Skip to main content
Crowdin CLI coverage expands with discovery, file operations, string listing, and TM pre-translate. Projects can pin the CLI version in i18n.yml, bucket paths resolve relative to the config file for monorepo workflows, and extract is faster on large React Intl catalogs. View the GitHub release · Full changelog

​
Breaking changes

None.

​
Features

  • Add an optional version field to i18n.yml (for example hyperlocalise@1.10.0) so projects can pin the CLI version, similar to packageManager in package.json. Commands fail with a clear error when the running binary does not match the pin. GitHub Actions can install the pinned version with hyperlocalise-version: config. (#2171)
  • Expand hyperlocalise crowdin with discovery and resource commands: branch list / branch add, config sources / config translations, file list, language list, glossary list / glossary upload (TBX), and tm list / tm upload (TMX) with import polling and interruptible long imports. (#2124)
  • Add crowdin string list for source strings, path-based crowdin file upload / file download / file delete, and TM-only crowdin auto-translate (alias pre-translate) with async status polling. Machine translation and AI pre-translate remain deferred. (#2137)

​
Fixes

  • check, status, pack, sync, and run now resolve bucket file paths relative to the i18n config file’s directory instead of the process working directory. This fixes failures when running from a monorepo root against a nested project config (for example nested-app/i18n.jsonc with src/locales/en/common.json). (#2170)
  • Reject i18n glob matches that escape the config root via directory symlinks, so status, check, pack, and run cannot read or rewrite files outside the project. (#2226)
  • Reject out-of-range PHP octal escapes (for example \400) instead of silently wrapping to NUL during locale file parsing. Valid escapes such as \377 still decode. (#2225)

​
Performance

  • React Intl extract is about 28% faster on glob file resolution (1.77 ms → 1.28 ms), with 53% fewer allocations (5,266 → 2,466 allocs/op) and 33% less memory (326 KB → 217 KB/op). Message extraction is about 13% faster (684 µs → 592 µs); end-to-end extract is about 9% faster with 240 fewer allocations per operation. (#2177, #2199)
  • CSV marshaling saves about 163 KB per operation when no extra keys are present by skipping slice allocation and sorting. (#2162)
  • Eval scoring reuses cached language.Parse results and avoids intermediate match slices during tag token counting (81 → 75–78 allocs/op in BenchmarkEvaluatorEvaluate). (#2127, #2144, #2224)

​
Tests

  • Version pin enforcement across CLI commands, YAML loading, and runsvc integration.
  • Config-relative path resolution for check, status, pack, and run --dry-run from a foreign CWD.
  • Crowdin string list, file resolve/upload/download/delete, TM pre-translate polling, import edges, and file/directory/string pagination.
  • Glob and directory-symlink escape regression for status, check, pack, and run.

​
Maintenance

  • Upgrade github.com/gobwas/glob to v1 and github.com/yuin/goldmark to v2; migrate extract glob matching and Markdown parsing to the new APIs. (#2168)
  • Bump Go dependencies, including github.com/jackc/pgx/v5 to v5.10.0. (#2229, #2206, #2201, #2131)
  • Update the Go build image to 1.27.1. (#2228)