Clarify timeStyle:long/full for Plain types

[arshaw]

Hello!

All the Plain* types accept timeZoneName/timeZone for their toLocaleString string methods but they're ignored:

const pdt = Temporal.PlainDateTime.from({
  year: 2025,
  month: 1,
  day: 1,
  hour: 12,
  minute: 13,
})

pdt.toLocaleString('en', {
  hours: '2-digit',
  minutes: '2-digit',
  seconds: '2-digit',
  timeZoneName: 'long',
  timeZone: 'America/Chicago',
})
// '1/1/2025, 12:13:00 PM'

However, when timeStyle is set to either 'full' or 'long', the time zone is included:

pdt.toLocaleString('en', {
  timeStyle: 'full',
  timeZone: 'America/Chicago',
})
// '12:13:00 PM Central Standard Time'

I this intentional? Or just a shortcoming of the reference polyfill?

If it is in fact a shortcoming, and the reference polyfill should not be doing this, I would recommend internally forcing timeStyle 'full' or 'long' to 'medium' to omit the timeZone.

I've tested to see whether the 'medium' is always simply the 'long' or 'full' minus the timezone, and it seems to be the case:

https://codepen.io/arshaw/pen/MWRYKqV?editors=0010

This technique doesn't work for Dzongkha/Esperanto in Firefox/Safari, but still probably okay to force it 'medium' for those locales nonetheless.

I'm happy to create a PR for this if applicable.

[ptomato]

This is a shortcoming in the reference code, I think. Time zone should never be printed for Plain types. But I can't remember off the top of my head if the spec says to throw in that case, or to ignore the timeZone and timeZoneName.

[ptomato]

I looked at the spec and I believe it should be ignored.

In step 57 of CreateDateTimeFormat we consult Table 27 to see which options are supported for each Temporal type, and only copy supported options from formatOptions to limitedOptions for use in determining the DTF pattern for that Temporal type.

[arshaw]

Thanks for the info @ptomato. The algorithm you describe for copying DateTimeFormat options makes sense.

The problem I'm posing is actually not very related to timeZone or timeZoneName. I was merely including timeZoneName in the first code example to prove how it IS ignored in the normal case. In the second code example, I was including timeZone in order to show a consistent reproduction result that didn't depend on the current user's locale. I probably should have made my code samples more minimal.

The crux of the problem is related to timeStyle. If you run the following code in the playground, you'll see how the time zone is displayed:

Temporal.Now.plainDateTimeISO().toLocaleString('en', {
  timeStyle: 'full',
})
// "3:14:26 PM Eastern Standard Time" for me

This is a shortcoming with the reference implementation specifically. The timeStyle:full/long options comes with the time zone baked in. My solution is to massage all timeStyle:full/long values to medium, which never shows the timezone. I'm happy to make a PR to the reference implementation if you feel this is a satisfactory workaround.

[ptomato]

Yes, please feel free! The reference polyfill already uses plenty of other workarounds in toLocaleString(). Unlike the rest of the code, toLocaleString() isn't written exactly like the spec, because we wanted to use the native DateTimeFormat objects as much as possible and not have to reimplement them.

[arshaw]

Got it, thanks @ptomato.

Then I assume this is a mistake in this test:

https://github.com/tc39/test262/blob/46fc2814306c75f671924c6830346bf76b3ac30b/test/staging/Intl402/Temporal/old/datetime-toLocaleString.js#L50

The timeZone option should be ignored. And definitely shouldn't subject a PlainDateTime to a DST shift, right?

That test is currently being ignored but I can make a fix so that when it comes back online it won't cause problems.

[ptomato]

It's not obvious to me that there's a desired behaviour that will meet everyone's expectations all the time. But reading the spec I agree that timeZone should be ignored there. So the test seems wrong. Good catch!

[anba]

I've created #2904 to sync the Intl.DateTimeFormat changes with latest ECMA-402. Hopefully #2904 also makes it a bit easier to understand how the date-time format selection works. For this specific case (with and without #2904):

Temporal.Now.plainDateTimeISO().toLocaleString("en", {
  timeStyle: "full",
})

the same time format is used for all types (Temporal.PlainTime, Temporal.PlainDateTime, Temporal.ZonedDateTime, Temporal.Instant). This means time zone information is included, even for plain Temporal types.

Without #2904, this step is relevant:

d. If bestFormat does not have any fields that are in fields, then
  i. Set bestFormat to null.

When the current type is Temporal.PlainTime, then bestFormat and fields have a non-empty intersection, because both include "hour", "minute", and "second", so bestFormat is not set to null.

With #2904, the time format selection should be more easy to understand when timeStyle is used, because there's only a single call to DateTimeStyleFormat.

---

That being said, I'm not sure the current spec matches previous consensus. For example the current spec requires that this throw a TypeError in HandleDateTimeTemporalTime:

new Intl.DateTimeFormat("en", {dateStyle: "long"}).format(new Temporal.PlainTime());

whereas this call succeeds and returns the string "12:00:00 AM":

new Intl.DateTimeFormat("en", {day: "numeric"}).format(new Temporal.PlainTime());

[ptomato]

Meeting, 2024-09-19: We believe this is probably just a bug in the JS code and that the spec text is correct, but that needs to be verified.

The principle is that Intl.DateTimeFormat resolves options for all fields, but if you format a limited type like PlainYearMonth, you only use the options for year and month.

Just to make sure we are all talking about the same thing, I made a table of expected results (in my locale and time zone) for formatting the various limited Temporal types with various DateTimeFormat options. Let me know if you disagree with anything in this table.

format options	Instant	PDT	PlainDate	PYM	PMD	PlainTime
{dateStyle: 'short'}	2024-09-19	2024-09-19	2024-09-19	2024-09	09-19	throw
{day: 'numeric'}	19	19	19	throw	19	throw
{timeStyle: 'long'}	12:12:00 p.m. PDT	12:12:00 p.m.	throw	throw	throw	12:12:00 p.m.
{hour: '2-digit'}	12 p.m.	12 p.m.	throw	throw	throw	12 p.m.

Note that the time zone name is printed when formatting Instant with a timeStyle that includes the time zone. This is intentional, even though Instant does not include a time zone, it is the Temporal analogue to legacy Date and behaves the same way in formatting.

If there are no objections to this table, I'll verify that this is actually what the implementation-defined parts of the spec text suggest, and that it is covered by test262 tests.