spec: valid year range in `NonISOMonthDayToISOReferenceDate`

[fabon-f]

NonISOMonthDayToISOReferenceDate has a step to validate whether fields.[[Year]] isn't too small or large:

i. If there exists no combination of inputs such that ! CalendarIntegersToISO(calendar, fields.[[Year]], ..., ...) would return an ISO Date Record isoDate for which ISODateWithinLimits(isoDate) is true, throw a RangeError exception.

My understanding is that Temporal.PlainMonthDay.from({ year: 275760, month: 12, day: 31, calendar: 'gregory' }) shouldn't fail since the maximum date belongs to the year 275760, but Temporal.PlainMonthDay.from({ year: 275761, month: 1, day: 1, calendar: 'gregory' }) should fail, according to the spec (the same thing would be applied for the lower bound and other calendars).

In reality, while SpiderMonkey and V8 raise an out-of-bound error for both cases, the reference polyfill accepts both. Also this discrepancy doesn't seem to be covered in test262.

[ptomato]

Thanks for reporting. SpiderMonkey and V8 are correct here. I've added test coverage in tc39/test262#4977.

[fabon-f]

Does "SpiderMonkey and V8 are correct" means that Temporal.PlainMonthDay.from({ year: 275760, month: 12, day: 31, calendar: 'gregory' }) (the case when year itself is not out-of-bounds but the specified date is out-of-bounds) should throw as opposed to the spec text?

[Manishearth]

Yes, and that was the behavior in the older spec text.

The fields given to PlainMonthDay.from() are fully evaluated (with constrain/reject behavior), and if that evaluation succeeds, the resulting M/D values are used with a reference year.

[ptomato]

I guess this was less clear in the new wording. I've pushed a commit to tc39/ecma402#1044 adjusting the sentence to clarify this.