In languages like Vietnamese, the diacritics are important in determing the sorting order (something like CollationDiacriticsV1Marker ) so loading additional data may help accuracy and might change the order.

To fix this, I could use something like (pseducode)

let local_provider = MultiForkByKeyProvider::new(vec![]);

Collator::try_new_with_buffer_provider(
    &local_provider, 
    LOCALE.into(), 
    CollatorOptions::new()
);

Wheather I am going in the correct direction or not, can you elaborate on this issue?

We have some tests here already: https://github.com/unicode-org/icu4x/blob/main/components/collator/tests/tests.rs

We should add some more tests there. This issue is to focus on the locales that load non-root data. You can see which locales do this by looking in the baked data files:

https://github.com/unicode-org/icu4x/tree/main/provider/baked/collator/data/macros

For example, in collator_dia_v1.rs.data, there is: static KEYS: [&str; 2usize] = ["und", "vi"]; which means that vi is the only locale in this case that uses non-root data. Ideally you can find a pair of strings that have different ordering in und versus vi due to this difference. This helps prove that the vi data supplement is being loaded correctly.

It would be good to have at least one test per data supplement. Some of these might already be covered, but we know vi is not covered due to #4165.

Ideally you can find a pair of strings that have different ordering in und versus vi due to this difference.

Now I get it, test cases should like this:

assert_eq!(collator.compare("à", "á"), Ordering::Less);       // For "vi"
assert_eq!(collator.compare("à", "á"), Ordering::Greater);    // For "und"

@ashu26jha Do you consider this issue complete?

I personally feel not to close this one as there are few TODOs at the end of the file which needs to be completed. For example writing the test for Tibetan, "sr and sr-Latn" etc.

Since #4633 was my first PR, I wanted to keep it short and easy to review ✌🏼