[Wayland-bugs] [Bug 72612] weston-subsurfaces does not render/display

[bugzilla-daemon at freedesktop.org]

https://bugs.freedesktop.org/show_bug.cgi?id=72612

--- Comment #6 from Neil Roberts <neil at linux.intel.com> ---
> There should not be a secret cairo EGL surface for the triangle widget. If
> there is, I think that should be fixed first. I guess it just hasn't been an
> issue until now.

I posted a patch to the mailing list to fix this:

http://lists.freedesktop.org/archives/wayland-devel/2013-December/012650.html

However it turns out I was wrong and that doesn't fix the problem. The problem
still happens even if you pass -n to disable the triangle surface.

The sequence of drawing without the triangle surface goes like this:

1. draw main surface
2. draw subsurface
3. swap subsurface
4. swap main surface

As mentioned in comment #1, the first thing Cairo does when flushing (AKA
swapping) the surface is to ‘acquire’ the GL context. After acquiring the
context it will bind the context again with the EGL surface for the Cairo
surface that is to be swapped. Acquiring the context binds whatever surface
Cairo used last. At step 4 the last surface that was bound is the subsurface.
That has a frame callback installed following the swap in step 3 which means
that when eglMakeCurrent is called it will block in get_back_bo. This will
block until the main surface is swapped which will never happen so it blocks
forever.

> I think I have assumed, that eglSwapInterval(0) would cause the EGL
> implementation to never stall indefinitely, but is that a feasible
> assumption?

Yes, that is correct. With eglSwapInterval(0) Mesa should only ever block on a
sync request and there should be no reason why that wouldn't come instantly.
This problem isn't really related to the eglSwapInterval(0) changes but instead
it is triggered by blocking on the frame callback in get_back_bo instead of in
eglSwapBuffers. That was a separate change independent of eglSwapInterval to
fix the 3 buffers problem.

> Btw. how about --with-cairo=image?

I haven't tested it but I assume using the image backend would fix the problem
because that doesn't have the mess with rebinding the EGL context.

I think the best fix would be to make Mesa not call get_back_bo from
eglMakeCurrent. The patch that added this call is this one:

http://cgit.freedesktop.org/mesa/mesa/commit/?h=db9c151d

I tried recklessly removing the call to intel_prepare_render to see if the
situation has changed since 2010 but it breaks quite catastrophically. It looks
like making this change would be a fair bit of work.

-- 
You are receiving this mail because:
You are the assignee for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.freedesktop.org/archives/wayland-bugs/attachments/20131219/5c6016f2/attachment.html>