Current scoping root used when matching "selector_matches" doesn't work well for shadow tree links

[a4sriniv]

Currently, when matching a selector against a link while processing document rule CSS selector predicates, we use the link's document as a scoping root. This would mean that elements inside other tree scopes (including shadow trees attached to the document) would always be excluded when matching these selectors and would never match.

(From spec, "When a selector is scoped, it matches an element only if the element is a descendant of the scoping root.").

I think we should be using the link's root as the scoping root instead. So for links inside a shadow tree, we would use the shadow root as the scoping root.

[jeremyroman]

Hrm, interesting. It's always been a little awkward because it seems clearly useful to encompass shadow DOM, and yet CSS selectors generally don't pierce that boundary.

1. What, if anything, is the difference between scoping to the element's root and not scoping here? (I assume whether :scope matches is one thing, but is that it?)
2. Are there drawbacks to allowing you to apply selectors inside the shadow roots like this? Should we instead consider "scoped" speculation rules? (My reservation about that idea is that it somewhat defeats the purpose, though.)

I suspect your resolution is correct, but wouldn't mind a second opinion from DOM folks if you haven't already asked.

[a4sriniv]

1. What, if anything, is the difference between scoping to the element's root and not scoping here? (I assume whether :scope matches is one thing, but is that it?)

AFAICT, the only difference is that :scope would only match against the document if a scoping root is not specified - and so a selector like :scope .foo would never match an element inside a shadow tree.

(From https://drafts.csswg.org/selectors-4/#the-scope-pseudo: "If there is no scoping root then :scope represents the root of the document (equivalent to :root)")