Gregorian#month_names and #weekday_names return partial data for 104 locales (en-CA, en-GB, en-IN, ...)

[nightsurgex2]

Summary

Worldwide::Calendar::Gregorian.month_names and .weekday_names return partial results for 104 locales, including en-CA, en-GB, en-IN, en-AU, en-NZ, en-ZA and every other en-001 descendant.

Worldwide::Calendar::Gregorian.month_names(locale: "en-CA", width: :abbreviated)
#=> ["Sept"]
# expected: ["Jan", "Feb", "Mar", "Apr", "May", "Jun", "Jul", "Aug", "Sept", "Oct", "Nov", "Dec"]

Worldwide::Calendar::Gregorian.weekday_names(locale: "se-FI")
#=> {mon: "mánnodat", tue: "disdat", thu: "duorastat", sat: "lávvordat"}
# expected: all seven, with sun/wed/fri inherited from `se`

Reproduced on worldwide 1.26.2 (current main, ad6eeefb), i18n 1.15.2, Ruby 3.4.8.

Cause

CLDR resolves locale data item by item: the effective data for a locale is the union of its own data with each of its ancestors', with the more specific locale winning per item (UTS #35 §4.1). Locale files in data/cldr/locales/ are stored as deltas accordingly — data/cldr/locales/en-CA/calendars.yml contains exactly:

months:
  stand_alone:
    abbreviated:
      9: Sept

I18n::Backend::Fallbacks#translate returns the first non-nil node it finds in the fallback chain. That is correct for leaves (months.stand_alone.abbreviated.1 still resolves to "Jan" from en) but wrong for a whole node: months.stand_alone.abbreviated matches en-CA's one-entry hash and the eleven inherited entries never get merged in.

month_names then calls .values on that hash, so the truncation is silent — lib/worldwide/calendar/gregorian.rb#L20.

Affected locales

805 known locales × 2 methods × 3 widths = 4830 combinations; 108 of them are wrong, spanning 104 distinct locales.

Method	Width	Count	Locales
month_names	:abbreviated	101	ar-DZ, ar-MA, ar-TN, en-001, en-150, en-AG, en-AI, en-AT, en-BB, en-BE, en-BM, en-BS, en-BW, en-BZ, en-CA, en-CC, en-CH, en-CK, en-CM, en-CX, en-CY, en-DE, en-DG, en-DK, en-DM, en-ER, en-FI, en-FJ, en-FK, en-FM, en-GD, en-GG, en-GH, en-GI, en-GM, en-GY, en-HK, en-IE, en-IL, en-IM, en-IN, en-IO, en-JE, en-JM, en-KE, en-KI, en-KN, en-KY, en-LC, en-LR, en-LS, en-MG, en-MO, en-MS, en-MT, en-MU, en-MV, en-MW, en-MY, en-NA, en-NF, en-NG, en-NL, en-NR, en-NU, en-NZ, en-PG, en-PK, en-PN, en-PW, en-RW, en-SB, en-SC, en-SD, en-SE, en-SG, en-SH, en-SI, en-SL, en-SS, en-SX, en-SZ, en-TC, en-TK, en-TO, en-TT, en-TV, en-TZ, en-UG, en-VC, en-VG, en-VU, en-WS, en-ZA, en-ZM, en-ZW, hi-Latn, hi-Latn-IN, se-FI, sr-Latn-ME, sr-Latn-XK
month_names	:narrow	3	ar-DZ, ar-MA, ar-TN
month_names	:wide	3	ar-DZ, ar-MA, ar-TN
weekday_names	:wide	1	se-FI

Worst cases return a single entry: hi-Latn and se-FI return 1 of 12 abbreviated month names; sr-Latn-ME and sr-Latn-XK return 5 of 12. en-GB and friends return 1 of 12.

Secondary issue: missing structural keys degrade into a plausible string

Worldwide::I18nExceptionHandler#degraded_translation (lib/worldwide/i18n_exception_handler.rb#L51-L57) turns any unresolved key into a humanized copy of its last segment. For a structural lookup that means a missing weekday table comes back as the string "wide" rather than raising:

Worldwide::Cldr.t("calendars.gregorian.days.stand_alone.wide", locale: :some_locale_without_data)
#=> "wide"

That is a reasonable default for leaf strings in a UI, but for a hash-valued CLDR node it converts a data gap into a plausible-looking value, and month_names then fails with NoMethodError: undefined method 'values' for an instance of String some distance from the cause. (This is how the same call sites failed on worldwide < 1.25.6, before the fiber-storage fix in 76762d1 restored the CLDR fallback chain on Ruby 3.2+.)

Proposed fix

Resolve structural CLDR keys with CLDR's inheritance rules — walk the fallback chain from the least specific ancestor to the most specific, deep-merging each locale's own contribution, and resolve <alias> nodes against the requested locale rather than the ancestor they were found in. Then have weekday_names / month_names raise if an entry is still missing rather than returning degraded data.

PR: #NNN

[nightsurgex2]

Draft PR with the fix and regression coverage: #582

[nightsurgex2]

Draft PR with the fix and regression coverage: #582