Hermes ICU4X

lib.rs

use widestring::U16CString;

use icu_locid::LanguageIdentifier;

fn canonicalize_utf16_locale(input: &[u16]) -> Vec<u16> {
    let utf8_result = String::from_utf16(&input).unwrap();
    let loc: LanguageIdentifier = utf8_result.parse().unwrap();
    let utf16_data: Vec<u16> = loc.to_string().encode_utf16().collect();
    utf16_data
}

#[no_mangle]
pub extern "C" fn canonicalize_locale(input: *const u16, len: usize) -> *mut u16 {
    let input_slice = unsafe { std::slice::from_raw_parts(input, len) };
    let output = canonicalize_utf16_locale(input_slice);

    let c_string = unsafe { U16CString::from_vec_unchecked(output).into_raw() };
    c_string
}

#[no_mangle]
pub extern "C" fn free_canonicalized_locale(ptr: *mut u16) {
    if !ptr.is_null() {
        let _ = unsafe { U16CString::from_raw(ptr) }; // This will drop the U16CString, freeing the memory.
    }
}

PlatformIntlApple.cpp

// https://github.com/facebook/hermes/blob/main/lib/Platform/Intl/PlatformIntlApple.mm#L319-L353

vm::CallResult<std::vector<std::u16string>> canonicalizeLocaleList(
    vm::Runtime &runtime,
    const std::vector<std::u16string> &locales) {
  // 1. If locales is undefined, then
  //   a. Return a new empty List
  // Not needed, this validation occurs closer to VM in 'normalizeLocales'.
  // 2. Let seen be a new empty List.
  std::vector<std::u16string> seen;
  // 3. If Type(locales) is String or Type(locales) is Object and locales has an
  // [[InitializedLocale]] internal slot, then
  // 4. Else
  // We don't yet support Locale object -
  // https://402.ecma-international.org/8.0/#locale-objects As of now, 'locales'
  // can only be a string list/array. Validation occurs in normalizeLocaleList,
  // so this function just takes a vector of strings.
  // 5. Let len be ? ToLength(? Get(O, "length")).
  // 6. Let k be 0.
  // 7. Repeat, while k < len
  for (const auto &locale : locales) {
    // 7.c.vi. Let canonicalizedTag be CanonicalizeUnicodeLocaleId(tag).
    auto parsedOpt = ParsedLocaleIdentifier::parse(locale);
    if (!parsedOpt)
      return runtime.raiseRangeError(
          vm::TwineChar16("Invalid language tag: ") +
          vm::TwineChar16(locale.c_str()));
    auto canonicalizedTag = parsedOpt->canonicalize();

    // 7.c.vii. If canonicalizedTag is not an element of seen, append
    // canonicalizedTag as the last element of seen.
    if (std::find(seen.begin(), seen.end(), canonicalizedTag) == seen.end()) {
      seen.push_back(std::move(canonicalizedTag));
    }
  }
  return seen;

PlatformIntlICU4X.cpp

extern "C" {
    const char16_t* canonicalize_locale(const char16_t* input, int len);
    void free_canonicalized_locale(char16_t* ptr);
}

vm::CallResult<std::vector<std::u16string>> getCanonicalLocales(
    vm::Runtime &runtime,
    const std::vector<std::u16string> &locales) {
  std::vector<std::u16string> seen;

  for (auto& loc : locales) {
    auto new_locale_ptr = canonicalize_locale(loc.c_str(), loc.length());
    std::u16string canonicalizedTag(new_locale_ptr);
    if (std::find(seen.begin(), seen.end(), canonicalizedTag) == seen.end()) {
      seen.push_back(std::move(canonicalizedTag));
    }
    free_canonicalized_locale(const_cast<char16_t*>(new_locale_ptr));
  }

  return seen;
}

results-20240206

Apple:
~/projects/hermes〉time ./build_release/bin/hermes ./test.js
        1.85 real         1.82 user         0.00 sys

ICU4X:
~/projects/hermes〉time ./build_release/bin/hermes ./test.js
        2.33 real         2.29 user         0.00 sys

test.js

for (let i = 0; i < 1000000; i++) {
  let result = Intl.getCanonicalLocales(["pl-pl", "de-de", "it-IT", "sr-Cyrl", "ja", "en-Latn-us", "de-at", "es-419", "und", "zh-CN"]);
}

Apple: 1.82

Split of ICU4X:

0. no-op - 0.63

1. Just iterating over locales and writing to seen if not seen - 0.80

2. Allocating Vec<u16> in Rust from the passed &[u16] - 0.94

3. Executing U16CString::from_raw_unchecked().into_raw(Vec<u16>) -> *mut u16 - 1.43

4. Assembling std::u16string from *mut u16 - 1.47

5. Adding free_locale - 1.58

6. Adding String::from_utf16(&[u16]) in canonicalize_utf16_locale - 1.85

7. Adding LanguageIdentifier parsing from String in canonicalize_utf16_locale - 2.07

8. Adding U16CString::from_vec_unchecked(Vec<u16>).into_raw() -> *mut u16 in canonicalize_utf16_locale - 2.30