A few locale and value combinations wrongly report PluralCategory::Other

[janhohenheim]

The following code fails:

let pr = PluralRules::try_new_unstable(
    &icu_testdata::unstable(),
    &locale!("pl").into(),
    PluralRuleType::Cardinal,
)
.unwrap();
let result = pr.category_for(1_usize);
assert_eq!(result, PluralCategory::One);

This happens because the result returned is PluralCategory::Other. However, when I check the CLDR, it says this:

Which leads me to believe that while 1.0_f64 might very well be PluralCategory::Other, 1_usize should be PluralCategory::One.
Am I missing something? Disclaimer: I don't speak any Polish, I'm simply porting some tests that used a different library and expected a One here.

[zbraniecki]

Thank you for the report. Investigating.

[janhohenheim]

Update: I ported some more of the tested plurals of this test written in C# to my crate using icu and many of them fail, reporting PluralCategory::Other instead of what the C# library expected. A quick glance at the CLDR table makes me think these are also reported wrong, e.g. for Icelandic 1_usize is also Other despite the table clearly saying otherwise.

I'm still unsure, did I misconfigure something here? My dependencies look like this:

[dependencies]
icu_plurals = { version = "1.2.0", features = ["std"] }
icu_locid = { version = "1.2.0", features = ["std"] }
icu_testdata = { version = "1.2.0", default-features = false, features = ["std", "icu_plurals"] }
icu_provider = { version = "1.2.0", features = ["std"] }
fixed_decimal = { version = "0.5.3", features = ["std", "ryu"] }

[janhohenheim]

Update 2: I added some debug prints and got the following failures:

Cardinal:
locale: "pl", value: 1, expected One but got Other
locale: "pl", value: 2, expected Few but got Other
locale: "pl", value: 3, expected Few but got Other
locale: "pl", value: 4, expected Few but got Other
locale: "pl", value: 5, expected Many but got Other
locale: "is", value: 1, expected One but got Other
locale: "is", value: 21, expected One but got Other
locale: "is", value: 31, expected One but got Other
locale: "is", value: 41, expected One but got Other
locale: "is", value: 51, expected One but got Other

Ordinal:
locale: "cy", value: 0, expected Zero but got Other
locale: "cy", value: 7, expected Zero but got Other
locale: "cy", value: 1, expected One but got Other
locale: "cy", value: 2, expected Two but got Other
locale: "cy", value: 3, expected Few but got Other
locale: "cy", value: 4, expected Few but got Other
locale: "cy", value: 5, expected Many but got Other

Interestingly, all failures report Other when they shouldn't

[zbraniecki]

@janhohenheim I believe you are using testdata provider which does not contain Polish, and falls back on one of the available locales in the testdata provider.

Please, read this tutorial and let me know if it helps you understand how to build the data provider for your needs - https://github.com/unicode-org/icu4x/blob/main/docs/tutorials/data_management.md

[zbraniecki]

@sffc wdyt about printing a warning when using testdata in release mode?

[janhohenheim]

For the record, my confusion about this came from trying to work with the code seen in https://docs.rs/crate/icu_plurals/latest, which does not mention what testdata means. Since I was compiling in debug mode, such a warning would not have helped.

That said: I'm now generating a mod with the linked tool and it generated code that uses icu, which brings in dependencies for a lot of subcrates I don't need. Can I somehow tell it through features or something similar that I'm only interested in plurals? I know I can do this for the keys used in the generator, but I'm talking about the icu dependency that is being generated.

[sffc]

@sffc wdyt about printing a warning when using testdata in release mode?

I'm not opposed to just always printing a message to stderr when the testdata provider is used

[janhohenheim]

Mind helping me again? I generated the module according to the documentation and created the following smoke test:

#[test]
fn smoke_test() {
    let pr = PluralRules::try_new_unstable(
        &UnstableProvider,
        &locale!("en").into(),
        PluralRuleType::Cardinal,
    )
    .expect("Failed to construct a PluralRules struct.");

    assert_eq!(pr.category_for(5_usize), PluralCategory::Other);
}

But it fails with Failed to construct a PluralRules struct.: Data(DataError { kind: MissingLocale, key: Some(DataKey{plurals/cardinal@1}), str_context: None, silent: false })
I don't quite get how this can happen, since I pass a locale.

[janhohenheim]

Update: nvm, just found this answer

[zbraniecki]

(converting to discussion for now)

[zbraniecki]

marking as resolved!