCH] user32: Add a timeout when sending WM_WINE_DESTROYWINDOW

[Alexandre Julliard]

Andrew Eikum <aeikum at codeweavers.com> writes:

> Backtraces of the two threads below. My understanding is the
> DestroyWindow is waiting for the second thread to handle the
> WM_WINE_DESTROYWINDOW, while the second thread is waiting for the
> first thread to call SetEvent.
>
> Windows doesn't send WM_WINE_DESTROYWINDOW, so DestroyWindow exits and
> the test calls SetEvent.
>
> There might be a race condition in the test, if the first thread sends
> WM_WINE_DESTROYWINDOW before the PostQuitMessage's WM_QUIT causes the
> second thread's event loop to exit. However, it deadlocks every time
> for me.

I don't think we want to simply ignore the failure and keep child
windows alive when their parent is gone. Probably the sequence could be
changed so that WM_WINE_DESTROYWINDOW is sent first, and WM_DESTROY is
sent in the context of the target thread.

-- 
Alexandre Julliard
julliard at winehq.org