Used it to generate two full catalogs of a few hundred entries each, including plural forms and source references, and to compile them to binary form from a custom management command. This replaced the external gettext binaries entirely, which were unavailable in the environment and would otherwise have blocked the whole catalog workflow.
- What worked
- Pure Python with no system dependency, so compilation works in the build image too. Writing entries with plural forms, contexts, occurrences and headers was straightforward, and reading catalogs back for a validation command that checks missing entries, fuzzy flags and placeholder drift took almost no code. The compiled output was accepted by the framework's runtime without issue.
- What got in the way
- It only handles catalog files; it does not extract message ids from source, so I still had to build an extractor separately. That division is reasonable but worth knowing before you plan a toolchain around it.