Fix non-ISO calendar month arithmetic

[justingrant]

Addresses three bugs with non-ISO date math. Fixes #1760. Fixes #1768.

This PR includes the following changes to the non-production polyfill:

• Fixes add/subtract when the starting day of the month was longer than the shortest month between the start and endpoint, e.g. adding a month to Jan 31 or subtracting a month from Mar 31 in the Gregorian calendar.
• Fixes until/since where largestUnit was 'months' or 'years' and the second argument was more than one month later than the first argument
• Fixes until/since where largestUnit was 'months' or 'years' where weeks were being incorrectly returned.
• Increases test coverage to cover these three fixes, plus all of datemath.mjs demitasse tests.

Because non-ISO calendar implementations are part of ECMA 402, not ECMA 262, this won't result in any spec changes nor docs changes. Only the non-prod polyfill is affected, but it's a polite to fix bug anyways so we won't have obviously broken results for some calendars.

[codecov]

Codecov Report

Merging #1761 (f6fc705) into main (e3f66bb) will increase coverage by 0.08%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##             main    #1761      +/-   ##
==========================================
+ Coverage   94.81%   94.89%   +0.08%     
==========================================
  Files          19       19              
  Lines       10764    10782      +18     
  Branches     1725     1729       +4     
==========================================
+ Hits        10206    10232      +26     
+ Misses        545      537       -8     
  Partials       13       13              

Flag	Coverage Δ
test262	73.30% <0.00%> (-0.11%)	⬇️
tests	90.14% <100.00%> (+0.13%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

Impacted Files	Coverage Δ
polyfill/lib/calendar.mjs	94.30% <100.00%> (+0.45%)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update e3f66bb...f6fc705. Read the comment docs.

[justingrant]

Converted to draft, because I think there may be a separate bug happening with subtraction of months. If there is, I'll fix it too before un-drafting this.

[justingrant]

Converted to draft, because I think there may be a separate bug happening with subtraction of months. If there is, I'll fix it too before un-drafting this.

Fixed the subtraction bug too and added tests for it. Removing draft status.

[ptomato]

I haven't looked closely at why this works but I think we should probably test gregory at least as well as we do iso8601, so it might be worth applying this patch https://gist.github.com/ptomato/adcedda30013fc750c7f4f57bd1427d1 to run the (hopefully exhaustive) datemath tests again with the gregory calendar. (The complication is that doing this reveals that there's a bug somewhere in non-ISO dateUntil as well, even without this PR)

[justingrant]

I think we should probably test gregory at least as well as we do iso8601, so it might be worth applying this patch https://gist.github.com/ptomato/adcedda30013fc750c7f4f57bd1427d1 to run the (hopefully exhaustive) datemath tests again with the gregory calendar. (The complication is that doing this reveals that there's a bug somewhere in non-ISO dateUntil as well, even without this PR)

Great idea! I'd prefer to do it in a separate PR though. I could fix dateUntil there, as well as any other bugs that show up. Want to file an issue to track, ideally with a failing dateUntil repro?

[ptomato]

I think we should probably test gregory at least as well as we do iso8601, so it might be worth applying this patch https://gist.github.com/ptomato/adcedda30013fc750c7f4f57bd1427d1 to run the (hopefully exhaustive) datemath tests again with the gregory calendar. (The complication is that doing this reveals that there's a bug somewhere in non-ISO dateUntil as well, even without this PR)

Great idea! I'd prefer to do it in a separate PR though. I could fix dateUntil there, as well as any other bugs that show up. Want to file an issue to track, ideally with a failing dateUntil repro?

I've done this in #1768.

[justingrant]

I think we should probably test gregory at least as well as we do iso8601, so it might be worth applying this patch https://gist.github.com/ptomato/adcedda30013fc750c7f4f57bd1427d1 to run the (hopefully exhaustive) datemath tests again with the gregory calendar. (The complication is that doing this reveals that there's a bug somewhere in non-ISO dateUntil as well, even without this PR)

Great idea! I'd prefer to do it in a separate PR though. I could fix dateUntil there, as well as any other bugs that show up. Want to file an issue to track, ideally with a failing dateUntil repro?

I've done this in #1768.

Adapting these tests and fixing that bug turned out to be really easy. So I just included the changes here. Also fixed another bug I found in the neighborhood.

[ptomato]

Thanks!