This is a regression since we're calculating the layers' visible rects based on their transform, while ignoring that the transform might change while animating.
Created attachment 123669 [details] Patch
Created attachment 123670 [details] Patch
Attachment 123669 [details] did not pass style-queue: Failed to run "['Tools/Scripts/check-webkit-style', '--diff-files', u'Source/WebKit2/ChangeLog', u'Source/WebKit..." exit_code: 1 Source/WebKit2/WebProcess/WebCoreSupport/WebGraphicsLayer.h:38: wtf includes should be <wtf/file.h> instead of "wtf/file.h". [build/include] [4] Source/WebKit2/WebProcess/WebCoreSupport/WebGraphicsLayer.h:38: Alphabetical sorting problem. [build/include_order] [4] Total errors found: 2 in 3 files If any of these errors are false positives, please file a bug against check-webkit-style.
Created attachment 124012 [details] Patch
Comment on attachment 124012 [details] Patch View in context: https://bugs.webkit.org/attachment.cgi?id=124012&action=review Seems ok to me > Source/WebKit2/WebProcess/WebCoreSupport/WebGraphicsLayer.cpp:551 > + // If this layer is part of an active transform animation, the visible rect might change, so we rather render the whole layer > + // until some better optimization is available. I would break the line after the second ,. Then they will have almost the same size > Source/WebKit2/WebProcess/WebCoreSupport/WebGraphicsLayer.cpp:553 > + if (selfOrAncestorHasActiveTransformAnimations()) > + return tiledBackingStoreContentsRect(); Did you do some measurements of this? regarding to the comment in the changelog > Source/WebKit2/WebProcess/WebCoreSupport/WebGraphicsLayer.cpp:725 > + if (!parent()) > + return false; > + return toWebGraphicsLayer(parent())->selfOrAncestorHasActiveTransformAnimations(); wouldnt it read nicer if (parent()) return toWeb....parent()... return false;
> > Source/WebKit2/WebProcess/WebCoreSupport/WebGraphicsLayer.cpp:553 > > + if (selfOrAncestorHasActiveTransformAnimations()) > > + return tiledBackingStoreContentsRect(); > > Did you do some measurements of this? regarding to the comment in the changelog No... if I had a viable alternative I'd measure it vs. this solution :) Problem is that if I'd calculate the visible rect based on all the transform keyframes, I could still reach a point where a huge layer is entirely visible. What we should probably do in that case is not have content-scale for huge transformed layers, but that should come later as it's tricky.
Created attachment 124015 [details] Patch
Created attachment 124017 [details] Patch
Comment on attachment 124017 [details] Patch Clearing flags on attachment: 124017 Committed r105934: <http://trac.webkit.org/changeset/105934>
All reviewed patches have been landed. Closing bug.