[css-pseudo-4] Should `::file-selector-button` be a `has-allowed pseudo element`

[lukewarlow]

::file-selector-button isn't dependent on any CSS for its continued existence so it seems a good candidate to be made a :has-allowed pseudo element?

Is there a reason it wasn't made one initially?

[tabatkins]

I agree, it should be.

Is there a reason it wasn't made one initially?

We were conservative with the list, is all. (And I think the pseudo's name hadn't been fully decided when we invented the concept.)

[dbaron]

I think this is a good testcase for having a has-allowed pseudo-element.

But I think #12630 (comment) applies here too. (We could also make :has(::pseudo) have a different default behavior from :has(tag) and test for pseudo-attachment rather than descendant combinator.)

[lukewarlow]

I agree that :has(::file-selector-button) matching the input's parent seems completely wrong. A new combinator feels the wrong approach, we should just make it match the originating element in this case.

[dshin-moz]

:has cannot be nested, so something like .form:has(.sub-form input:has(::file-selector-button)) wouldn't work, which is maybe unfortunate?

[dbaron]

Maybe we want :has-pseudo() or :attached() or something similar that accepts pseudo-elements, and is allowed inside of :has(), and represents only the "this element has one of these pseudo-elements". (And then we wouldn't allow pseudo-elements in :has() at all, but would allow them inside of :has-pseudo() inside of :has().

[emilio]

Yeah, I like the :has-pseudo() approach, it's also consistent with :has-slotted for example...

[bramus]

+1 on :has-pseudo()

[dbaron]

BTW, my concern with the :has-pseudo() name as I was suggesting it is that it's only for pseudo-elements and not pseudo-classes.

[emilio]

I'd be ok with :has-pseudo-element() tbh. A bit longer but... :)

[lukewarlow]

But it would be good to still be able to do :has-bruce(::file-selector-button:hover)

[dbaron]

Yeah, I think that's an issue, although there are two ways we could take it:

1. we do need to make this work with :has() in some way, or
2. :has-pseudo-element() isn't actually different from :has() except for the combinator rules at the start -- it's just :has() where the implied combinator at the start is pseudo-element-of (and the first thing better be a pseudo-element) but otherwise it's just like :has().

(If we go with (2) we need to decide whether the :has-pseudo-element(not-a-pseudo) is a parse error or just fails to match.)