[css-fonts-3] Computed value of font-family isn't quite right

[fantasai]

The computed value of 'font-family' is listed “as specified” but actually, its defined more clearly further down as being a list, each item consisting of either a <generic-family> keyword or a string. (Non-generic font family names specified as keywords get converted to a string.) CSS Fonts 4 gets this correct.

[fantasai]

Or maybe the prose below the propdef table is more correct? Either way they're not very consistent. One says unquoted sequences of identifiers compute to a string and the other says they don't.

Testcase: http://software.hixie.ch/utilities/js/live-dom-viewer/?saved=6226

[upsuper]

The implementation status of this is quite complicated. Browsers try to preserve identifiers if specified so to some extent, but don't agree with each other.

IIRC, Blink preserves identifier when there is only a single identifier, otherwise it serializes it as string. Gecko does a bit more and tries to preserve identifier sequence if there is no abnormal whitespaces via escaping (e.g. leading whitespace, trailing whitespace, more than one whitespaces in the middle).

It's not clear whether that is necessary or useful, though. And I'm also not sure whether this specific detail needs to be fully spec'ed.

[svgeesus]

Hmm.
Firefox gives "quoted family", unquoted family, sans-serif
Chrome "quoted family", "unquoted family", sans-serif (extra quotes)
Edge "quoted family",unquoted family,sans-serif (no spaces)

[tabatkins]

Edge is def wrong - we've always mandated spaces after commas in serializations.

While I prefer "always quote" as a behavior, nobody does that; given the choices, I slightly prefer Chrome's behavior, simply because it's simplest to understand, I think. (font-family's munging of identifiers is so, so bizarre.)

I do think this should be specified; at minimum, we have to choose a single behavior to mandate for Typed OM's serialization.

[svgeesus]

/me at least no-one double quoted the quoted strings

[fantasai]

I think my preference would be to make the computed value a list of strings and/or generic-family keywords, and for each string track a flag on whether it should be quoted for getComputedStyle() (and limit this specifically to getComputedStyle(), for compat).

If we can get away with Chrome's more limited behavior, that seems easier to understand. If we could get away with quoting everything, that'd be the best. Then we can drop the flag. But I think Chrome would have to try deploying that to see if it's Web-compatible first.