[rubas]

Before

Cldr.Config.number_symbols_for/2 builds Cldr.Number.Symbol with struct/2, which drops every key the struct does not define. With ex_cldr_numbers 2.38.3 it drops these CLDR symbols:

# nb, nn, no, sv and their regional locales: 7 records
time_separator_alt_variant: "."
# every locale: 840 records (fr-BE "≃", nb "ca.")
approximately_sign: "~"
# fr-CH and de-AT: 1 record each
currency_decimal: ".", currency_group: "."

Problem

• Nobody decided to drop these symbols. They disappear because the struct has no field for them.
• When CLDR adds a symbol, ex_cldr drops it and nobody sees it.

After

@excluded_number_symbols [:time_separator_alt_variant]
@optional_number_symbols [:approximately_sign, :currency_decimal, :currency_group]

def number_symbol!(symbols) do
  unsupported = @optional_number_symbols -- Map.keys(struct(Cldr.Number.Symbol))
  struct!(Cldr.Number.Symbol, Map.drop(symbols, @excluded_number_symbols ++ unsupported))
end

Any other symbol that the struct does not define raises a KeyError.

Details

Symbol	Records	Decision	Why
time_separator_alt_variant	7	Left out by name	The "." that nb, nn, no, and sv use as an alternative time separator. No formatter reads it.
approximately_sign	840	Kept when Cldr.Number.Symbol has the field	Real data: in 60 records it differs from the "approximately" pattern (fr-BE sign "≃", pattern "≈{0}"). No ex_cldr_numbers version has the field yet.
currency_decimal, currency_group	1 each	Kept when Cldr.Number.Symbol has the field	fr-CH "." and de-AT ".". ex_cldr_numbers adds the fields in elixir-cldr/cldr_numbers#66.

• Compatibility: Cldr.Number.Symbol lives in ex_cldr_numbers, and ex_cldr does not depend on it. The optional symbols follow the struct, so this works with ex_cldr_numbers 2.38.3 and with the fields from Use CLDR currency separators when formatting a currency cldr_numbers#66, and the two can be released in any order.
• Check: test/number_symbol_test.exs builds the symbols of every locale in priv/cldr/locales (about 5 s), so a CLDR update with a new symbol fails the test suite. Other tests: sv leaves out its "." time separator, and an unknown symbol (digit_marker) raises.
• test/support/number_symbol.ex stands in for the ex_cldr_numbers 2.38.3 struct in the test env.
• ex_cldr_numbers mix test against this branch (path dependency): upstream main 5232 passed, Use CLDR currency separators when formatting a currency cldr_numbers#66 5234 passed. Its test backend compiles all 766 locales, which runs number_symbols_for!/2 for each.
• mix test passes (13359). mix format --check-formatted fails on main too (Elixir 1.20.4), in 21 files, on lines this change does not touch. The changed code is formatted.
• The ex_cldr_numbers backend (lib/cldr/number/backend/symbol.ex) builds the same struct with its own struct/2, so it still drops these symbols. It could call Cldr.Config.number_symbol!/1.

---

AI disclosure: AI models helped us find this issue, write the change, and review it.

[kipcole9]

Thanks for the PR, @rubas - the symbol table in your write-up is very useful. Since ex_cldr is now in maintenance mode I am trying to avoid changes users could see. This change doesn’t alter any output today: struct/2 already drops the same symbols. Its effect would be a new KeyError while consumers compile their backends if the data ever gains a symbol, and that’s a failure mode I’d rather not add.

elixir-cldr/cldr_numbers#66 doesn’t depend on it either, since struct/2 keeps the currency fields once Cldr.Number.Symbol defines them. So I’m closing this PR, but I’ll look at elixir-cldr/cldr_numbers#66 on its own merits. Thanks again!