feat(i18n): skip do-not-translate entries in backfill_po

The backfill only checked for an empty msgstr, so it filled entries a human
translator had deliberately left blank — e.g. the ru catalog's "# Не переводить"
(do-not-translate) markers on literal tokens. This translated strings that must
stay identical to the source: icon names (`bolt` -> the  Explore control
icon), d3 enum values (`step-after`/`step-before`), API field names
(`error_message`), SQL keywords, code constants, and example placeholders.

Add `_is_do_not_translate`, consulted when collecting untranslated entries. An
entry is skipped if its msgid is in the curated `DO_NOT_TRANSLATE` set (seeded
from the ru translator's markers; language-independent) or if it carries an
explicit do-not-translate translator comment in any catalog. Skipped entries are
reported and left untranslated (source-token fallback).

Adds unit tests for the curated set, translator-comment honoring (incl. the
Cyrillic marker), and the normal-entry pass-through.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
This commit is contained in:
Amin Ghadersohi
2026-07-01 18:24:21 -07:00
parent 8be255de40
commit fb8af9caec
3 changed files with 95 additions and 0 deletions

View File

@@ -342,6 +342,12 @@ uses Claude AI to generate draft translations for any missing entries. All
AI-generated strings are marked `#, fuzzy` and tagged with an attribution
comment so that human reviewers know they need to be checked before merging.
The script never touches entries that must stay literal — msgids in its curated
`DO_NOT_TRANSLATE` set (icon names, enum values, SQL keywords, API field names,
example placeholders) and any entry carrying an explicit do-not-translate
translator comment (e.g. the ru catalog's `# Не переводить`). Those are left
untranslated so they fall back to the source token.
#### Prerequisites
```bash