[CSS Blending] Blending operation is not isolated when z-index on parent changes dynamically

• Component: CSS
• Status: RESOLVED FIXED

[Mihai Tica]

Consider the following scenario:

parent element and child element having -webkit-mix-blend-mode other than normal.
Set z-index and position for the parent from Javascript. Blending is not isolated, ie the blend operation is performed with all the backdrop.
Instead, blending should only be performed with the contents of the parent element.

[Mihai Tica]

Created attachment 225147 [details]
Test case

Repaint test. Place it it fast/repaint

[Mihai Tica]

*** Bug 129305 has been marked as a duplicate of this bug. ***

[Mihai Tica]

Created attachment 225457 [details]
Not for review: reimplemented isolation with logic added in RenderLayerCompositor

This is a proof of concept patch.

--Implementation--
The implementation correctly handles all cases, however, it resides in RenderLayerCompositor.
An alternative implies adding blending as a descendant dependent flag, similar to https://codereview.chromium.org/23503046/diff/59001/Source/core/rendering/RenderLayer.cpp, however, going upwards provides better performance.

--Tests--
The first test test validates that the background of the body is fully painted, and m_isolatesBlending doesn't affect the root renderer.
The second tests validates that isolation is correctly handled. Note that there the third test-case fails due to https://bugs.webkit.org/show_bug.cgi?id=129480. This dependency should be fixed for all cases where RenderLayer.paintsWithTransparency should return true.

[Mihai Tica]

@Simon: could you please have a look over the last patch? Would this approach be ok?
If not, do you have any suggestions?

[Simon Fraser (smfr)]

Comment on attachment 225457 [details]
Not for review: reimplemented isolation with logic added in RenderLayerCompositor

Did you see if this logic fits into RenderLayer::updateDescendantDependentFlags()?

[Mihai Tica]

(In reply to comment #5)
> (From update of attachment 225457 [details])
> Did you see if this logic fits into RenderLayer::updateDescendantDependentFlags()?

I've investigated this and it looks like it would be feasible to use the updateDescendantDependentFlags logic.

Would using this be ok? I'm guessing that hooking here implies no additional traversing, thus being more performant than the implementation from my last patch.

[Mihai Tica]

Created attachment 226015 [details]
First try at isolation with descedantDependentFlag

[Mihai Tica]

Created attachment 226033 [details]
Patch

[Mihai Tica]

Created attachment 226124 [details]
Add guards in RenderLayer.cpp

[Mihai Tica]

@Simon: Could you please review this? Thanks

[Mihai Tica]

Created attachment 227071 [details]
Patch

[WebKit Commit Bot]

Attachment 227071 [details] did not pass style-queue:


ERROR: Source/WebCore/rendering/RenderLayer.cpp:849:  Multi line control clauses should use braces.  [whitespace/braces] [4]
Total errors found: 1 in 9 files


If any of these errors are false positives, please file a bug against check-webkit-style.

[Mihai Tica]

Created attachment 227167 [details]
Patch

[WebKit Commit Bot]

Comment on attachment 227167 [details]
Patch

Clearing flags on attachment: 227167

Committed r165970: <http://trac.webkit.org/changeset/165970>

[WebKit Commit Bot]

All reviewed patches have been landed.  Closing bug.