Reflecting USVString attributes seems to be broken

[Ms2ger]

Content attribute values can be any WTF-16 value, so you can't return them from USVString-typed IDL attributes.

We might want to try to make all content attributes valid unicode, but unless we do that, we can't use USVString for things that can return attribute values directly.

[domenic]

Hmm. We should fix the definition of Reflect to run the convert to USVString algorithm from Web IDL, I think?

[domenic]

Or just go back to DOMString. @annevk, thoughts?

[annevk]

I was thinking this could be part of Reflect. With formalized slots this becomes even easier since both the content and IDL attribute are inputs to a slot of sorts.

[domenic]

I'd like to do this before we work on [Reflect]. It sounds like you are in favor of saying something like "if reflecting a USVString, convert the result to a USVString before returning it" to the section defining reflection?

[zcorpan]

What should happen on setting?

[annevk]

Yeah, when setting it ends up setting some internal slot (might even have to be a URL record for Blob URLs) which we'd define as converting to USVString and then passing it on to the URL parser.

[zcorpan]

OK, so the content attribute should just get set to the given value without conversion to USVString? If so it sounds like the type should still be DOMString.

[annevk]

Yeah, that sounds correct actually. Only the internal slot can be non-DOMString.

[domenic]

Wait, what? I thought we'd want to use the normal behavior when setting a USVString attribute. That is, I'd expect

const a = document.createElement("a");
a.ping = "\uDC01";
assert(a.getAttribute("ping") === "\uFFFD");
assert(a.ping === "\uFFFD");

[annevk]

Does that make sense when setAttribute() doesn't?

[domenic]

I mean, it makes as much sense as all the other non-transparent-passthrough logic reflecting does, like the numeric censoring-to-0. It seems like a similar situation to me: JavaScript has a simple unrestricted "string" or "number" type; the spec decides it wants a restricted subset of strings (USVString) or numbers (unsigned long); the setter is then responsible for censoring values outside that range, and the getter responsible for parsing/converting values until they fit in that range.

[annevk]

Fair.

[zcorpan]

@domenic SGTM. It wasn't clear to me what behavior we wanted for this. :-)