[cssom] Please define parsing a group of selectors more clearly

[bzbarsky]

https://drafts.csswg.org/cssom/#parsing-selectors says to

To parse a group of selectors means to parse the value using the selectors_group production defined in the Selectors specification and return either a group of selectors if parsing did not fail or null if parsing did fail.

but there is no selectors-group in https://drafts.csswg.org/selectors-4/ (it was there in selectors-3). Furthermore, this is vague about how namespace prefixes should be handled, if at all.

Plausible answers are to just reuse https://drafts.csswg.org/selectors-4/#parse-a-selector or define some variant of it which considers the @namespace rules from the relevant sheet or something.

[bzbarsky]

In particular, https://drafts.csswg.org/selectors-4/#parse-a-selector itself is kinda vague about what it means for a namespace prefix to be "declared"...

[jyc]

Here's what some browsers currently do:

• Firefox doesn't implement setting selectorText (Bug #37468)
• Chrome implements setting selectorText, and seems to use namespaces from the stylesheet.
• Safari implements setting selectorText, but does not seem to use namespaces from the stylesheet.

If setting selectorText is implemented and namespaces from the stylesheet are used, then both sentences should be green in this snippet: https://jsfiddle.net/qqfsjLzp/1/

[tabatkins]

Should "parse a selector" care about namespace prefixes? That is, should I make the language there more specific, or remove it entirely? I think it's context-specific, which means the calling context has to manually check the namespace prefixes in the returned selector for itself. I think I can remove the "unknown simple selector" thing too, then? It might be legitimately useful to have a generic selector-parsing algo that doesn't care if a pseudo-class is unknown.

[bzbarsky]

As long as we fix all the consumers, that would be ok.

[tabatkins]

I think CSSOM is the only consumer right now.

[bzbarsky]

querySelector/querySelectorAll is a consumer for sure.

[tabatkins]

In my head I'd temporarily thought those were defined by CSSOM too. Indeed they are not. ^_^