Move `icu_plurals::rules::reference` to datagen?

[robertbastian]

Seems to be datagen-only code?

Both icu_plurals::rules and icu_datetime::pattern have parallel reference and runtime modules, presumably reference is not used at runtime?

[sffc]

@zbraniecki

[sffc]

In 2.0 I'm adding https://unicode-org.github.io/icu4x/rustdoc/icu_datetime/neo_pattern/struct.DateTimePattern.html which is a polished API around datetime::pattern::reference. I don't know if we need/want a public API around plurals::pattern::reference. I'm not super happy with it being doc(hidden).

[zbraniecki]

The reason I added reference is that it allows for canonical parse/edit/serialize loopback on the rules/patterns.

This is used by datagen but the target audience is wider - customers who want to manipulate a Plural Rule should be able to acquire this parser/serializer pair from icu_plurals. I'm ok for this to be behind a flag, but I'd be against moving it out of the respective crate.

[sffc]

@zbraniecki, do you want the whole module exported with all its (fairly deep) AST APIs and things, or would you be happy with a high-level API with just a few functions like parse and serialize?

[robertbastian]

• @robertbastian I did a bunch of effort to add from_str impls, but I didn't do them here. If the only purpose for these types is datagen, they should live in datagen.
• @zbraniecki The target audience of these modules is someone who wants to operate on the Unicode plural rule format.
• @robertbastian As soon as you want to do that, you should enable the datagen feature.
• @sffc What I did in DateTime was create a special type driven by use cases. https://unicode-org.github.io/icu4x/rustdoc/icu_datetime/neo_pattern/struct.DateTimePattern.html
• @robertbastian It's inconsistent across components. There are some components where we have these nice APIs, and others where we don't and just put it in the datagen crate.
• @zbraniecki The use case is, like, a client who wants to go and make their own data tooling. I'm a software engineer, and I have an XML file with CLDR. The ticket is, take this data and change M to N. I need to write a little script to create an AST and transform it.
• @sffc I've had experiences working with clients who are tempted to reach for these to do the wrong thing in datetime, which is why I made the higher-level abstraction with guardrails.
• @zbraniecki I think Unicode should provide this functionality. I don't know if it needs to be in ICU4X itself.
• @robertbastian Maybe we make a new crate that includes the CLDR Serde impls and also these functions. cldr_utils? Or just cldr?
• @sffc There are types that are potentially shared with runtime crates, such as PatternItem and Field.
• @robertbastian It seems cleaner to have a CLDR type and a runtime type, anyway.

Conclusion for 2.0:

• Make sure these types are private or doc(hidden), or in general that this issue does not block 2.0
• Work toward the general direction of moving these to another crate, but this doesn't need to block 2.0

LGTM: @sffc @robertbastian @zbraniecki