Time separator is also numbering system dependent?

[anba]

CLDR defines a numbering system dependent time separator in ldml/numbers/symbols/timeSeparator.

For example, da-u-nu-latn uses the time separator ".", whereas da-u-nu-arabext falls back to the root time separator "٫".

But when reading the timeSeparator definition in https://unicode.org/reports/tr35/tr35-59/tr35-numbers.html#Number_Symbols, it's not clear to me when ldml/numbers/symbols/timeSeparator should actually be used resp. if we should include the numbering system when looking up the time separator. UTS 35 has this note:

No CLDR locales are known to have a need to specify timeSeparator symbols that depend on number system; [...]. In the meantime, since CLDR data consumers can still request the timeSeparator symbol. it should match the symbol actually used in the timeFormats and availableFormats items.

The default (= "latn" numbering system [1]) time separator seems to match what's used in <timeFormats> and <availableFormats>, except for fr-CA:

• https://github.com/unicode-org/cldr/blob/46417d02b129b8c3dffce8688672eef2a1bf74d5/common/main/fr_CA.xml#L2021
• https://github.com/unicode-org/cldr/blob/46417d02b129b8c3dffce8688672eef2a1bf74d5/common/main/fr_CA.xml#L2088

fr-CA uses separate time separators for each time unit, i.e. "h", "min", and "s".

UTS 35 also defines durationUnit, which uses a single time separator which isn't numbering system dependent.

I'm currently a bit lost due to the conflicting definitions from UTS 35. Can we make this a bit more clear in the spec?

---

[1] "latn" numbering system and not <defaultNumberingSystem> for cases like ur-IN, which uses "arabext" as its default numbering system, but doesn't override ldml/numbers/symbols[@numberSystem="arabext"]/timeSeparator, so it's using the default from the root locale, which means "٫".

[sffc]

TG2 consensus: https://github.com/tc39/ecma402/blob/master/meetings/notes-2022-12-08.md#time-separator-is-also-numbering-system-dependent-108