Odoo Module I18N

i18n workflow

Most OCA repos wire OCA/odoo-pre-commit-hooks' oca-checks-po --fix into their pre-commit suite (e.g. via invoke lint in Doodba-based projects), catching/auto-fixing on every commit touching .po/.pot files: po-requires-module (missing #. module: MODULE comment), po-python-parse-printf/po-python-parse-format (placeholder count mismatch between msgid and msgstr — a real runtime risk, not just style), po-duplicate-message-definition, po-duplicate-model-definition, po-syntax-error, and po-pretty-format (alphabetical order, 78-column wrap, clearing a msgstr identical to its msgid). Regenerate strings first (below), then run your lint/pre-commit suite to clean up formatting — don't hand-fix things it already auto-fixes.

Regenerating .pot/.po files

Never hand-edit .pot files. The underlying OCA tool is click-odoo-makepot (part of click-odoo-contrib):

click-odoo-makepot --addons-dir=<path> -m <module_name> \
  --msgmerge --purge-old-translations

Many project setups wrap this — e.g. Doodba-based projects expose invoke updatepot -m <module_name> (also --all/-r <repo_name>), which additionally stops Odoo first, strips volatile POT-Creation-Date/ PO-Revision-Date headers (so diffs stay clean), and runs the lint suite against the touched i18n/*.po* files automatically. Use your project's wrapper if it has one instead of reimplementing this by hand.

--msgmerge merges into existing .po files instead of clobbering translator work, and is what you want by default; only skip it if you deliberately want to regenerate from scratch (rare — it discards existing translations). Only pass --fuzzy-matching when you specifically want Odoo to guess-match similar strings as fuzzy.

Writing translatable code

• User-facing strings in Python: wrap with _("...") from odoo.tools.translate (or the model's env._), never build the string via f-string/%-formatting around the _() call — interpolate inside it (_("Hello %s", name)), not _("Hello %s") % name. A standard OCA .pylintrc (≥16) enforces this family of checks: translation-format-interpolation, translation-fstring-interpolation, translation-positional-used, translation-not-lazy, translation-contains-variable, translation-too-few-args/translation-too-many-args, translation-unsupported-format, translation-format-truncated.
• XML: translatable attributes (string, help, placeholder, ...) are translated automatically — don't add a manual t-translation unless doing something unusual with QWeb.
• Never concatenate translated fragments (_("Foo") + _("Bar")) — translators need the full sentence in one string to produce correct grammar in other languages.

OCA conventions

• .po files are typically synced with Weblate by the OCA infra on the actual upstream repos — avoid unrelated reformatting of .po files beyond what your regeneration tool itself produces, to keep diffs reviewable/mergeable there.
• Keep the module's own .pot up to date whenever translatable strings change — don't let it silently drift; regenerate it (e.g. invoke updatepot -m <module> in Doodba-based projects) as part of finishing any change that adds or edits user-facing strings.