Bug 231832 – Make sure child layers of top layer elements are rendered and correctly z-ordered (top-layer-stacking.html fails)

• Status: RESOLVED
• Resolution: FIXED
• Component: Layout and Rendering

[Tim Nguyen (:ntim)]

I suspect something is wrong in RenderLayer::rebuildZOrderLists

[Radar WebKit Bug Importer]

<rdar://problem/84316561>

[Tim Nguyen (:ntim)]

Created attachment 441461 [details]
Patch

[Simon Fraser (smfr)]

Comment on attachment 441461 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=441461&action=review

I think the real bug here is that RenderLayer::collectLayers() just early returns when establishesTopLayer() is true, so top layer layers never get any children.

> Source/WebCore/rendering/RenderLayer.cpp:768
> +    collectLayersForChildren(includeHiddenLayers, false, posZOrderList, negZOrderList, accumulatedDirtyFlags);

This is called the "boolean trap": a true/false argument that is hard to understand here at the call site. We generally avoid this by using enums.

> Source/WebCore/rendering/RenderLayer.cpp:803
> +                auto startOfSorting = posZOrderList->end();
> +                layer->collectLayersForChildren(includeHiddenLayers, true, posZOrderList, negZOrderList, accumulatedDirtyFlags);
> +                std::stable_sort(startOfSorting, posZOrderList->end(), compareZIndex);

I don't think this is right. This will append all descendant layers to 'posZOrderList' but what needs to happen is that 'layer' here, which is a stacking context, builds its own positive and negative z-order lists.

> Source/WebCore/rendering/RenderLayer.cpp:827
> +        auto& layerList = (zIndex() >= 0 || hasTopLayerAncestor) ? positiveZOrderList : negativeZOrderList;

Here all descendants of top layer layers append to the positiveZOrderList but this is wrong.

[Simon Fraser (smfr)]

Created attachment 441466 [details]
Patch

[EWS]

Committed r284314 (243110@main): <https://commits.webkit.org/243110@main>

All reviewed patches have been landed. Closing bug and clearing flags on attachment 441466 [details].