Documentation on zero-copy deserialization of locale types.

Locale and LanguageIdentifier are highly structured types that cannot be directly stored in a zero-copy data structure, such as those provided by the zerovec module. This page explains how to indirectly store these types in a zerovec.

There are two main use cases, which have different solutions:

1. Lookup: You need to locate a locale in a zero-copy vector, such as when querying a map.
2. Obtain: You have a locale stored in a zero-copy vector, and you need to obtain a proper Locale or LanguageIdentifier for use elsewhere in your program.

Lookup

To perform lookup, store the stringified locale in a canonical BCP-47 form as a byte array, and then use Locale::strict_cmp() to perform an efficient, zero-allocation lookup.

To produce more human-readable serialized output, you can use PotentialUtf8.

use icu::locale::Locale;
use potential_utf::PotentialUtf8;
use zerovec::ZeroMap;

// ZeroMap from locales to integers
let data: &[(&PotentialUtf8, u32)] = &[
    ("de-DE-u-hc-h12".into(), 5),
    ("en-US-u-ca-buddhist".into(), 10),
    ("my-MM".into(), 15),
    ("sr-Cyrl-ME".into(), 20),
    ("zh-TW".into(), 25),
];
let zm: ZeroMap<PotentialUtf8, u32> = data.iter().copied().collect();

// Get the value associated with a locale
let loc: Locale = "en-US-u-ca-buddhist".parse().unwrap();
let value = zm.get_copied_by(|uvstr| loc.strict_cmp(uvstr).reverse());
assert_eq!(value, Some(10));

Obtain

Obtaining a Locale or LanguageIdentifier is not generally a zero-copy operation, since both of these types may require memory allocation. If possible, architect your code such that you do not need to obtain a structured type.

If you need the structured type, such as if you need to manipulate it in some way, there are two options: storing subtags, and storing a string for parsing.

Storing Subtags

If the data being stored only contains a limited number of subtags, you can store them as a tuple, and then construct the LanguageIdentifier externally.

use icu::locale::subtags::{Language, Region, Script};
use icu::locale::LanguageIdentifier;
use icu::locale::{
    langid,
    subtags::{language, region, script},
};
use zerovec::ZeroMap;

// ZeroMap from integer to LSR (language-script-region)
let zm: ZeroMap<u32, (Language, Option<Script>, Option<Region>)> = [
    (5, (language!("de"), None, Some(region!("DE")))),
    (10, (language!("en"), None, Some(region!("US")))),
    (15, (language!("my"), None, Some(region!("MM")))),
    (
        20,
        (language!("sr"), Some(script!("Cyrl")), Some(region!("ME"))),
    ),
    (25, (language!("zh"), None, Some(region!("TW")))),
]
.into_iter()
.collect();

// Construct a LanguageIdentifier from a tuple entry
let lid: LanguageIdentifier =
    zm.get_copied(&25).expect("element is present").into();

assert_eq!(lid, langid!("zh-TW"));

Storing Strings

If it is necessary to store and obtain an arbitrary locale, it is currently recommended to store a BCP-47 string and parse it when needed.

Since the string is stored in an unparsed state, it is not safe to unwrap the result from Locale::try_from_utf8(). See icu4x#831 for a discussion on potential data models that could ensure that the locale is valid during deserialization.

As above, to produce more human-readable serialized output, you can use PotentialUtf8.

use icu::locale::langid;
use icu::locale::Locale;
use potential_utf::PotentialUtf8;
use zerovec::ZeroMap;

// ZeroMap from integer to locale string
let data: &[(u32, &PotentialUtf8)] = &[
    (5, "de-DE-u-hc-h12".into()),
    (10, "en-US-u-ca-buddhist".into()),
    (15, "my-MM".into()),
    (20, "sr-Cyrl-ME".into()),
    (25, "zh-TW".into()),
    (30, "INVALID".into()),
];
let zm: ZeroMap<u32, PotentialUtf8> = data.iter().copied().collect();

// Construct a Locale by parsing the string.
let value = zm.get(&25).expect("element is present");
let loc = Locale::try_from_utf8(value);
assert_eq!(loc, Ok(langid!("zh-TW").into()));

// Invalid entries are fallible
let err_value = zm.get(&30).expect("element is present");
let err_loc = Locale::try_from_utf8(err_value);
assert!(matches!(err_loc, Err(_)));