[Piglit] New test for partial blits after a buffer swap

[Federico Mena Quintero]

On Wed, 2013-10-30 at 07:00 -0700, Paul Berry wrote:
> On 29 October 2013 14:46, Eric Anholt <eric at anholt.net> wrote:

>         
>            /* FIXME: Does EGL support front buffer rendering at all?
>         */
> 
> 
> FYI, we discussed this issue back in May but didn't really come to a
> clear consensus of what we wanted to do about it:
> http://lists.freedesktop.org/archives/mesa-dev/2013-May/040059.html.
> (Note that the replies happened in July so the mesa-dev archives don't
> thread them well.  See
> http://lists.freedesktop.org/archives/mesa-dev/2013-June/subject.html#40129 for the replies.)

Nice; thanks for digging this up.  I'm glad to see that it's a known
problem.

... Except that in the mutter/clutter/cogl triumvirate, EGL is not
used :)  They run through GLX as far as I can tell.

AFAIK, clutter/cogl like to do a blit from the back to the front buffer
(or use the CopySubBuffer extension if available) to avoid a full frame
repaint.  This helps when running the desktop shell in a VM (without
acceleration), or over the network, or over some VNC-like thing (or on a
funny platform like IBM mainframes with no acceleration).

I hope that the above can answer the "is this desirable" questions from
the thread you linked.  I can live with either the blits being fixed, or
with CopySubBuffer being added for the software rendering case.

I'll try to see if patching the code to copy the fake front to the real
front is sufficient.  I don't really know the right place to do this,
but let's see.

  Federico