Implement spellchecking all translations in a particular language

[Changaco]

I don't see a way to do that in the UI, and the only issue I could find that mentions "spell" is #443, but it's about a particular tool that does more than a simple spellcheck.

[Changaco]

To illustrate, what I'm looking for is the Weblate equivalent of aspell -l $lang -x check "$file".

[Changaco]

Temporary workaround: an aspell wrapper script written in python which supports basic spellchecking of .po files (requires Babel). See also this superuser question.

[nijel]

Indeed spell checking is currently not there. I was hoping to have it with #443, but I'd still not sure about pulling Java into the chain just for this...

[Changaco]

Why pull anything in? Just use whatever is installed on the server: LanguageTool, aspell, or other; and warn the server admin if there isn't any of that, and also if the spellchecker only has a few language packs installed.

[nijel]

Yes that would be perfect, it just needs to be implemented ;-).

[Changaco]

I'm afraid I currently don't have the time to implement it myself.

[nijel]

Probably easiest approach would be to use pyenchant and it's SpellChecker. It has simpl API and uses whatever backend is available on a system.

[Bachstelze]

Pyenchant sounds good, but they are currently searching for a new maintainer. Are there any alternatives or should we help to maintain it?

[Changaco]

I don't recommend using pyenchant since it adds a dependency to the enchant C library, which is nothing more than a thin wrapper around various spellcheckers. That abstraction layer would be easier to maintain and distribute if written in pure Python.

In another project I ended up using https://pypi.org/project/hunspell/ instead of pyenchant, but Weblate needs a library that provides tokenizers (functions that split a text into words), whereas this hunspell module only provides functions that analyze a single word. pyenchant only includes an English tokenizer.

[Bachstelze]

We could use polyglot that provides tokenization for 165 languages. Hunspell looks great, but I don't know which languages and dictionaries are maintained.

[nijel]

In case we ever decide to go this way, I think integrating with something like LanguageTool is the way to go.

[nijel]

Duplicate of #443