These test cases ensure that DST disambiguation does not take place on
intermediate values that are not the start or the end of the calculation.
Note that NormalizedTimeDurationToDays is no longer called inside
Temporal.Duration.prototype.add/subtract, so a few tests can be deleted.
Other tests need to be adjusted because NormalizedTimeDurationToDays is
no longer called inside Temporal.ZonedDateTime.prototype.since/until via
DifferenceZonedDateTime, although it is still called as part of rounding.
In addition, new tests for the now-fixed edge case are added, and for the
day corrections that can happen to intermediates.
See https://github.com/tc39/proposal-temporal/pull/2760
This adds tests specifically for every kind of case that changes due to
the tweak to the date difference algorithm: differences from a longer
month to a shorter month, when the months are adjacent, in the same year
but not adjacent, and in different years.
Also adds tests for a case that does *not* change, but would trip on an
incorrectly implemented algorithm: when the intermediate months value
falls at the end of February.
There was incidental coverage of the change to the date difference
algorithm in other tests. Those are adjusted, as well.
Normative change: https://github.com/tc39/proposal-temporal/pull/2759
Consensus in February 2024
Adapts the tests that checked arbitrarily long loops, to now check that an
exception is thrown if the loop would happen.
Adds tests that exercise the newly added checks on return values of
getPossibleInstantsFor and getOffsetNanosecondsFor that limit UTC offset
shifts to 24 hours or less.
Also updates some step numbers in related tests.
The existing tests didn't cover some edge cases where implementations
have to compute the exact result of `numerator / denominator`, where at
least one of `numerator` and `denominator` can't be exactly represented
by an IEEE-754 double precision floating point value.
"precision-exact-mathematical-values-5.js" gets added in #3961, so the
new tests from this commit start at "precision-exact-mathematical-values-6.js".
NormalizedTimeDurationToDays can no longer loop indefinitely, because at
a certain point we will hit the upper bound of MAX_SAFE_INTEGER, so rename
the test to reflect that it can loop an arbitrary but limited number of
times.
Add a test for the RangeError condition in NormalizedTimeDurationToDays
when the time zone calculates a day length that is not a safe integer
number of nanoseconds.
While editing these tests, rename them to match the current name of the AO
and make sure the step numbers are up to date. (Normally I wouldn't care
so much about that, but these tests can be pretty confusing so it's good
to be able to refer to the spec text.)
Adapts or removes tests that relied on creating durations that are now out
of range. Adds new tests for maximum in-range and minimum out-of-range
durations.
A few results change because the algorithm previously used for rounding
didn't always add duration units to dates in RFC 5545 order, and we also
introduce a special case for rounding with largestUnit years or months and
smallestUnit weeks.
In order to fixtc39/proposal-temporal#2563, we added invocations of
BalanceDurationRelative after some invocations of RoundDuration. These
cause observable calendar calls, which must be accounted for in some
existing tests.
See https://github.com/tc39/proposal-temporal/issues/2563
The old behaviour was encoded in one test in staging, but the behaviour of
largestUnit in duration rounding has changed since that test was written.
Therefore I'm assuming that toString() should've been updated when that
happened.
Built-in non-ISO calendars require either monthCode/day, or month/day plus
some form of year specification.
This adds test coverage for each of the categories listed in
https://github.com/tc39/proposal-temporal/issues/2664, of which some must
currently reside in the test/intl402/ folders.
We'll do this for now, then separately work on migrating all of the tests
that require a non-ISO8601 calendar but aren't dependent on it being any
particular calendar.
In the AO DisambiguatePossibleInstants, a PlainDateTime instance is passed
to user code. This instance should have the built-in ISO 8601 calendar.
Here are some tests that ensure it does.
See tc39/proposal-temporal#2671.
In PlainYearMonth arithmetic, we need to find the end of the month when
adding a negative duration or subtracting a positive one. This end of the
month can be out of range.
Test case based on one provided by Anba. See issue:
https://github.com/tc39/proposal-temporal/issues/2700
This removes several loopholes where it was possible to return particular
values from user calls that would cause infinite loops, or calculate
zero-length days.
This allows removing several tests, as calendar.dateAdd() is no longer
called in several places, and it's no longer possible to create a
situation where BigInt arithmetic is observable in NanosecondsToDays.
Note the monkeypatch of getPossibleInstantsFor in test/built-ins/Temporal/
TimeZone/prototype/getInstantFor/argument-builtin-calendar-no-array-
iteration.js.
Other than that, all the tests are basically identical.
Adds new tests to order-of-operations.js in Duration.round and
Duration.total, to exercise the code path where previous to this normative
change, relativeTo would have been converted to PlainDate 3x and 2x,
respectively.
