cc: Don't subtract transparent overlay from damage.

cc: Don't subtract transparent overlay from damage.

After processing overlay candidates we used to remove the overlay rects
from the damage rect.
The assumption was that all the overlays would be opaque.

Since we started allowing alpha blended overaly, we should avoid
subtracting from the damage when blending might be allowed.
Otherwise any element animating below a transparent overlay might
not be updated.

We haven't noticed any problem so far since partial update was disabled
on Mali. crrev.com/2829543003 enables empty swap, surfacing the problem
when the damage region is a subset of a transparent overlay.

BUG=705290
CQ_INCLUDE_TRYBOTS=master.tryserver.blink:linux_trusty_blink_rel

Review-Url: https://codereview.chromium.org/2842383003
Cr-Commit-Position: refs/heads/master@{#467727}
(cherry picked from commit 5f26315f43d9d4214df8af3deb263a02fc47db5d)

R=dcastagna@chromium.org

Review-Url: https://codereview.chromium.org/2845193003 .
Cr-Commit-Position: refs/branch-heads/3071@{#273}
Cr-Branched-From: a106f0abbf69dad349d4aaf4bcc4f5d376dd2377-refs/heads/master@{#464641}
Committed: https://chromium.googlesource.com/chromium/src/+/db31db28a02894025e1939362d75aff4665d0b8a

[reveman, 2017-04-27 21:53:58]

reveman@chromium.org changed reviewers:
+ dcastagna@chromium.org

[Daniele Castagna, 2017-04-27 21:54:57]

lgtm

[reveman, 2017-04-27 21:55:27]

Description was changed from

==========
cc: Don't subtract transparent overlay from damage.

After processing overlay candidates we used to remove the overlay rects
from the damage rect.
The assumption was that all the overlays would be opaque.

Since we started allowing alpha blended overaly, we should avoid
subtracting from the damage when blending might be allowed.
Otherwise any element animating below a transparent overlay might
not be updated.

We haven't noticed any problem so far since partial update was disabled
on Mali. crrev.com/2829543003 enables empty swap, surfacing the problem
when the damage region is a subset of a transparent overlay.

BUG=705290
CQ_INCLUDE_TRYBOTS=master.tryserver.blink:linux_trusty_blink_rel

Review-Url: https://codereview.chromium.org/2842383003
Cr-Commit-Position: refs/heads/master@{#467727}
(cherry picked from commit 5f26315f43d9d4214df8af3deb263a02fc47db5d)
==========

to

==========
cc: Don't subtract transparent overlay from damage.

After processing overlay candidates we used to remove the overlay rects
from the damage rect.
The assumption was that all the overlays would be opaque.

Since we started allowing alpha blended overaly, we should avoid
subtracting from the damage when blending might be allowed.
Otherwise any element animating below a transparent overlay might
not be updated.

We haven't noticed any problem so far since partial update was disabled
on Mali. crrev.com/2829543003 enables empty swap, surfacing the problem
when the damage region is a subset of a transparent overlay.

BUG=705290
CQ_INCLUDE_TRYBOTS=master.tryserver.blink:linux_trusty_blink_rel

Review-Url: https://codereview.chromium.org/2842383003
Cr-Commit-Position: refs/heads/master@{#467727}
(cherry picked from commit 5f26315f43d9d4214df8af3deb263a02fc47db5d)

R=dcastagna@chromium.org

Review-Url: https://codereview.chromium.org/2845193003 .
Cr-Commit-Position: refs/branch-heads/3071@{#273}
Cr-Branched-From:
a106f0abbf69dad349d4aaf4bcc4f5d376dd2377-refs/heads/master@{#464641}
Committed:
https://chromium.googlesource.com/chromium/src/+/db31db28a02894025e1939362d75...
==========

[reveman, 2017-04-27 21:55:28]

Committed patchset #1 (id:1) manually as
db31db28a02894025e1939362d75aff4665d0b8a.

[Xiaosong, 2017-07-26 08:54:47]

On 2017/04/27 21:55:28, reveman wrote:
> Committed patchset #1 (id:1) manually as
> db31db28a02894025e1939362d75aff4665d0b8a.

The empty swap
(https://chromium.googlesource.com/chromium/src.git/+/bba53b15c45ce0fa66d3343b...)
is broken if we don't subtract the overlay damage from the root damage rect.

I am trying to verify the empty swap with a web page with video playback inside,
in this test case, the video playback is rendered via overlay and the static
background of web page is rendered via primary plane. then found that
output_frame.sub_buffer_rect is always non-empty because we don't subtract the
overlay damage from the root damage rect.
The TextureDrawQuad generated by video_layer_impl is considered as non-opaque
because the video_layer_impl is not set as contents_opaque.

[Xiaosong, 2017-07-26 08:57:24]

On 2017/07/26 08:54:47, Xiaosong wrote:
> On 2017/04/27 21:55:28, reveman wrote:
> > Committed patchset #1 (id:1) manually as
> > db31db28a02894025e1939362d75aff4665d0b8a.
> 
> The empty swap
>
(https://chromium.googlesource.com/chromium/src.git/+/bba53b15c45ce0fa66d3343b...)
> is broken if we don't subtract the overlay damage from the root damage rect.
> 
> I am trying to verify the empty swap with a web page with video playback
inside,
> in this test case, the video playback is rendered via overlay and the static
> background of web page is rendered via primary plane. then found that
> output_frame.sub_buffer_rect is always non-empty because we don't subtract the
> overlay damage from the root damage rect.
> The TextureDrawQuad generated by video_layer_impl is considered as non-opaque
> because the video_layer_impl is not set as contents_opaque.

reveman@, can you help to explain this? I am not sure whether it's a bug or an
expected behavior.

[reveman, 2017-07-26 13:43:51]

That's the expected behavior for non-opaque overlays today. Could be improved in
the future but requires more advanced tracking of damage.