CalendarDateToISO: Constrain behaviour for regnal years?

[anba]

It'd be nice to have some more detailed information how constraining should work for regnal years in the Japanese calendar.

1. Dates before the start of the era (from SpiderMonkey WIP implementation):

js> Temporal.PlainDate.from({calendar: "japanese", era: "reiwa", eraYear: 1, month: 1, day: 1}).toString()
"2019-05-01[u-ca=japanese]"
js> Temporal.PlainDate.from({calendar: "japanese", era: "reiwa", eraYear: 1, month: 1, day: 1}).era
"reiwa"
js> Temporal.PlainDate.from({calendar: "japanese", era: "reiwa", eraYear: 1, month: 1, day: 1}, {overflow: "reject"}).toString()
typein:1:20 RangeError: calendar field "eraYear" is too large: 1

Whereas the spec polyfill changes to the previous era (for both constrain and reject overflow behaviour):

js> Temporal.PlainDate.from({calendar: "japanese", era: "reiwa", eraYear: 1, month: 1, day: 1}).toString()
"2019-01-01[u-ca=japanese]"
js> Temporal.PlainDate.from({calendar: "japanese", era: "reiwa", eraYear: 1, month: 1, day: 1}).era       
"heisei"

2. Dates after the end of the era (from SpiderMonkey WIP implementation):

js> Temporal.PlainDate.from({calendar: "japanese", era: "heisei", eraYear: 100, month: 1, day: 1}).toString()
"2019-04-30[u-ca=japanese]"
js> Temporal.PlainDate.from({calendar: "japanese", era: "heisei", eraYear: 100, month: 1, day: 1}).era
"heisei"
js> Temporal.PlainDate.from({calendar: "japanese", era: "heisei", eraYear: 100, month: 1, day: 1}, {overflow: "reject"}).toString()
typein:1:20 RangeError: calendar field "eraYear" is too large: 100

Spec polyfill has (for both constrain and reject overflow behaviour):

js> Temporal.PlainDate.from({calendar: "japanese", era: "heisei", eraYear: 100, month: 1, day: 1}).toString()
"2088-01-01[u-ca=japanese]"
js> Temporal.PlainDate.from({calendar: "japanese", era: "heisei", eraYear: 100, month: 1, day: 1}).era       
"reiwa"

Questions I've asked myself:

• Does overflow behaviour affect era/eraYear?
• If yes, does constrain overflow constrain to the start/end of the era?
• If yes, does reject overflow reject dates before/after the era?

[justingrant]

Meeting 2024-05-30:

• For constrain we'll treat each era "proleptically": dates before the era starts are mapped to the previous era (without changing the ISO date due to constraining), and dates after the era ends are mapped to later era (again without changing ISO date).
• For reject, throw if it's before or after the era boundaries.

@gibson042 to make a PR. Because current behavior is unspecified, this will not be a normative change.

[anba]

I've just found tc39/ecma402#540 which actually wants a different behaviour for reject, namely that reject doesn't throw for dates after the end of the era. I don't know if this should imply that reject should also not throw for dates before the start of the era, though.

[ptomato]

@gibson042 Will you have time to work on this in the next while? If not, I can take it over.

[gibson042]

I think this should extend #2940, but don't plan to start it until after that lands (or is at least in a final form).

[gibson042]

As I went to start on this, I realized that I'm not certain where it landed. To clarify, I think there are three dimensions to consider:

• overflow: Developer choice of "constrain" vs. "reject" might affect behavior (keeping in mind that it applies not just to from and with, but also to add/subtract arithmetic).
• stability: Many calendars have stable eras that are not expected to change (e.g., Gregorian, Coptic, Ethiopic, Minguo/ROC), while others (particularly regnal calendars such as Japanese) frequently add new eras and occasionally add past eras.
• directionality: A year later than the end of an era may be differentiated from a year earlier than the beginning. And as an added wrinkle, if they are so differentiated, backwards-counting inverse eras might logically classify "year value less than the chronologically latest year of the era" as either (e.g., the polyfill currently interprets { calendar: "gregory", era: "ce", eraYear: 0, month: 1, day: 1 } as 0000-01-01 [1 BCE] but rejects { calendar: "gregory", era: "bce", eraYear: 0, month: 1, day: 1 } with a "Years in bce era must be positive" RangeError rather than interpreting it as 0001-01-01 [1 CE]).

tc39/ecma402#540 suggests that overflow should be irrelevant for unstable/regnal calendars (specifically, that out-of-bounds era/eraYear combinations are always mapped into the appropriate era), and presumably for stable calendars as well (in which case directionality would also become irrelevant). That makes sense to me, but it does mean that e.g. Temporal.PlainDate.from({ calendar: "gregory", era: "bce", eraYear: 0, month: 2, day: 1 }, { overflow: "reject" }) will return a date with a different era and eraYear rather than interpreting "reject" as justifying a RangeError (as it would with e.g. day: 30) and I'd want to update docs in addition to the spec guidance.

Does anyone want to argue for an approach that would sometimes throw Errors rather than remapping?

[ptomato]

I think you are right that remapping is the natural expected behaviour for eras. In other words, overflow: 'balance'. (Unlike for days that exceed the month, where we struggled to find use cases for balance.)

To take an example:

Temporal.PlainDate.from('2019-06-01').withCalendar('japanese').subtract({ years: 1 }, { overflow: 'reject' })

I cannot imagine any possible scenario in which it would be good that this throws. Likewise, for the same example with { overflow: 'constrain' }, I cannot imagine any possible scenario in which it would be good that this returns 2019-05-01 (i.e. constraining to the start date of the era) or 2019-06-01 (i.e., subtracting 1 year and then constraining the year to the first year of the era.)