Normative: Correctly respect largestUnit option in duration rounding

[ptomato]

Note: this PR is stacked on top of #2567 and #2519. The only relevant commit for this issue is the topmost one, "Normative: Balance durations after rounding in until()/since()/toString()".

In cases where a number rounded up to the next unit, we would previously return a result that didn't respect the largestUnit option in the same way that Duration.prototype.round() would.

To fix this, we need to call BalanceDateDurationRelative after RoundDuration in the until()/since() methods that deal with date units (time units already behaved correctly).

In Duration.prototype.toString() where time units did not already behave correctly, we need to call BalanceDuration. I was not sure whether it was intentional that we didn't call this, but as far as I can determine from in toString() was "controls rounding, works same as round()", but the status quo was that it worked differently from round().)

Test262 tests are in a branch which is also stacked on top of tests for #2519. I will make it into a PR after we have merged #2519. Tests are in tc39/test262#3956

Closes: #2563

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Comparison is base (ee62aee) 96.36% compared to head (652788d) 96.38%.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #2571      +/-   ##
==========================================
+ Coverage   96.36%   96.38%   +0.02%     
==========================================
  Files          21       21              
  Lines       12535    12615      +80     
  Branches     2345     2356      +11     
==========================================
+ Hits        12079    12159      +80     
  Misses        401      401              
  Partials       55       55              

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[ptomato]

This change achieved consensus today at the May 2023 TC39 plenary meeting. Due to the PR being stacked on top of #2519, though, it would be better to merge this after merging that one first.

[justingrant]

Is this PR now part of #2645?

[ptomato]

Is this PR now part of #2645?

No, it won't be able to land until all of #2519 has landed. (I didn't/don't want to write a separate version of this PR to work correctly in the pre-Calendar-Method-Record world.)

[cjtenny]

Looks to me like it correctly does what it says on the tin - with one potential bug to address, see comment.

[jasonwilliams]

To fix this, we need to call BalanceDurationRelative after RoundDuration in the until()/since() methods that deal with date units (time units already behaved correctly).

Did you mean to say "we need to call BalanceDateDurationRelative" here?

PR looks good to me, I've spoken offline about potentially merging some logic into roundDuration but that will need some big changes including more parameters so may not be worth it for now.

[ptomato]

Did you mean to say "we need to call BalanceDateDurationRelative" here?

Indeed, thanks! Fixed.

[ptomato]

Tests are merged now, so this can be merged.