Re: winex11.drv: Allow grabed window belonging to a different thread in send_mouse_input.

[Henri Verbeet]

On 8 March 2016 at 17:01, Jacek Caban  wrote:
> Signed-off-by: Jacek Caban
>
> If we call XGrabPointer on a window belonging to a different thread, we
> may end up handling its events with different display and being unable
> to associate X11 window with HWND.

Is calling SetCapture() on a window from a different thread really
supposed to be valid? Assuming it is, wouldn't it then be easier to
grab the window from the correct thread instead? (Similar to how e.g.
WM_X11DRV_CLIP_CURSOR works.)

winex11.drv: Allow grabed window belonging to a different thread in send_mouse_input.

[Jacek Caban]

Hi Henri,

On 03/09/16 16:02, Henri Verbeet wrote:
> On 8 March 2016 at 17:01, Jacek Caban  wrote:
> > Signed-off-by: Jacek Caban
> >
> > If we call XGrabPointer on a window belonging to a different thread, we
> > may end up handling its events with different display and being unable
> > to associate X11 window with HWND.
>
> Is calling SetCapture() on a window from a different thread really
> supposed to be valid?

Yes, I think so. Also note that the HWND we're capturing is not
necessary the same as X11 window we grab (it may be its child belonging
to a different thread than its parent), so Windows restrictions don't
apply here.
> Assuming it is, wouldn't it then be easier to
> grab the window from the correct thread instead? (Similar to how e.g.
> WM_X11DRV_CLIP_CURSOR works.)

Yes, we could do that, although I'm not sure this sort of complication
is the right thing in this case. I thought more about this and I think
that another solution will be better.

I should add that even if we call XGrabPointer from a different thread,
we don't necessary end up receiving events in a different thread (or a
different connection to be precise). According to my testing, in most
cases we still receive events in the thread owning the window. There
are, however, some cases when we don't. I can reproduce receiving in a
different thread with my tests, but even then it's not consistent. When
I tried to trace X11 calls, it started to behave differently and
received events in owning thread. It sounds like some sort of race on
X11side. My previous patch fixed it by accepting both behaviours in
Wine. Instead, we could try harder to get a consistent behaviour from X
server. I will send a patch that achieves that.

Thanks,
Jacek