What is a key?

[faassen]

icu4x-datagen offers a --keys option. I'm interested in generating just collation information, but I need to cover a wide range of languages as I need to build a general collation lookup mechanism. My hope is that I can slice down the generated data by controlling the keys that get generated by using specific keys like collator/data@1. But I have in fact no idea what keys are required for what; is there a way for me to find out? I've done some digging but haven't found enlightenment yet.

(The alternative is that I generate Rust code and rely on dead code elimination. Would that work if I do care about all possible locales?)

[Manishearth]

Generally we just recommend using --keys-for-bin to have it detect the keys you need from the binary. If this is tricky in your situation we'd love to know!

I do have an open issue about being clearer about this #2685

Currently if you want to figure out which keys are needed, check out the _unstable constructors of each type you are constructing, look at the marker bounds, and see what the value of KEY is.

[faassen]

Thanks!

I'm building a library that implements a programming language that needs access dynamic to collations (language-driven). I figured --keys-for-bin was not going to work for that use case, but I'd love to be mistaken.

[Manishearth]

It will work: language-driven is fine, "keys" are for the type of data, not the language/locale stuff. You can still select the languages you need with the other flags on the datagen command line.

ICU4X is structured such that for a normal binary keys-for-bin will always contain all the keys that the program may need.

[sffc]

To add to @Manishearth's answer, --keys-for-bin is best if you get it to work, but to do it manually, first look at the docs:

then click on each marker bound, like CollationSpecialPrimariesV1Marker, and in the documentation it tells you the key string, like “collator/prim@1”