Fixing TZ exemplar city generation

[robertbastian]

#3181

CLDR 43 removes trivial display names entries, so iterating through display names will no longer work. We instead iterate through the BCP47/Tzid map.

The way this code deals with aliases seems to be unintentional, as it takes the alphabetically last value. I've preserved this behaviour, but it's something we should figure out.

[robertbastian]

@nordzilla could you shed some light on the alias logic?

[robertbastian]

I've rewritten the logic. It now uses the first TZDB ID in the alias field that has an exemplar city (previously I think it was implicitly the last), and falls back to using the last part of the TZDB ID (i.e. America/Los_Angeles -> Los Angeles) if there are no display names.

This leads to some data diffs, i.e. kipho changes from Kanton to Enderbury, because its aliases are Pacific/Enderbury Pacific/Kanton, Montreal is added through the fallback (it doesn't seem to have display names?), and English also gains Currie, Santa Isabel, and Honolulu through the fallback (other languages already have them, no idea why these aren't included in English in CLDR 42).

[robertbastian]

Kanton becoming Enderbury is wrong, but expected, as Pacific/Enderbury is listed before Pacific/Kanton. This problem is tracked in https://unicode-org.atlassian.net/browse/CLDR-14453. en works around it by having removed the display name for Pacific/Enderbury, so that Pacific/Kanton is used.

[sffc]

Note: There are 2 definitions of the canonical IANA alias for each time zone; one is the CLDR alias, which is more historical and always stable, and the other is the IANA alias, which tends to be more modern but is subject to change. The order of the aliases in the bcp47 file does not seem to correspond to whether it is the CLDR alias or the IANA alias. I know there are open issues about this, as well as an ECMA-402 proposal in the works to settle on whose alias is the one the industry should use.

[robertbastian]

There's https://unicode-org.atlassian.net/browse/CLDR-14453

[robertbastian]

I'm not sure if we should include the last commit, it might lead to problems in the future. We should fix the data instead.

[robertbastian]

It already doesn't work with CLDR43 as that doesn't allow us to distinguish between explicitly set and key-derived values.

[sffc]

I'm not sure if we should include the last commit, it might lead to problems in the future. We should fix the data instead.

Until CLDR gets their act together, any of the following are reasonable behaviors:

1. Get the name for the first alias; err if it cannot be found (special case montreal)
2. Get the name for the first alias; don't add anything to the icu4x data struct if it cannot be found
3. Get the name for the first alias that has a name

I think the PR is currently implementing 1, right?

[robertbastian]

This does:

• Get the name of the first alias
• Get the city for the alias, if found return
• If the alias starts with Etc, skip
• If the alias is a single segment, skip
• Use the last segment of the alias

The last case currently only happens for the three English names that get added here, but will happen a lot in CLDR 43