Yes, the compile time is quite long when using all locales. And the memory requirement is also linear with respect to each locale - perhaps up to 1Mb per locale. So with over 700 locales that can be quite a lot of memory.

However a couple of things to consider:

1. While compile time is quite long (in the minutes range), this should only happen once on your development machine, and once in CI after which you would cache the relevant modules. And then once perhaps for production.
2. The full list of locales includes a lot of locales that are not in modern use. You could trim the list quite a bit to those in modern use. Of course if you are expecting to support formatting really all (as you noted) then this isn’t much help.

(I’m the author of. most of the ex_cldr_* libraries)

That’s what I have expected - that it won’t be simple.

You are really an inspiring fellow human being.

Yes, that’s a fair point indeed.

I am exploring (in my head) how a “version 3” might dynamically load locales. The complication is that number formatting (and some other areas like date time formatting) are actually driven by compile-time code generation. So it’s not as simple as just loading some data. So it’s not simple - and it’s not going to happen right now.

The locale data used to be structured in such a way that determining the “modern use” was straight forward. The data is organized differently now - I will dig into it and report back on a suggestion.