NudgeToCalendarUnit: CreateTemporalDate is fallible

[anba]

Steps 3.a-d:

Let isoResult1 be BalanceISODate(dateTime.[[Year]] + duration.[[Years]], dateTime.[[Month]] + duration.[[Months]], dateTime.[[Day]]).
b. Let isoResult2 be BalanceISODate(dateTime.[[Year]] + duration.[[Years]], dateTime.[[Month]] + duration.[[Months]], dateTime.[[Day]] + duration.[[Days]]).
c. Let weeksStart be ! CreateTemporalDate(isoResult1.[[Year]], isoResult1.[[Month]], isoResult1.[[Day]], calendarRec.[[Receiver]]).
d. Let weeksEnd be ! CreateTemporalDate(isoResult2.[[Year]], isoResult2.[[Month]], isoResult2.[[Day]], calendarRec.[[Receiver]]).

The CreateTemporalDate calls are fallible.

[anba]

I also don't (yet?) quite understand why the duration fields are interpreted as ISO-calendar units, so it might be help to add a note why directly calling BalanceISODate instead of going through AddDate is allowed.

[ptomato]

@anba That may actually be a mistake on my part. At first glance, it looks like we'd need to call CalendarDateAdd for isoResult1 and then BalanceISODate to just add duration.[[Days]] to isoResult1 to obtain isoResult2. Thanks for catching it; I guess we should treat it as your post-merge code review.

As for the original issue, I assumed the CreateTemporalDate calls would not be fallible because isoResult2 would be the equivalent plain date for destEpochNs, which would be within the limits.

(But if an arbitrary date could be returned from CalendarDateAdd, that no longer holds.)

[anba]

As for the original issue, I assumed the CreateTemporalDate calls would not be fallible because isoResult2 would be the equivalent plain date for destEpochNs, which would be within the limits.

If duration.[[Years]] is let's say 1_000_000, then isoResult1 is outside the valid date-time limits, so the ISODateTimeWithinLimits check in CreateTemporalDate will throw a RangeError.

[ptomato]

But if you had a duration with calendar units that large, wouldn't it throw already before it got to this point? I should try stepping through this, to make sure...

[anba]

DifferenceDate in DifferenceTemporalPlainDate calls user-code for user-defined calendars, so the duration can have arbitrary large values, DifferenceTemporalPlainDate then calls RoundRelativeDuration, which in turn calls NudgeToCalendarUnit. The duration values are unchanged along this call chain, so large values should be possible.

(I haven't yet implemented this part in SpiderMonkey, so I'm not 100% sure, but this looks plausible when just reading the spec text.)

[ptomato]

I think the BalanceISODate / CalendarDateAdd distinction doesn't matter here because weeks in all of the supported calendars are always 7 days, so the result is the same no matter whether you add ISO years/months or calendar years/months to get weeksStart. Nonetheless, it's conceptually incorrect and we should change it.

[ptomato]

(The CreateTemporalDate calls are gone after #2925, so that part no longer applies.)