Normative: Validate unit-valued options after all options are read

[sffc]

See #3116

This preserves the original call to GetOption, but it always calls it with the full list of units, and I pull the subset-checking logic into a separate AO.

I have two commits: an editorial change to split the AO in two, and a small normative change to move the validation later down at the call sites.

@nekevss @Manishearth

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 96.85%. Comparing base (a10d38d) to head (80852fe).

Additional details and impacted files

@@           Coverage Diff           @@
##             main    #3130   +/-   ##
=======================================
  Coverage   96.85%   96.85%           
=======================================
  Files          21       21           
  Lines        9983     9983           
  Branches     1829     1829           
=======================================
  Hits         9669     9669           
  Misses        268      268           
  Partials       46       46           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[sffc]

The code that changes behavior with this PR is something like:

obj = {
    get largestUnit() {
        console.log("get largestUnit");
        return "years";
    },
    get smallestUnit() {
        console.log("get smallestUnit");
        return "seconds";
    },
}

Temporal.PlainTime.from("00:00").round(obj)

Current behavior: console logs "get largestUnit" and then RangeError is thrown.

New behavior: console logs both "get largestUnit" and "get smallestUnit", and then RangeError is thrown.

Note: the RangeError is because PlainTime does not allow "years" as a largestUnit. There is no difference in behavior if get largestUnit() returns an invalid string such as "INVALID".

(as noted previously, there doesn't appear to be Test262 coverage for this)

[gibson042]

The intended changes as I understand them do seem both reasonable and narrow, but I think the algorithms steps fail to properly implement them.

[Manishearth]

This is roughly what I was imagining, thanks for doing all this work!

As Richard has noted, handling unset/default values may be tricky. The way I was expecting it to work was that the get allows you to specify REQUIRED, and Validate doesn't need to care, which roughly appears to be what you're doing here.

[ptomato]

I'm surprised that this didn't show up in more places. But I guess smallestUnit is often the last option read anyway, because S is relatively late in the alphabet. So I'm pleasantly surprised at the small extent of the normative change. (I am assuming this PR is the full extent of the changes!)

All in all I'd still prefer not to do this. (It isn't a bug in the proposal, it sets a new precedent that TC39 hasn't really shown appetite for so far, and I just don't see the alternative workaround in temporal_rs as that big of a deal.) But it's small and targeted enough that if I'm the only opposing voice, I won't stand in the way.

Logistical note: to make the best impression when presenting to TC39, I'd recommend squashing the fixup commits into their appropriate editorial or normative commits, then splitting out the editorial commit into its own PR which I can review quickly and merge. That way, there's less for delegates to review and it's much clearer what is changing.

[sffc]

This PR fixes the issues involving unit-valued options, which is the current scope of the issue and all I'm asking for at this time.

Yes, often smallestUnit is the final read, so no change was needed in those cases.

Yes, splitting the editorial change into its own PR is something I'm happy to do.

[dminor]

@anba Do you have any concerns about this change?

[sffc]

I opened #3135 for the editorial changes.

[ptomato]

The editorial changes are merged, and I rebased this. So now it's just the minimal set of normative changes.

[anba]

@anba Do you have any concerns about this change?

I don't think so. It'll require some small refactoring because the current SM implementation sometimes uses AUTO as a placeholder to represent absent values. This approach is no longer possible when "auto" must always initially be parsed as a valid Temporal unit option. But apart from that I don't see any implementation issues.

(The SM implementation already parses the unit options accepting all units and only later check which units are actually allowed. )