Duration.prototype.{round,total}: Missing checks for too large date-time values

[anba]

Temporal.Duration.prototype.round, steps 27.f-h:

f. Let isoDateTime be CombineISODateAndTimeRecord(plainRelativeTo.[[ISODate]], MidnightTimeRecord()).
g. Let targetDateTime be CombineISODateAndTimeRecord(targetDate, targetTime).
h. Set internalDuration to ? DifferencePlainDateTimeWithRounding(isoDateTime, targetDateTime, calendar, largestUnit, roundingIncrement, smallestUnit, roundingMode).

Temporal.Duration.prototype.total, steps 12.f-h:

f. Let isoDateTime be CombineISODateAndTimeRecord(plainRelativeTo.[[ISODate]], MidnightTimeRecord()).
g. Let targetDateTime be CombineISODateAndTimeRecord(targetDate, targetTime).
h. Let total be ? DifferencePlainDateTimeWithTotal(isoDateTime, targetDateTime, calendar, unit).

isoDateTime and targetDateTime can be be outside the valid date-time limits, which breaks assertions in DifferencePlainDateTimeWithRounding and DifferencePlainDateTimeWithTotal when calling DifferenceISODateTime.

(This is already covered by test262 tests, but probably didn't get noticed because the polyfill doesn't implement the assertion steps.)

[ptomato]

@anba Are you sure it already has test coverage? I tried adding the assertion steps to the polyfill, and got no new test failures. Which ones fail on your end?

[anba]

After updating more code to match the current spec text, the assertions no longer trigger. So you're correct, this is missing test262 coverage.

Edit: Forgot to reenable the test262 tests which were failing because of this missing check. 😅

This is covered by

• built-ins/Temporal/Duration/prototype/total/relativeto-string-limits.js
• built-ins/Temporal/Duration/prototype/round/relativeto-string-limits.js

When the input is "-271821-04-19", this step:

Let isoDateTime be CombineISODateAndTimeRecord(plainRelativeTo.[[ISODate]], MidnightTimeRecord()).

creates a date-time valid outside the valid ISO date-time valids, because the time part is midnight.

[ptomato]

So here's what I think is happening. The assertion is not tripped in those two test262 tests because the duration is zero, and DifferencePlainDateTimeWithRounding/DifferencePlainDateTimeWithTotal have an early return for the case of two identical PlainDateTimes. However, if you run the same operations with e.g. const instance = new Temporal.Duration(0, 0, 0, 0, 0, /* minutes = */ 5); the assertion is tripped. So I think there really is no test coverage for this, and maybe your code skipped the early return?

I will first adapt the three relativeto-string-limits.js tests so that they would cover this situation.