[css-selectors-5]: move from :heading(<An+B>#) to :heading(<integer>#)

[keithamus]

Closes #12599

[AtkinsSJ]

<integer> would allow things like calc() in here. This probably only wants to permit literal integers? IDK, maybe calculations on env vars could be useful.

Though... I don't think CSS has an "integer literal" type anyway. Just a <number-token> one.

[keithamus]

Really? https://drafts.csswg.org/css-values-4/#integers doesn't seem to suggest that, instead saying it's a subset of <number-token>.

[AtkinsSJ]

I think the important phrase there is "When written literally". It's not super clear in that phrasing, but you can definitely use calc() for other things that take <integer>, for example z-index

[keithamus]

I'm sure calc() isn't allowed inside of selectors but I cannot find the relevant prose in spec to say where calc() is allowed.

Perhaps @tabatkins can point us in the right direction.

[AtkinsSJ]

I don't think we've previously had a situation where a numeric type is allowed in a selector, though I might have missed that. I'm definitely nit-picking on technicalities here. 😅 But yeah Tab will know.

[keithamus]

I guess calc() is supported for the <length> of a media query (although not a selector, it demonstrates that calc() has support outside of declaration values).

Perhaps I can update the spec prose here to explicitly call out that only literal <integer> values are allowed and calc() would result in a parse error? I can also update the tests to reflect this.

[zcorpan]

https://drafts.csswg.org/css-values-4/#math says

A math function represents a numeric value, one of:
...

• <integer>
  [...] and can be used wherever such a value would be valid.

Are arguments to functional selectors "values"? It's not super-clear to me if they should apply.

[tabatkins]

Yeah, selectors are kinda a weird space here. As we add more "generic" value-handling functions (like ident()), we'll have to make a general decision on whether they're allowed in places like a rule's selector. I lean towards disallowing, matching how An+B only allows literal ints.

So I guess, just to be clear, I recommend using <number-token>, with prose specifying that the token's type flag has to be "integer", a la the an+b definition.

(Looking at this, I just realized we have two competing definitions of <integer>, one from Syntax and one from Values, which have different definitions. That's not great. I should scope the Syntax definitions there to be specifically for an+b.)

[keithamus]

I opted for <level> as a discrete type just for a little more clarity. Hopefully this works, happy to use <number-token> instead though.

[SebastianZ]

Nit: The commit message and title misspell "heading" as "headng".

Sebastian