[css-cascade-6] Tree-crossing @scope

[andruud]

The current definition of in scope requires an element to be an inclusive descendant of the scoping root. This means an element must be in the same tree as the scoping root to be "in scope":

@scope (#somehost) {
  :scope::part(p) { } /* Can never match */
}

However, browsers don't match the spec at the moment: https://wpt.fyi/results/css/css-cascade/scope-part.html?label=master&label=experimental&aligned&view=interop&q=label%3Ainterop-2025-scope

We don't behave consistently for all tree-crossing stuff; ::slotted() cases aren't working, at least not in Chrome and Firefox.

The question is, should we change WPTs/browers, or should we change the definition of what "in scope" means in the spec? cc @dshin-moz, @mdubet

[tabatkins]

Ah, we should change the definition. That definition is meant to just make sure that things in the same tree that aren't in scope aren't matched; it wasn't intended to restrict cross-tree matching, so long as the elements in the other tree could be considered reasonably "in the scope".

For shadow trees, I think this is very easy - if the shadow host is in scope, the shadow tree is in scope too.

[andruud]

For shadow trees, I think this is very easy - if the shadow host is in scope, the shadow tree is in scope too.

It's not enough for ::slotted(). If we want that to work, we'll need to say "inclusive flat tree descendant".