Layer flashing and poor perf during scrolling of message list on gmail.com and hotmail.com - overlap testing needs to constrained to clipping scopes

[Simon Fraser (smfr)]

Layer flashing and poor perf during scrolling of message list on gmail.com and hotmail.com - overlap testing needs to constrained to clipping scopes

[Simon Fraser (smfr)]

Created attachment 370354 [details]
Patch

[Simon Fraser (smfr)]

<rdar://problem/49403082>

[Antti Koivisto]

Comment on attachment 370354 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=370354&action=review

> Source/WebCore/rendering/LayerOverlapMap.cpp:180
> +    auto* clippingScope = findClippingScopeForLayers(enclosingClippingLayers);
> +    if (!clippingScope)
> +        return false;

Does this happen and what does it mean if it does? Could we assert here instead?

> Source/WebCore/rendering/LayerOverlapMap.cpp:246
> +OverlapMapContainer::ClippingScope* OverlapMapContainer::findClippingScopeForLayers(const Vector<LayerOverlapMap::LayerAndBounds>& enclosingClippingLayers) const
> +{
> +    ASSERT(enclosingClippingLayers.size());
> +    ASSERT(enclosingClippingLayers[0].layer.isRenderViewLayer());
> +
> +    const auto* currScope = &m_rootScope;
> +    for (unsigned i = 1; i < enclosingClippingLayers.size(); ++i) {
> +        auto& scopeLayerAndBounds = enclosingClippingLayers[i];
> +        auto* childScope = currScope->childWithLayer(scopeLayerAndBounds.layer);
> +        if (!childScope)
> +            return nullptr;
> +
> +        currScope = childScope;
> +    }
> +
> +    return const_cast<ClippingScope*>(currScope);

As discussed, maybe this can be replaced with a side hashmap? That might make the code easier to follow and avoid passing the full enclosingClippingLayers stack.

> Source/WebCore/rendering/RenderLayerCompositor.cpp:1811
> +    for (const auto* currLayer = layer.parent(); currLayer; currLayer = currLayer->parent()) {

Not a fan of the 'currFoo' naming.

[Simon Fraser (smfr)]

https://trac.webkit.org/r245602