Bug 76489 – REGRESSION: WebGL doesn't show up in CSS reflections

• Status: RESOLVED
• Resolution: FIXED
• Component: WebGL

[Simon Fraser (smfr)]

The pixel result of compositing/webgl/webgl-reflection.html shows that WebGL doesn't show in CSS reflections.

[Filip Pizlo]

This appears to pass as of r115915

[Simon Fraser (smfr)]

Actually it still fails. The output of NRWT is confusing when pixel testing is not active.

[Simon Fraser (smfr)]

Specifically "Tests expected to fail but passed:" incorrectly lists tests that have expected IMAGE failure, even when pixel testing is not active.

[Antoine Quint]

This isn't just a problem with DRT, this does not work in Safari with a simple test.

[Radar WebKit Bug Importer]

<rdar://problem/26541236>

[Antoine Quint]

Created attachment 280089 [details]
Testcase

The attached test case shows that the background color (blue) of the WebGL <canvas> is reflected but the actual WebGL content (fuchsia).

[Antoine Quint]

I wonder if we should be doing extra work in PlatformCALayerCocoa::clone() so that the contentsLayer of the new newLayer is set. My guess is that the replication layer doesn't have a call to setContentsToPlatformLayer() like the replicated layer does.

[Dean Jackson]

(In reply to comment #7)
> I wonder if we should be doing extra work in PlatformCALayerCocoa::clone()
> so that the contentsLayer of the new newLayer is set. My guess is that the
> replication layer doesn't have a call to setContentsToPlatformLayer() like
> the replicated layer does.

Without looking at the code, I would guess this should work the same as video. I think you're right that it does that via setContentsToPlatformLayer.

[Antoine Quint]

In PlatformCALayerRemoteCustom::clone, I also see this which looks suspicious in order to get reflections working on iOS:

    if (layerType() == LayerTypeWebGLLayer) {
        clonedLayer = adoptNS([[CALayer alloc] init]);
        // FIXME: currently copying WebGL contents breaks the original layer.
        copyContents = false;
    }

[Antoine Quint]

My current understanding of this issue is that reflections are updated to match their replicated layer's content in GraphicsLayerCA::layerDidDisplay where we iterate through the layer clones. There, we check whether the layer that displayed was the layer itself or its content layer. In the case of a <canvas> element using WebGL, we only get notifications for the layer, not the content layer. However, the content layer is the one which has its backing CALayer's contents set to the WebGLLayer's content, so without this notification the code doesn't update the content layer's clones.

I'm not sure why these notifications aren't fired.

As for <video> elements, their reflections work out due to specific work done in PlatformCALayerCocoa::clone to set the AVPlayer of the cloned layer (the replica) to the original layer (the replicated layer).

Not sure yet if we should add WebGL-specific code in PlatformCALayerCocoa::clone or get GraphicsLayerCA::layerDidDisplay to be called with the content layer of a GraphicsLayerCA backing a WebGL <canvas> element. My hunch is the latter, which should trigger once the WebGLLayer has been drawn.

[Antoine Quint]

The problem was indeed the lack of GraphicsLayerCA::layerDidDisplay notification for the content layer. The issue was that the [CALayer display] override in WebGLLayer did not call dispatch the notification the way WebLayer does.

[Antoine Quint]

This doesn't fix iOS though, which is probably due to PlatformCALayerRemoteCustom::clone.

[Antoine Quint]

Created attachment 280357 [details]
Patch

[Antoine Quint]

I raised https://bugs.webkit.org/show_bug.cgi?id=158314 for iOS support.

[Dean Jackson]

Comment on attachment 280357 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=280357&action=review

> Source/WebCore/ChangeLog:12
> +
> +        * platform/graphics/mac/WebGLLayer.mm:

Mention the existing test.

[Antoine Quint]

Created attachment 280430 [details]
Patch for landing

[WebKit Commit Bot]

Comment on attachment 280430 [details]
Patch for landing

Clearing flags on attachment: 280430

Committed r201639: <http://trac.webkit.org/changeset/201639>

[WebKit Commit Bot]

All reviewed patches have been landed.  Closing bug.