[selectors] Validity of selectors nesting :has(), :is() and :has()

[faceless2]

We're redoing our :has() implementation due to #7676 and #7344 and I hit this case:

div, x:has(:is(:has(:foo))) { background: red }
div, x:has(:is(:has(div))) { background: red }

I'm not sure which of these statements is true

1. Both selectors are valid; although :has(:foo) and nested :has() are invalid, the forgiving nature of :is() means it's ignored.
2. The first selector is valid, because the inner :has() is invalid so it's ignored. The second is invalid; it's a nested :has()
3. Both selectors are invalid; there's a nested :has()

Safari goes for option 1, Chrome is option 3, and I think it should probably be option 2 - the forgiving nature of :is() seems like it should take priority to me.

Testcase at https://jsbin.com/roxukig/edit?html,output

[tabatkins]

As currently written, the spec requires the inner :has() to be the one that's invalid. The outer :has() just provides a parsing context that causes the invalidity; it doesn't care, itself, about why something is invalid inside of it. (And avoiding this sort of inconsistency is why the spec is written as it is, rather than invalidating the outer :has().)

So in both cases, the selectors are valid, but the :is() discards its arguments due to the nested :has() being invalid. So your option 1 is correct.

[faceless2]

No action required so closing, but noting that as of today Chrome and Safari both go with Option 1, Firefox is option 3.