sffc
commented
1 year ago If we use the 24-month numbering: we need to figure out date arithmetic. We do not have many invariants when it comes to what calendars do with date arithmetic. It's already the case that CLDR calendars can change the day number when adding/subtracting months (e.g. August 31 + 1 month) and the month number when adding/subtracting years (e.g. 1 Adar II + 1 year). I don't immediately see anything wrong with these hindu calendars changing the ordinal month by 2 when the duration says to add 1 month.
If we use 12-month numbering: nothing really wrong here in the data model; it's just a big headache when formatting. We would need to pull more special casing into the formatting module when we would really prefer as much as possible to keep calendar-specific logic in the calendar module.
Might I suggest the following month code scheme for the 24-month system:
- M01X = waxing phase of the first month
- M01Y = waning phase of the first month
- M02X = waxing phase of the second month
- ...
For leap and merge months:
- M01M = waxing phase of the first+second merge month
- M01N = waning phase of the first+second merge month
- M01P = waxing phase of leap first month
- M01Q = waning phase of leap first month
The table becomes:
| x | Wax | Wane | Wax | Wane | Wax | Wane | Wax | Wane | Wax | Wane |
|---|---|---|---|---|---|---|---|---|---|---|
| Amanta | M03X (Jyeshta) | M03Y | M04X (Ashad) | M04Y | M05P (Extra Shravan) | M05Q | M05X (Shravan) | M05Y | M06X (Bhadrapad) | M06Y |
| Purnimanta | M03X | M04Y | M04X | M05Y | M05P | M05Q | M05X | M06Y | M06X | M07Y |
Manishearth
FrankYFTang
ptomato
robpalme
anba
Nepal uses the Vikram Samvat calendar as an official calendar; this is a problem we will have to deal with eventually.
Hindu lunisidereal calendars are a bit strange, in ways that may affect the model put forth in Temporal, as well as the data model used in CLDR.
Similar to the Chinese calendar, they allow for leap months at any time of the year. We already have a solid solution for this, with month codes like M01L, M02L, etc.
There are three principal ways in which they are strange
Month naming and deleted/merged months
Sometimes months with a leap year will also have a deleted month
This in and of itself isn't a huge deal for our model. The tricky thing is that technically the month with the double transition gets named after both months, e.g. in 1983 there was a month called "Pausha-Magha kshaya", i.e. "Pausha merged with 'lost' Magha". We may wish to produce alternate month codes for these. (perhaps something like M10M)
These months are rare so we can treat them as a special case or just by the name of the first month if we wish.
Some traditions also use a special day name for the "regular" month (this is the Adar II/Adar II/Adar problem again).
Day naming and merged days
The Hindu calendar has 29-30 day lunar months, however they are each split into a waxing and a waning phase. Each day of the month has a name that is derived from the number. (1 to 14) You'll have names like "Shravan Waxing First", "Shravan Waning Fourteenth", "Shravan New Moon" and "Shravan Full Moon". Note that "First" is the second day of the fortnight/month since it will be preceded by a new or full moon.
Day 14 doesn't always occur as a distinct day; for shorter (29-day) months one of the fortnights will have merged day 14 with day 13. I don't yet know what that day is called; I think people call it by both names though not necessarily at once. On calendars it seems to be marked as "day 13/14", and I'm not sure how that is spelled out.
CLDR does already have support for day names. However this is more complicated logic than usual, since the day names aren't a straight up index into the ordinal day in the month. Days 1 through 13 (of the month) will cleanly map to "Waxing New Moon" to "Waxing Day Twelve", but after that it really depends on the precise month since the waxing fortnight may or may not have a distinct Day Thirteen/Fourteen.
One way of solving this problem is treating each fortnight as a unique "month", giving us 24 months in a year (plus two more when there is a leap month). This does solve the problem rather neatly, though we'd still need to figure out day naming for merged Thirteenth/Fourteenth.
This does somewhat match how the days get named, too, it's not as much "Waxing Day Five" as it is "Day Five of the Waxing Phase of Shravan".
Purnimanta vs amanta and split months
Broadly speaking, there are two traditions for month reckoning. Amanta traditions count months from new moon to new moon (like Chinese/Hebrew/Islamic), and Purnimanta traditions count months from full moon to full moon ("purna" is cognate with "full"). We would not treat these two traditions as the same calendar; so we don't have to deal with that, but there's an interesting bit of weirdness that comes out of their need to be consistent.
See, despite being seemingly incompatible traditions, they do end up being used for interchange. This works because the way they are set up, both calendars agree on the month name for the waxing phase of each cycle, but the waning phases differ. E.g. here, the waxing Chaitra is shared between two calendars, but the waning Chaitra occurs either before it or after it, depending.
Most festival dates are on the waxing phase, so most dates interchange directly, and for festivals in the waning phase (e.g. Janmashtami they get two dates listed (e.g. "Amanta: Shravan Krishna Ashtami, Purnimanta: Bhadrapad Krishna Ashtami"[^1]).
So far, still fine, as long as we maintain these calendars as separate calendars.
The real trickiness comes in with the leap months. Leap months are shadow versions of the regular month; roughly speaking none of the festivals occur then, but if the entire month has a significance the leap month does too, and monthly/fortnightly observances (e.g. stuff you do on the eleventh of each fortnight) still happen.
For this to work out in a synchronous form, the Purnimanta and Amanta traditions need to agree on where the leap month is. So if the following is a snippet of a regular year:
Then this is the format of a leap year with leap Shravan (like 2023):
In the Purnimanta calendar, the leap month gets inserted inside the regular month! This is the only way to maintain the correspondence between the two calendars: the leap months match up, and the waxing phases of the regular months match up, and the waning phases of the regular months are "one off".
This is another piece of evidence that the Hindu calendars probably should be modeled as ones with 24 months, since I can't even begin to figure out how date arithmetic would work with split months.
Modeling the Hindu calendars as ones with 24 months, however, bring up a new question: what should Temporal month APIs do? Ultimately, even though the fortnights behave more like months than the real months; people do associate the full 29/30-day months with the term "month" (and its translations). If you told someone that the month is "Shravan Waxing" that would not make sense to them (that is the fortnight, not the month).
Anyway, a lot to think about here, and perhaps worth splitting into sub-issues. I figured I ought to brain-dump the research I'd done so at least there's something to refer to for the future.
cc @FrankYFTang @sffc @ben-allen
[^1]: The "Krishna" in the festival name is only semi-related to the "Krishna" in the day name; "Krishna" means "dark" and is used to talk about the waning phase, this festival just happens to be for someone named after darkness.