[Aymericr]

What does this PR do?

The release workflow's translation gate, bun scripts/i18n.ts release-check, could not pick a base once a new locale was bundled. hasCatalogs required a catalog for every TARGET_LOCALES entry, and no earlier commit has the new locale's catalog. So after #1184 (Turkish) or #1200 (Italian) merges, the latest release tag, the origin/main fallback and the catalogs' first commit are all rejected, and the next release fails:

release-check: no base with bundled PO catalogs (@pascal-app/editor@1.0.7 has no catalogs; origin/main and the catalogs' first commit are unavailable or HEAD); pass --base <rev>

--base @pascal-app/editor@1.0.7 is rejected the same way. PR CI doesn't see this: i18n:check compares with the PR's base commit, where a missing locale is already allowed.

• A base qualifies when it has a catalog for any target locale (every → some in hasCatalogs). A base from before the catalogs is still rejected, by the default lookup and by --base.
• Locales with a catalog at the base are compared exactly as before. A locale without one has nothing reviewed there to keep: release-check names it (New locales since @pascal-app/editor@1.0.7, with no reviewed entries to keep: tr), and its compiled pack is still scanned for drafts.
• --base-dir <dir> with no target catalog now fails like --base (--base-dir <dir> has no bundled PO catalogs to compare with). Before, it compared with an empty base and passed.
• i18n/README.md: one sentence on new locales in the release-check paragraph.
• Test fixtures: the fixture repos turn off commit and tag signing. With signing configured globally, every fixture commit went through the signer (slow, or a prompt per commit; up to 15 s each here, past bun's 5 s test timeout), and with tag.gpgSign=true, git tag <name> fails for want of a message. Fixture directories are now removed after each test, as in the other script tests.

No catalog, translation or review marker changes here, and nothing in this PR reviews or approves a tr or it entry.

How to test

1. bun test ./scripts/i18n-literals.test.ts: 51 pass. The new tests cover:

   • a locale bundled after the latest release tag;
   • regressions in the other locales and drafts in the new locale's pack still failing;
   • --base predating a locale but not the catalogs;
   • a --base-dir without catalogs failing.

   The first three fail on main with the error above; the --base-dir one fails without its guard (exit 0 instead of 1).

2. In a throwaway clone, merge a locale PR and point origin/main at the merge, as the release run after that merge sees it:

   git fetch origin pull/1184/head:pr-1184
   git switch --detach origin/main && git merge --no-edit pr-1184
   git update-ref refs/remotes/origin/main HEAD
   bun scripts/i18n.ts release-check

   On main this fails with the error above. After git merge --no-edit origin/cursor/i18n-release-check-new-locale-b052, it prints Comparing with @pascal-app/editor@1.0.7 and New locales since @pascal-app/editor@1.0.7, with no reviewed entries to keep: tr, then passes. Same with Add Italian (it) as a bundled locale (unreviewed AI drafts) #1200 (it), and with both merged (tr, it).

3. In that tree, mark fr's reviewed editor.buildPanel.advanced as #, fuzzy with Review: required: release-check fails with fr: "editor.buildPanel.advanced" is reviewed on the base branch with the same English and must stay reviewed. Compiling a draft into the fr, tr or it pack fails with <locale>: compiled pack ships unreviewed "…".

bun run ci passed locally on be3f503: biome, i18n:check, skills validation, typecheck, package and root tests, build, and the production smoke test. On the final head, the root i18n suite (80 pass), i18n:check and bun run check pass locally, and CI passes.

Screenshots / screen recording

N/A — non-visual change.

Checklist

• I've tested this locally with bun dev
• My code follows the existing code style (run bun check to verify)
• I've updated relevant documentation (if applicable)
• This PR targets the main branch

Locale PRs

This unblocks the release after #1184 (tr) or #1200 (it) merges. It doesn't make either one ready:

• Against current main, i18n:check fails for both until their catalogs are refreshed. tr is missing 1,192 messages and has 37 obsolete; it is missing 129 and has 24 obsolete.
• Both edit the same TARGET_LOCALES line and the same locale lists in scripts/i18n-literals.test.ts, so the second to land has to merge those lines.
• Their fork CI runs are waiting for approval.
• Their drafts still need native-speaker review before they ship.

Review

• Codex's GitHub review ran when this PR was opened and finished on be3f503 with no findings. It hasn't re-run on the two follow-up commits.
• A fresh-context review by another agent approved be3f503. Its P2 (a --base-dir without catalogs passed silently) and two nits are addressed in 1e2ba69 and 8f97c54, which it then approved with no P0–P2 findings.
• Cursor Bugbot passed on 8f97c54 with no comments.

Next owner: Maintainer

  

---

Note

Medium Risk
Changes release workflow gating and base-selection logic for translation regressions; mistakes could let bad packs ship or block valid releases.

^{Reviewed by Cursor Bugbot for commit 8f97c54. Bugbot is set up for automated code reviews on this repo. Configure here.}

[chatgpt-codex-connector]

Codex Review Summary: ✅ Completed

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-11-09T01:25:51.362777Z	be3f503	PR opened

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.