[MrYawe]

Fixes the English fallback reported in #68.

Locale.parent/1 builds the root tag with cldr_locale_id: :und, then with_cldr_locale_id/1 validates "und" again. When :und isn't a supported locale, that validation resolves it to the default locale, so every parent chain ended at :en instead of :und. As a result, RBNF's reject_und_spellout_stub/3 never saw root, and English answered for any spellout rule set the locale lacked. For example, Rbnf.to_string(2_000_000, :spellout_cardinal, locale: :ru) returned "two million". with_cldr_locale_id/1 now keeps root as :und, the same way cldr_locale_id_from/1 already resolves a bare root tag.

The new tests in supported_locales_test.exs remove :und from the supported locales. They check that parent chains still end at :und, and that the ru request returns UnknownRbnfRuleError. Both fail on main. The RBNF test erases the cached ru chain first, because put_supported_locales/1 doesn't clear it.

This doesn't change the nb/no question in #68.

[kipcole9]

Thank you, this is correct and the tests are exactly the right ones.

I decided to fix it one level further back in f10f824, by making :und a supported locale whether or not it is configured, which also covers validate_locale("und") outside the parent walk and clears RBNF’s cached chains when the supported set changes.

Your test cases are in supported_locales_test.exs in that shape. With :und always supported the extra with_cldr_locale_id/1 clause is no longer reached, so I will close this in favour of f10f824.

Thank you again, much appreciated. Localize 1.5.0 will be out in the next 24 hours, including this fix.