Add icu_preferences util

[zbraniecki]

No description provided.

[zbraniecki]

This is ready for review.

The deps in components have been cleaned up and documented, the util iteself is not yet well documented as I prefer to first get aligned on the API.

[robertbastian]

Looks good overall

[sffc]

Not a full review yet, just spot-checking

[sffc]

Discussion notes:

• @sffc - The list of collator keys is not all that we support. And we should be able to support other custom collator keys.
• @zbraniecki - The list is easy to extend; I would like to make the list more comprehensive when we actually add the collator preferences. As far as adding a custom variant, that's definitely something we could consider. I would like to defer those specific questions to later because they are compatible with the architecture of icu_preferences.
• @robertbastian We use TinyStr a lot internally. But you want to switch to using Subtag.
• @zbraniecki - My preference would be to converge around Subtag, because that's an ICU4X type.
• @robertbastian Subtag is slightly different; it forces lowercase.
• @sffc And it requires 2 characters.
• @robertbastian Maybe we could promote them from being doc(hidden) to being experimental.

[sffc]

Discussion: I don't see a good reason to have a moving into_single_subtag function since Subtag is Copy. Seems like as_single_subtag(&self) would be fine.

[zbraniecki]

The value is that you don't need to copy even - you destruct into a single Subtag. Let me know if that convinces you, I'll replace it with as_single_subtag().copied() otherwise.

[sffc]

Maybe just make as_single_subtag return Option<Subtag>.

[sffc]

Optional: rename the function to_single_subtag.

[zbraniecki]

as per https://rust-lang.github.io/api-guidelines/naming.html#ad-hoc-conversions-follow-as_-to_-into_-conventions-c-conv

as_ -> Free       ->  borrowed -> borrowed
to_ -> Expensive  ->  borrowed -> borrowed
                      borrowed -> owned (non-Copy types)
                      owned -> owned (Copy types)
into_ -> Variable ->  owned -> owned (non-Copy types)

We have:

• as_single_subtag(&self) -> Option<&Subtag>
• into_single_subtag(self) -> Option<Subtag>

I think as_ and into_ fit - to_ could work if source was Copy, but it's not, so passing it has to consume (to_ should take reference).

[sffc]

Well it's arguable that it is working on Copy types because it returns Some if the thing is a single Subtag (Copy). But I don't feel strongly about this and it need not block the PR

[zbraniecki]

I checked that. The closest reference in Stdlib I could find it Cow which has:

• to_mut(&mut self) -> &mut <B as ToOwned>::Owned
• into_owned(self) -> <B as ToOwned>::Owned (vs. OSstring::to_owned(&mut self) -> T`.

The latter has also the following description:

Creates owned data from borrowed data, usually by cloning.

So my read is that to_ is meant to be expensive, it takes a borrowed value. Our case is cheap, it takes an owned value.

I agree that it's not a clear cut, but I don't take &self and I don't clone. to_X(self) -> T is only for primitive types such as f64::to_bits.

I'll keep into_*.

[sffc]

My comment last time was a "here and elsewhere" but the elsewhere didn't happen :)