Bug 100263 – [chromium] Enable fixed position compositing on Android DRT

• Component: New Bugs
• Status: RESOLVED FIXED

[Sami Kyöstilä]

[chromium] Enable fixed position compositing on Android DRT

[Sami Kyöstilä]

Created attachment 170419 [details]
Patch

[Sami Kyöstilä]

Having some trouble seeing the effects of this locally, so let's run it through the bots. I did check with debug logs that the appropriate flags are getting set.

[WebKit Review Bot]

Comment on attachment 170419 [details]
Patch

Attachment 170419 [details] did not pass chromium-ews (chromium-xvfb):
Output: http://queues.webkit.org/results/14563296

New failing tests:
compositing/geometry/fixed-position-transform-composited-page-scale-down.html

[Adam Barth]

This is likely a patch for jamesr to review.

[Sami Kyöstilä]

Looks like we don't run Android layout tests on the bots yet. Running locally, I see 3 types of failures:

- Red/blue color channels are swapped.
- Repaint tests don't work with the compositor.
- Some compositor tiles have have different low-order bits, most likely because of inaccuracies in linear interpolation. I think this is expected because GL doesn't guarantee bit-exactness.

I'll try to fix the red/blue issue first.

[Sami Kyöstilä]

Created attachment 170611 [details]
Patch

Fix red/blue color channel issue.

I also checked the tile texel filtering problem on three different devices and got three different results, so I think we'll have to live with it. Same goes for repaint tests, because all of them need to be rewritten to use the new mechanism from bug 97801.

[Peter Beverloo]

Comment on attachment 170611 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=170611&action=review

Cool, thanks Sami! Do we know how many tests are failing after the color adjustments?

> Tools/DumpRenderTree/chromium/TestShell.cpp:685
> +    // in software rendering mode but BGRA if accelerated compositing is active.

Isn't Alexandre working on software rendering (albeit through the compositor)?

[WebKit Review Bot]

Comment on attachment 170611 [details]
Patch

Attachment 170611 [details] did not pass chromium-ews (chromium-xvfb):
Output: http://queues.webkit.org/results/14566587

New failing tests:
compositing/geometry/fixed-position-transform-composited-page-scale-down.html

[Sami Kyöstilä]

(In reply to comment #7)
> (From update of attachment 170611 [details])
> View in context: https://bugs.webkit.org/attachment.cgi?id=170611&action=review
> 
> Cool, thanks Sami! Do we know how many tests are failing after the color adjustments?

I'm still running the tests locally, but from the failures so far I'd estimate about 7100 :)

> > Tools/DumpRenderTree/chromium/TestShell.cpp:685
> > +    // in software rendering mode but BGRA if accelerated compositing is active.
> 
> Isn't Alexandre working on software rendering (albeit through the compositor)?

Yes he is, but at least initially that's not going to be the mode used by Chrome on Android. And as you said, it's still going through the compositor, so we'd need some new mechanism to detect that mode.

Thinking about this a little more, the root issue here is that the compositor readback gives back pixels with the wrong component order. Let me take another pass to figure out why that is happening. Also, I'll see what's up with that failing test on cr-linux.

[WebKit Review Bot]

Comment on attachment 170611 [details]
Patch

Attachment 170611 [details] did not pass chromium-ews (chromium-xvfb):
Output: http://queues.webkit.org/results/14544604

New failing tests:
compositing/geometry/fixed-position-transform-composited-page-scale-down.html

[Sami Kyöstilä]

Created attachment 170653 [details]
Patch

Fixing R/B ordering issue separately in bug 98647. compositing/geometry/fixed-position-transform-composited-page-scale-down.html fails with and without this patch.

[WebKit Review Bot]

Comment on attachment 170653 [details]
Patch

Attachment 170653 [details] did not pass chromium-ews (chromium-xvfb):
Output: http://queues.webkit.org/results/14566692

New failing tests:
compositing/geometry/fixed-position-transform-composited-page-scale-down.html

[Sami Kyöstilä]

Anyone care to review?

[James Robinson]

Comment on attachment 170653 [details]
Patch

Sure

[Peter Beverloo]

Comment on attachment 170653 [details]
Patch

The failing test should be unrelated, the CQ should verify that. Setting CQ+ on request.

[WebKit Review Bot]

Comment on attachment 170653 [details]
Patch

Rejecting attachment 170653 [details] from commit-queue.

New failing tests:
compositing/geometry/fixed-position-transform-composited-page-scale-down.html
Full output: http://queues.webkit.org/results/14623401

[Peter Beverloo]

Comment on attachment 170653 [details]
Patch

Trying again.

[WebKit Review Bot]

Comment on attachment 170653 [details]
Patch

Clearing flags on attachment: 170653

Committed r133607: <http://trac.webkit.org/changeset/133607>

[WebKit Review Bot]

All reviewed patches have been landed.  Closing bug.