Changeset 245147 in webkit

Timestamp:

May 9, 2019 11:26:24 AM (5 years ago)

Author:

Simon Fraser

Message:

fast/hidpi/video-controls-in-hidpi.html sometimes asserts in WK1
​https://bugs.webkit.org/show_bug.cgi?id=197695

Reviewed by Zalan Bujtas.

With the backing-sharing changes that landed in r245058, some WebKit1 tests with media controls asserted in
RenderLayerBacking::computeParentGraphicsLayerRect() because a layer would have a m_ancestorClippingLayer,
but backgroundClipRect() would return an infinite rect.

This happened when a layer tree change caused the layer's compositing ancestor to no longer isolate
composited blending (which affects the behavior of RenderLayerCompositor:clippedByAncestor()), but we failed
to mark its composited children as needing the configuration update which would eliminate their m_ancestorClippingLayers.

The fix is to call setChildrenNeedCompositingGeometryUpdate() when isolatesCompositedBlending changes. We don't haev
setChildrenNeedCompositingConfigurationUpdate(), but setChildrenNeedCompositingGeometryUpdate() has the desired side-effect.

I was unable to make a standalone test case for this, but the code is exercised by media control tests.

• rendering/RenderLayerCompositor.cpp:

(WebCore::RenderLayerCompositor::computeCompositingRequirements):

Location:

trunk/Source/WebCore

Files:

• ChangeLog (modified) (1 diff)
• rendering/RenderLayerCompositor.cpp (modified) (1 diff)

trunk/Source/WebCore/ChangeLog

@@ +1 @@
+2019-05-09  Simon Fraser  <simon.fraser@apple.com>
+
+        fast/hidpi/video-controls-in-hidpi.html sometimes asserts in WK1
+        https://bugs.webkit.org/show_bug.cgi?id=197695
+
+        Reviewed by Zalan Bujtas.
+
+        With the backing-sharing changes that landed in r245058, some WebKit1 tests with media controls asserted in
+        RenderLayerBacking::computeParentGraphicsLayerRect() because a layer would have a m_ancestorClippingLayer,
+        but backgroundClipRect() would return an infinite rect.
+        
+        This happened when a layer tree change caused the layer's compositing ancestor to no longer isolate
+        composited blending (which affects the behavior of RenderLayerCompositor:clippedByAncestor()), but we failed
+        to mark its composited children as needing the configuration update which would eliminate their m_ancestorClippingLayers.
+        
+        The fix is to call setChildrenNeedCompositingGeometryUpdate() when isolatesCompositedBlending changes. We don't haev
+        setChildrenNeedCompositingConfigurationUpdate(), but setChildrenNeedCompositingGeometryUpdate() has the desired side-effect.
+
+        I was unable to make a standalone test case for this, but the code is exercised by media control tests.
+
+        * rendering/RenderLayerCompositor.cpp:
+        (WebCore::RenderLayerCompositor::computeCompositingRequirements):
+
 2019-05-09  Brent Fulgham  <bfulgham@apple.com>
 

trunk/Source/WebCore/rendering/RenderLayerCompositor.cpp

@@ -956 +956 @@
 
 #if ENABLE(CSS_COMPOSITING)
+    bool isolatedCompositedBlending = layer.isolatesCompositedBlending();
     layer.setHasNotIsolatedCompositedBlendingDescendants(childState.hasNotIsolatedCompositedBlendingDescendants);
+    if (layer.isolatesCompositedBlending() != isolatedCompositedBlending) {
+        // isolatedCompositedBlending affects the result of clippedByAncestor().
+        layer.setChildrenNeedCompositingGeometryUpdate();
+    }
+
     ASSERT(!layer.hasNotIsolatedCompositedBlendingDescendants() || layer.hasNotIsolatedBlendingDescendants());
 #endif