Bug 140177 – Artifacts when absolute positioned element with transform moves positions

• Component: New Bugs
• Status: RESOLVED CONFIGURATION CHANGED

[Carlos Liam]

Created attachment 244155 [details]
Screenshot of the bug

I don't know exact specifics of the bug.

http://aarzee.me/stackinteractive.html

Click on the 'folder' icon and click and drag one of the colored icons (without switching positions with another icon) to reproduce.

If the element is NOT transformed while 'held' the artifacts do not appear.

[Benjamin Poulain]

I can reproduce on ToT. We don't repaint the invalidated area, resulting in stale pixels everywhere.

[Radar WebKit Bug Importer]

<rdar://problem/19402030>

[Simon Fraser (smfr)]

http://trac.webkit.org/log/trunk/?rev=151645&stop_rev=151626

[alan]

Created attachment 244304 [details]
Test reduction.

[Simon Fraser (smfr)]

This happens because we repaint the layer in RenderLayerCompositor::updateBacking() after the style change that starts the transition. So the layer already has the new transform and transform origin.

[Simon Fraser (smfr)]

We can work around this by repainting using the layer's cached repaint rect I think.

[Simon Fraser (smfr)]

Created attachment 244401 [details]
Test with positioned descendant of the moving box

[Simon Fraser (smfr)]

Fix is something like this:

diff --git a/Source/WebCore/rendering/RenderLayer.cpp b/Source/WebCore/rendering/RenderLayer.cpp
index 1e9f4955775fa120f39cca40126296089e043746..8688273887be2d6dba3d741afe925a02071de05b 100644
--- a/Source/WebCore/rendering/RenderLayer.cpp
+++ b/Source/WebCore/rendering/RenderLayer.cpp
@@ -4094,12 +4094,14 @@ void RenderLayer::paintLayerContents(GraphicsContext* context, const LayerPainti
     bool needToAdjustSubpixelQuantization = setupFontSubpixelQuantization(context, didQuantizeFonts);
 
     // Apply clip-path to context.
-    bool hasClipPath = setupClipPath(context, paintingInfo, offsetFromRoot, rootRelativeBounds, rootRelativeBoundsComputed);
+    LayoutSize columnAwareOffsetFromRoot = toLayoutSize(convertToLayerCoords(paintingInfo.rootLayer, LayoutPoint(), AdjustForColumns));
+    bool hasClipPath = setupClipPath(context, paintingInfo, columnAwareOffsetFromRoot, rootRelativeBounds, rootRelativeBoundsComputed);
 
     LayerPaintingInfo localPaintingInfo(paintingInfo);
 
     GraphicsContext* transparencyLayerContext = context;
-    std::unique_ptr<FilterEffectRendererHelper> filterPainter = setupFilters(context, localPaintingInfo, paintFlags, offsetFromRoot, rootRelativeBounds, rootRelativeBoundsComputed);
+
+    std::unique_ptr<FilterEffectRendererHelper> filterPainter = setupFilters(context, localPaintingInfo, paintFlags, columnAwareOffsetFromRoot, rootRelativeBounds, rootRelativeBoundsComputed);
     if (filterPainter) {
         context = filterPainter->filterContext();
         if (context != transparencyLayerContext && haveTransparency) {

[Simon Fraser (smfr)]

Sorry, that was meant for another bug.

[Ahmad Saleem]

@Simon & @Alan - are you able to reproduce this with reductions? I cannot reproduce it on Safari Technology Preview 234.

[alan]

I can't reproduce it anymore.

[Ahmad Saleem]

(In reply to alan from comment #11)
> I can't reproduce it anymore.

Should we close it?

[Simon Fraser (smfr)]

Yes, let's close.