[cssom-view] elementFromPoint, elementsFromPoint, and caretPositionFromPoint should not return an element inside a shadow tree

[rniwa]

elementFromPoint and elementsFromPoint should not return an element inside a shadow tree. Instead, it should look for the highest shadow host of the element and return that instead so that it doesn't leak an element in shadow trees.

See the shadow DOM specification.

More precisely, once these methods are added on DocumentOrShadowRoot interface, then we need to retarget the element we found against the context object.

[rniwa]

@annevk @hayatoito @zcorpan

[rniwa]

Okay, there's another issue with the current definition either in CSS OM or shadow DOM specification. When the result of hit testing results in a Text node which is a direct child of ShadowRoot, what should we return? It looks like Chrome currently returns null but that seems broken. Given the definition of retargeting, it should be returning the shadow host instead.

But none of this is clear because either specification really says anything about in which tree the node is found (e.g. composed tree versus flat tree; probably the latter) and when & where the retargeting happens.

[rniwa]

Another edge case to consider is when the hit testing results in a Text node which is assigned to a slot. I think we should return the slot when elementFromPoint is called from within the shadow tree and the shadow host when it's called outside the shadow tree.

[rniwa]

I'm going to implement the proposed behavior in https://bugs.webkit.org/show_bug.cgi?id=162882 because whatever Chrome/Blink currently implements is awfully broken.

[rniwa]

caretPositionFromPoint is even weirder. It’s not clear to me what should happen to an offset when the node is retargeted.

[hayatoito]

Hmm. It is not good that we would not have interoperable implementations.

@yosinch FYI. He should know better than me about how Blink implemented these APIs.