Should ECMA-402 spec text for time zone canonicalization refer to CLDR or to IANA as authoritative?

[justingrant]

CLDR in unicode-org/cldr#3105 will soon provide the ability to fetch modern canonical IDs for currently-problematic time zones like Asia/Calcutta and Europe/Kiev. ICU will also be adding an API to expose the IANA canonical ID. This will enable V8 and JSC to finally expose modern IANA canonical names like SpiderMonkey does, but without the separate IANA-based overrides that SpiderMonkey has had to maintain.

Before this change, it didn't make sense to have normative spec text in 402 to define which IDs should be primary (the new 262 term for canonical time zone ID) vs. non-primary. But with this CLDR/ICU change, it's finally practical to specify normative rules for determining which IDs are primary (and, if not, which primary ID they resolve to).

I can draft a PR with this normative text for discussion, but first there's one main question to answer: should we specify the rules solely as using CLDR data, solely as using IANA data, or should we specify rules for which IANA IDs are canonical that just happens to match what CLDR is doing?

As an illustration, here's two possible directions that we could take this spec text. Don't worry about the particular text used (it's very rough and will change) but the general approach of depending on CLDR vs. depending on IANA is where I'm most looking for feedback.

Which one is better? @sffc @gibson042 @anba @Constellation @FrankYFTang

Option A - Defer to CLDR

The Unicode Common Locale Data Repository (CLDR) defines which available named time zone identifiers are primary or non-primary, as well as which non-primary time zone identifiers resolve to which primary time zone identifiers. The following exceptions are applied:

• For historical reasons, "UTC" is a primary time zone identifier while "Etc/UTC", "Etc/GMT", and "GMT" (and the identifiers that resolve to them) are non-primary time zone identifiers that resolve to "UTC".
• CLDR identifiers that are not present in the IANA Time Zone Database ("Etc/Unknown", "Canada/East-Saskatchewan", and "US/Pacific-New") are not supported.
• "Factory" is not supported.

The following spec text would replace the steps of https://tc39.es/proposal-temporal/#sup-availablenamedtimezoneidentifiers:

1. Let result be a new empty List.
2. For each <type>element that contains an alias attribute in timezone.xml in the Unicode Common Locale Data Repository (CLDR), do
   1. Let aliases be a new List, populated by splitting the space-delimited alias attribute.
   2. If an iana attribute is present, let primary be the String value of that attribute; otherwise, let primary be the first element in aliases.
   3. If primary is one of "Etc/UTC", "Etc/GMT", or "GMT", set primary to "UTC".
   4. If primary is not "Etc/Unknown" nor "Factory" , then
      1. Let record be the Time Zone Identifier Record { [[Identifier]]: primary, [[PrimaryIdentifier]]: primary }.
      2. Append record to result.
      3. For each element identifier in aliases, do
         1. If identifier is not primary and is not one of "Canada/East-Saskatchewan" nor "US/Pacific-New", then
            1. Set record to the Time Zone Identifier Record { [[Identifier]]: primary, [[PrimaryIdentifier]]: primary }.
            2. Append record to result.
3. Sort result lexicographically by UTF-16 code unit of each element's [[Identifier]] field, in ascending order
4. Assert: result contains a Time Zone Identifier Record r such that r.[[Identifier]] is "UTC" and r.[[PrimaryIdentifier]] is "UTC".
5. Return result.

Option B - Define using IANA only

Each Zone in the IANA Time Zone Database must be a primary time zone identifier and each Link name in the IANA Time Zone Database must be a non-primary time zone identifier that resolves to its corresponding Zone name, with the following exceptions:

• For historical reasons, "UTC" must be a primary time zone identifier. "Etc/UTC", "Etc/GMT", and "GMT", as well as all Link names that resolve to any of them, must be non-primary time identifiers that resolve to "UTC".
• Any Link name in the TZ column of zone.tab of the IANA Time Zone Database must be a primary time zone identifier.
• Any Link name that represents a geographical area entirely contained within the territory of a single ISO 3166-2 country code must resolve to a primary identifier that also represents a geographical area entirely contained within the territory of the same ISO 3166-2 country code. For example, "Atlantic/Jan_Mayen" must resolve to "Arctic/Longyearbyen".
• The following legacy POSIX identifiers must resolve to their Continent/City equivalents, as shown in the table below:

Legacy POSIX Zone Name	Primary Time Zone Identifier
EST	Etc/GMT+5
MST	Etc/GMT+7
HST	Etc/GMT+10
EST5EDT	America/New_York
CST6CDT	America/Chicago
MST7MDT	America/Denver
PST8PDT	America/Los_Angeles
WET	Europe/Lisbon
CET	Europe/Berlin
MET	Europe/Vienna
EET	Europe/Athens

Option C - Define IANA rules, but explain how to use timezone.xml

We could also merge both (A) and (B): define the IANA *and* explain how to use timezone.xml data to satisfy those rules. I'm not going to draft text for this yet because I'm unsure if anyone would want it, but I wanted to include this option here for discussion.

[gibson042]

https://tc39.es/ecma402/ is careful to recommend use of CLDR data without ever requiring it, and I don't want to change that. My preference would be following the existing recommendation pattern for time zone resolution.

[anba]

• Legacy POSIX identifiers that refer to a single fixed UTC offset, such as "EST" or "WET", must be non-primary time zone identifiers that resolve to the corresponding the corresponding Zone name, starting with "Etc/GMT", for the same UTC offset. For example, "EST" must resolve to "Etc/GMT+5".

As mentioned in the CLDR PR, WET, CET, MET, and EET are not fixed-offset time zones and therefore can't be mapped to some Etc/GMT±hh time zone. IOW only EST, HST, and MST can be mapped to Etc/GMT±hh time zones.

Option A will lead to adding the following time zones IDs:

Name	Kind
Etc/Unknown	Zone
Canada/East-Saskatchewan	Link to America/Regina
US/Pacific-New	Link to America/Los_Angeles

I don't think these IDs should be added. Etc/Unknown was already discussed in the June meetings. Canada/East-Saskatchewan has been removed from IANA because it's a misnomer. US/Pacific-New has also been removed from IANA, because it was intended for a possible US time zone rule change which never happened.

[justingrant]

• Legacy POSIX identifiers that refer to a single fixed UTC offset, such as "EST" or "WET", must be non-primary time zone identifiers that resolve to the corresponding the corresponding Zone name, starting with "Etc/GMT", for the same UTC offset. For example, "EST" must resolve to "Etc/GMT+5".

As mentioned in the CLDR PR, WET, CET, MET, and EET are not fixed-offset time zones and therefore can't be mapped to some Etc/GMT±hh time zone. IOW only EST, HST, and MST can be mapped to Etc/GMT±hh time zones.

@anba Good points. I updated (B) in the OP accordingly. I also changed from prose to using a table to indicate how we'll handle all those legacy IDs. Better?

Option A will lead to adding the following time zones IDs:

Name Kind
Etc/Unknown Zone
Canada/East-Saskatchewan Link to America/Regina
US/Pacific-New Link to America/Los_Angeles
I don't think these IDs should be added. Etc/Unknown was already discussed in the June meetings. Canada/East-Saskatchewan has been removed from IANA because it's a misnomer. US/Pacific-New has also been removed from IANA, because it was intended for a possible US time zone rule change which never happened.

I agree. I updated (A) to exclude them. I also turned (A) into more formal spec text.

https://tc39.es/ecma402/ is careful to recommend use of CLDR data without ever requiring it, and I don't want to change that. My preference would be following the existing recommendation pattern for time zone resolution.

@gibson042 Could you explain more about why a recommendation is needed as opposed to normative text? The problem we want to solve is deviation between implementations, so seems reasonable to limit wiggle room. Or am I missing something that makes this distinction not matter?

[gibson042]

Basically, ECMA-402 just isn't the right place for such constraints because of https://tc39.es/ecma402/#sec-api-overview

The API can be added to an implementation of the ECMAScript 2024 Language Specification (ECMA-262 15^th Edition, or successor) in whole or in part. This specification introduces new language values observable to ECMAScript code (such as the value of a [[FallbackSymbol]] internal slot and the set of values transitively reachable from %Intl% by property access), and also refines the definition of some functions specified in ECMA-262 (as described below). Neither category prohibits behaviour that is otherwise permitted for values and interfaces defined in ECMA-262, in order to support adoption of this specification by any implementation of ECMA-262.

ECMA-402 may add behavior that is not prohibited by ECMA-262, but may not prohibit behavior that is allowed by ECMA-262 except where ECMA-262 specifically delegates to it (as in toLocaleString methods). And CLDR is one source capable of providing localization-relevant information, but is not the only such source—an implementation that relies directly upon IANA data should not be considered non-conforming. Also note that ECMA-402 intentionally allows flexibility as a general rule (e.g., the output of a formatter is not considered stable across implementations or time), and deviations from that flexibility require strong motivation.

[justingrant]

TG2 meeting 2023-10-12: Option C above is preferred, with the caveat that we should make the spec text be a recommendation not a normative requirement.

I'll prepare a PR. For specific detailed concerns like those raised in comments above, they're narrow enough that TG2 thought that we can hash them out in PR review.

[sffc]

TG2 discussion: https://github.com/tc39/ecma402/blob/master/meetings/notes-2023-10-12.md#should-ecma-402-spec-text-for-time-zone-canonicalization-refer-to-cldr-or-to-iana-as-authoritative-825

[sffc]

@justingrant The conclusion from the October meeting was that you would "draft a PR with a recommendation based on Option C that we can discuss further."