Patch that "fixes" compositor-x11

[Bill Spitzak]

Kristian Høgsberg wrote:

> Argh, you're right, thanks.  But you said that just adding the count
> and returning that alone didn't fix it, right?  I wonder if we need a
> 
>         xcb_flush(compositor->conn);
> 
> in x11_output_repaint() after the eglSwapBuffers() call.  Maybe just
> before returning from
> x11_compositor_handle_event()... not sure.

I can certainly try inserting that call at various places and see if it 
fixes it.

> If you only handle one
> event at the time in x11_compositor_handle_event(), it doesn't sound
> like a missing flush issue.  Maybe try to force
> x11_compositor_next_event() to always use xcb_poll_for_event(), that
> is, don't use xcb_poll_for_queued_event()?

It is already like this because I don't have xcb_poll_for_queued_event 
according to configure. I also tried removing the #if and it did not 
link so I know I don't have it.

I am guessing the "mask" is what epoll said was happening to the fd, is 
that right? Doesn't that mean we can assume there are no events if READ 
is not on?