For fallback of nsDisplayFilter, currently we always repaint the whole display list inside the nsDisplayFilter in each transaction. There are two problems here. One is we shouldn't repaint all. The other one is that the 'reftest-wait' mechanism couldn't work without the invalidation info. We should address the two problems in this bug. A related testcase is "layout/reftests/bugs/796847-1.svg".

Comment on attachment 8908456 [details] Bug 1393376 - (Part 1) Paint the item when its type is TYPE_FILTER during WebRenderFallbackData generating; https://reviewboard.mozilla.org/r/180102/#review187278 ::: gfx/layers/wr/WebRenderLayerManager.cpp:551 (Diff revision 2) > PixelCastJustification::WebRenderHasUnitResolution); > > - nsRegion invalidRegion; > + bool needPaint = true; > nsAutoPtr<nsDisplayItemGeometry> geometry = fallbackData->GetGeometry(); > > - if (geometry) { > + if (geometry && !fallbackData->IsInvalid() && aItem->GetType() != DisplayItemType::TYPE_FILTER) { Please add a comment about what the special case for TYPE_FILTER is for. ::: gfx/layers/wr/WebRenderLayerManager.cpp:570 (Diff revision 2) > if (!lastBounds.IsEqualInterior(clippedBounds)) { > invalidRegion.OrWith(lastBounds); > invalidRegion.OrWith(clippedBounds); > } > } > + needPaint = (invalidRegion.IsEmpty())? false: true; This can just be needPaint = !invalidRegion.IsEmpty();

Comment on attachment 8908457 [details] Bug 1393376 - (Part 2) Retain BasicLayerManager in WebRenderFallbackData for layer invalidation computation; https://reviewboard.mozilla.org/r/180104/#review187280

Pushed by ryanvm@gmail.com: https://hg.mozilla.org/integration/autoland/rev/94c2ea91c36f (Part 1) Paint the item when its type is TYPE_FILTER during WebRenderFallbackData generating; r=jrmuizel https://hg.mozilla.org/integration/autoland/rev/b2c7917c24e9 (Part 2) Retain BasicLayerManager in WebRenderFallbackData for layer invalidation computation; r=jrmuizel

https://hg.mozilla.org/mozilla-central/rev/94c2ea91c36f https://hg.mozilla.org/mozilla-central/rev/b2c7917c24e9