Normative: Fix Instant rounding modes

[ptomato]

As described in the documentation, the proposal champions' intention was
for rounding of exact times to consider the Big Bang as the "zero" point,
rather than the arbitrary zero point of the Unix epoch. The spec text
didn't reflect this correctly.

Closes: #2191

[ptomato]

Draft until presented at TC39.

[codecov]

Codecov Report

Merging #2210 (aa6a5c2) into main (8f90398) will not change coverage.
The diff coverage is n/a.

@@           Coverage Diff           @@
##             main    #2210   +/-   ##
=======================================
  Coverage   89.95%   89.95%           
=======================================
  Files          19       19           
  Lines       10901    10901           
  Branches     1683     1683           
=======================================
  Hits         9806     9806           
  Misses       1081     1081           
  Partials       14       14           

Flag	Coverage Δ
test262	83.47% <ø> (ø)
tests	80.46% <ø> (ø)

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

---

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 8f90398...aa6a5c2. Read the comment docs.

[rkirsling]

Thanks, makes sense to me!

[ptomato]

This reached consensus at the July 2022 TC39 plenary meeting.

[anba]

This change invalidated some other tests, for example https://github.com/tc39/test262/blob/main/test/built-ins/Temporal/ZonedDateTime/prototype/since/roundingincrement-non-integer.js:

DifferenceTemporalZonedDateTime calls DifferenceInstant with ns1 = 1_000_000_000_000_000_005n and ns2 = 1_000_000_000_000_000_000n. DifferenceInstant then calls RoundTemporalInstant with ns2 - ns1 = -5n and unit = "nanosecond". RoundNumberToIncrementAsIfPositive is then called with x = -5, increment = 2, and roundingMode = "trunc".

RoundNumberToIncrementAsIfPositive computes:

1. Let quotient be x / increment.
2. Let unsignedRoundingMode be GetUnsignedRoundingMode(roundingMode, false).
3. Let r1 be the largest integer such that r1 ≤ quotient.
4. Let r2 be the smallest integer such that r2 > quotient.
5. Let rounded be ApplyUnsignedRoundingMode(quotient, r1, r2, unsignedRoundingMode).
6. Return rounded × increment.

→

1. quotient = -5 / 2 = -2.5
2. unsignedRoundingMode = zero
3. r1 = floor(quotient) = -3
4. r2 = floor(quotient) + 1 = -2
5. rounded = r1 = -3
6. return = -3 * 2 = -6

With ApplyUnsignedRoundingMode:

1. If x is equal to r1, return r1.
2. Assert: r1 < x < r2.
3. Assert: unsignedRoundingMode is not undefined.
4. If unsignedRoundingMode is zero, return r1.

→

with x = -2.5, r1 = -3, r2 = -2

1. Not taken
2. ...
3. ...
4. return r1

So the test needs to be changed as follows:

- TemporalHelpers.assertDuration(result, 0, 0, 0, 0, 0, 0, 0, 0, 0, 4, "roundingIncrement 2.5 floors to 2");
+ TemporalHelpers.assertDuration(result, 0, 0, 0, 0, 0, 0, 0, 0, 0, 6, "roundingIncrement 2.5 floors to 2");

But I'm not sure this change was intentional by this PR.

[ptomato]

Whoops. I don't think it was ever correct to call RoundTemporalInstant on the difference between two instants, which should be a duration. I'll open a new issue for this.

[FrankYFTang]

After I try to change my v8 code to sycn with 2210 my build break the following tests, I am not 100% sure that is because the same issue @anba mentioned or I have some bugs in my patch or some other pre-existing issues but I would like to mention these tests here so you will consider them when you work on #2320

  'built-ins/Temporal/Instant/prototype/since/roundingincrement-non-integer': [FAIL],
  'built-ins/Temporal/Instant/prototype/since/roundingincrement-wrong-type': [FAIL],
  'built-ins/Temporal/Instant/prototype/since/roundingmode-undefined': [FAIL],
  'built-ins/Temporal/Instant/prototype/since/roundingmode-wrong-type': [FAIL],
  'built-ins/Temporal/Instant/prototype/since/smallestunit-wrong-type': [FAIL],
  'built-ins/Temporal/ZonedDateTime/prototype/since/roundingincrement-non-integer': [FAIL],
  'built-ins/Temporal/ZonedDateTime/prototype/since/roundingincrement-wrong-type': [FAIL],
  'built-ins/Temporal/ZonedDateTime/prototype/since/roundingmode-undefined': [FAIL],
  'built-ins/Temporal/ZonedDateTime/prototype/since/roundingmode-wrong-type': [FAIL],
  'built-ins/Temporal/ZonedDateTime/prototype/since/smallestunit-wrong-type': [FAIL],
  'staging/Temporal/Instant/old/since': [FAIL],
  'staging/Temporal/Instant/old/until': [FAIL],

[FrankYFTang]

@anba - do you see the same set of breakage or different?

[anba]

@anba - do you see the same set of breakage or different?

Yes, I see exactly the same failures.