Bug 77557

Created attachment 124958 [details] Patch

Comment on attachment 124958 [details] Patch Attachment 124958 [details] did not pass chromium-ews (chromium-xvfb): Output: http://queues.webkit.org/results/11387524 New failing tests: compositing/scaling/tiled-layer-recursion.html

FYI, the failure from chromium-ews seems like a flake since I have not been able to reproduce it locally. I'll try again tomorrow though.

Comment on attachment 124958 [details] Patch R=me, let's let the commit queue take a crack at that patch to see if it helps.

Comment on attachment 124958 [details] Patch Rejecting attachment 124958 [details] from commit-queue. New failing tests: compositing/scaling/tiled-layer-recursion.html Full output: http://queues.webkit.org/results/11394983

Ok, I figured out what was up with the failing test. The page scale is affecting the scroll bars and they are getting scaled after being drawn, causing some filtering artifacts. First I thought this caused by some GTK theme weirdness on my machine but that was a red herring. Currently the overflow control layers (scroll bars) are configured appliesPageScale=false. This is a signal to GraphicsLayerChromium to configure the underlying LayerChromium with a content scale to implement the actual page scale. However, since the LayerChromium content scale is only updated when 1) the layer transform changes or 2) the page scale for the layer or a parent of it changes, the content scale never ends up getting set for the overflow controls layer and the scroll bars look as they should. My change was to always apply the content scale for GraphicsLayerChromium, and this broke the scroll bars since they were now scaled. I've now revised the patch to set appliesPageScale=true for overflow controls and made sure changing that flag is properly reflected in LayerChromium and onwards.

Created attachment 126544 [details] Patch

Changed title since this the changes are not limited to Chromium any more. Let me know if you think I should split the patch into two pieces.

Comment on attachment 126544 [details] Patch View in context: https://bugs.webkit.org/attachment.cgi?id=126544&action=review r- because of the added ambiguity. > Source/WebCore/ChangeLog:9 > + 1. Layer contents are scaled by the painter (e.g., non-composited > + content layer) "the painter" is not a term we use in WebKit. I don't really know what you mean by it. > Source/WebCore/ChangeLog:37 > + Reviewed by NOBODY (OOPS!). This should go above the description. > Source/WebCore/platform/graphics/GraphicsLayer.cpp:86 > + , m_needsPageScale(true) Currently layers which are descendants of the layer with m_appliesPageScale=true get scaled. Put another way, the layer with m_appliesPageScale=true is the root of the scale. Now that you've added m_needsPageScale, there is ambiguity. What does it mean for a layer which is ancestral to the m_appliesPageScale=true layer to have m_needsPageScale=true? Maybe we should pull out m_appliesPageScale entirely, and have RLC store which layer is the scale root. Instead of m_needsPageScale, you could then have m_ignoresPageScale, which is set to true for overflow control layers.