Build Identifier: Mozilla/5.0 (Windows NT 6.1; WOW64; rv:2.0b13pre) Gecko/20110314 Firefox/4.0b13pre ID:20110314030412 This was originally reported by http://bugzilla.mozilla.gr.jp/show_bug.cgi?id=6787 Image element with position:fixed is drawn on scroll bar. Reproducible: Always Steps to Reproduce: 1. Start Minefield with new profile 2. Open attached sample html 3. Actual Results: Image element with position:fixed is drawn on scroll bar Expected Results: Image element with position:fixed sgould not be drawn on scroll bar Regression window: Works: http://hg.mozilla.org/mozilla-central/rev/16d3af58499c Mozilla/5.0 (Windows NT 6.1; WOW64; rv:2.0b10pre) Gecko/20110116 Firefox/4.0b10pre ID:20110117133144 Fails: http://hg.mozilla.org/mozilla-central/rev/aa611c72c337 Mozilla/5.0 (Windows NT 6.1; WOW64; rv:2.0b10pre) Gecko/20110115 Firefox/4.0b10pre ID:20110117140637 Pushlog: http://hg.mozilla.org/mozilla-central/pushloghtml?fromchange=16d3af58499c&tochange=aa611c72c337

In local build, build from d41e933f589a : fails build from 16d3af58499c : works Regressed by: d41e933f589a Matt Woodrow — Bug 586683 - Part 4 - Move images into a separate ImageLayer when they are the only item in a ThebesLayer. r=roc a=blocking2.0

roc, can we ship with this?

I think we can ship with this, but should fix it!

Roc: renominate this, please, if you disagree with our assertion, but we didn't think it was bad enough to block. Joe thought it might represent a deeper and worse bug in Layers, though.

I agree we can ship with this. It's almost certainly not a deep layers bug.

The bug is simply that we don't clip a layerized image, at all. Oops. Fortunately it can't escape from the Web page because the page is in a container layer that is clipped.

My reftest for this bug exposed this other layers bug.

Comment on attachment 519524 [details] [diff] [review] Fix MarkLeafLayersCoveredByOpaque >+// aClipRect and aRegoin are in that global coordinate system. aRegoin

Comment on attachment 519525 [details] [diff] [review] Set clip rect on layerized images nsRefPtr<ImageContainer> ContainerState::ThebesLayerData::CanOptimizeImageLayer(LayerManager* aManager) { - if (!mImage) { + if (!mImage || !mImageClip.mRoundedClipRects.IsEmpty()) { return nsnull; } return mImage->GetContainer(aManager); } Should we do the IsRectClippedByRoundedCorner & RemoveRoundedCorners dance here? void ContainerStat if (imageContainer) { nsRefPtr<ImageLayer> imageLayer = CreateOrRecycleImageLayer(); imageLayer->SetContainer(imageContainer); data->mImage->ConfigureLayer(imageLayer); + if (data->mImageClip.mHaveClipRect) { + nsPresContext* presContext = mContainerFrame->PresContext(); + nscoord appUnitsPerDevPixel = presContext->AppUnitsPerDevPixel(); + nsIntRect clip = data->mImageClip.mClipRect.ToNearestPixels(appUnitsPerDevPixel); + imageLayer->IntersectClipRect( + data->mImageClip.mClipRect.ToNearestPixels(appUnitsPerDevPixel)); + } layer = imageLayer; Assert no rounded rects in the clip here?

(In reply to comment #11) > Should we do the IsRectClippedByRoundedCorner & RemoveRoundedCorners dance > here? Let's not bother. When layers can clip to rounded corners, we won't be doing this anyway. > void > ContainerStat > > if (imageContainer) { > nsRefPtr<ImageLayer> imageLayer = CreateOrRecycleImageLayer(); > imageLayer->SetContainer(imageContainer); > data->mImage->ConfigureLayer(imageLayer); > + if (data->mImageClip.mHaveClipRect) { > + nsPresContext* presContext = mContainerFrame->PresContext(); > + nscoord appUnitsPerDevPixel = presContext->AppUnitsPerDevPixel(); > + nsIntRect clip = > data->mImageClip.mClipRect.ToNearestPixels(appUnitsPerDevPixel); > + imageLayer->IntersectClipRect( > + data->mImageClip.mClipRect.ToNearestPixels(appUnitsPerDevPixel)); > + } > layer = imageLayer; > > Assert no rounded rects in the clip here? OK.

Comment on attachment 519524 [details] [diff] [review] Fix MarkLeafLayersCoveredByOpaque >From: Robert O'Callahan <robert@ocallahan.org> >Bug 641770. Ensure that the cliprect for a layer is interpreted in the coordinate system of its container in MarkLeafLayersCoveredByOpaque. r=bas > >diff --git a/gfx/layers/basic/BasicLayers.cpp b/gfx/layers/basic/BasicLayers.cpp >--- a/gfx/layers/basic/BasicLayers.cpp >+++ b/gfx/layers/basic/BasicLayers.cpp >@@ -1237,42 +1237,49 @@ TransformIntRect(nsIntRect& aRect, const > gfxRect gr = gfxRect(aRect.x, aRect.y, aRect.width, aRect.height); > gr = aMatrix.TransformBounds(gr); > aRect = (*aRoundMethod)(gr); > } > > // This implementation assumes that GetEffectiveTransform transforms > // all layers to the same coordinate system. It can't be used as is > // by accelerated layers because of intermediate surfaces. >+// aClipRect and aRegoin are in that global coordinate system. typo nit: aRegion

http://hg.mozilla.org/mozilla-central/rev/2622ae93dfe8 http://hg.mozilla.org/mozilla-central/rev/c96c353c9d5e

Verified with Mozilla/5.0 (Windows NT 6.1; rv:2.2a1pre) Gecko/20110404 Firefox/4.2a1pre