Make reflect work for ElementInternals

[annevk]

Other changes:

• Remove reflection of unrestricted double as it is buggy and unused.
• The DOMString getter steps did not account for a missing attribute.
• The native accessibility semantics map was renamed to the internal content attribute map as it's now a more general reflection concept.

Fixes #8442.

Follow-up:

• Define how custom elements are exposed to AT w3c/core-aam#152
• GC for cached attr-associated elements? #8545
• Shadow DOM and ElementInternals ARIAMixin properties #8544
• Formalize content attribute reflection #3238
• ARIAMixin has many integer attributes with string types and uses DOMString? incorrectly w3c/aria#1110
• Use positive number instead of non-negative number greater than zero #8926
• Formalize default value reflection concept #8927
• Investigate number parser reuse #8928
• Consider making the reflected target always the interface the IDL attribute is on #8930

Corresponding ARIA PR: w3c/aria#1876.

---

/common-dom-interfaces.html ( diff )
/custom-elements.html ( diff )
/dom.html ( diff )
/infrastructure.html ( diff )

[domenic]

This seems like a good start, but it has some issues... it's unclear how far we can get with a patch that just makes things a little better, versus doing the full rewrite of all reflection that I think both of us have had kicking around in our head for a while. If you do want to shave some yaks, I'm very supportive!

[domenic]

This looks promising. Sorry for the long delay, and I'll try to be more responsive this time around!

[annevk]

If this generally looks good I'll make a more concerted effort to apply this throughout the whole section and tidy up some of the language as well as start working on the corresponding PRs that will be needed.

[domenic]

Looks good! Please proceed :)

[annevk]

Now that all the algorithms are converted:

• Read through the section once more to see if it still makes sense.
• Look at what needs updating around ElementInternals and perhaps have some kind of note pointing here.
• Briefly investigate [Reflect] annotations to see how much additional work that would be at this point.
  • Formalize content attribute reflection #3238 (comment) still looks reasonable and I could see defining some kind of macro on top of this PR to tackle what is stated there. However:
    • The real value would be with enumerated attributes (especially nudging people in the direction of "limited to known values") and those are not tackled.
    • It doesn't handle the reflected target being a different object. I suspect that has to remain prose-only forever.
  • If we got [Reflect] sorted we would get rid of "support ElementInternals" again. That does seem like a worthy goal, but at this point I think I rather get there incrementally unless we have a very clear plan.
• Figure out and write the updates to ARIA. Ideally as close to an editorial PR as possible.

Meanwhile I'd still appreciate drive-by review of course of what I managed to get done today.

[annevk]

While looking at ARIA I hit a bit of a snag, ARIA has a lot of spec-UB: w3c/aria#1110 (comment).

• Then, when I was wondering if some could maybe be changed to use DOMString instead, I noticed that DOMString reflection does not account for the lack of a content attribute at the moment! I'm relatively sure that should be returning the empty string. That would be another normative "bug fix" made in this PR.

My current thinking is to leave some of the spec-UB in ARIA, but at least make it adopt this new framework. Thoughts?

One other thing I wanted to raise, should we rename "native accessibility semantics map" to something more neutral such as "internal content attribute map" so it can be used for other internal state as well, as needed? We could also do that later as there's not going to be many call sites.

[domenic]

• Then, when I was wondering if some could maybe be changed to use DOMString instead, I noticed that DOMString reflection does not account for the lack of a content attribute at the moment! I'm relatively sure that should be returning the empty string. That would be another normative "bug fix" made in this PR.

Before #7956 it said

If a reflecting IDL attribute is a DOMString or USVString attribute but doesn't fall into any of the above categories, then the getting and setting must be done in a transparent, case-preserving manner.

which doesn't seem better. So yeah I guess this bug has been around for a long time.

One other thing I wanted to raise, should we rename "native accessibility semantics map" to something more neutral such as "internal content attribute map" so it can be used for other internal state as well, as needed? We could also do that later as there's not going to be many call sites.

Yes, I'd be in favor of that. Even if we never plan on expanding it to other attributes, it's a bit weird to have it be accessibility-named now that there's a general framework.

My current thinking is to leave some of the spec-UB in ARIA, but at least make it adopt this new framework. Thoughts?

Seems reasonable, but we should probably make addressing this the next priority in this area.

[domenic]

Looks great, with some clarity-improvement suggestions. Happy to do another round of review though if needed.

[domenic]

LGTM except maybe that intro bullets thing. I'll also check out the ARIA PR.