Shane F. Carr
Shane F. Carr
@robertbastian confirms consensus on the first bullet point: "exemplar_chars lives in icu_locale" but not bullet points 2-3.
This is still in the 2.0 critical path because we are not in alignment about whether adding it as a default feature in the future is a breaking change.
Is that issue resolved if we move forward with the previously proposed `icu_locale_fallback` crate?
The issue was in the data crates. The data crates can use `icu_locale_fallback` instead of `icu_locale`, avoiding the circular dependency. Right?
My understanding of the issue you raised was `icu_locale -> icu_locale_data -> icu_locale`, which we can change to `icu_locale -> icu_locale_data -> icu_locale_fallback`
About the exemplar chars feature: I discussed it with @robertbastian and I'm convinced by this argument: "we don't feature-gate classes in other crates that we don't plan to split out...
You're right. The one user we know we have for exemplar chars uses it for spell checking. I corrected my list to say "spell checkers" instead of "layout engines". I...
> I still don't see how having these features is not going to be messy, but if I'm overruled I'm overruled. I prefer to achieve consensus and mutual understanding of...
Suggestion: something I needed in #5510 is a dataful enum ULE. Maybe `OptionULE` could be generalized.
So the most basic dataful enum representation is to spend a byte on the discriminant and then have a single VarULE value (we should consider supporting regular ULEs, too, but...