Mixing Cyrillic and Latin languages breaks multi-language checking

[judaew]

When setting jinx-languages to include both a Cyrillic and a Latin-based language (for example, "en_US uk_UA"), only the Ukrainian (Cyrillic) part is actually checked. English words are ignored completely.

If I instead set two Latin-based languages (e.g. "en_US de_DE"), both work perfectly.

This suggests the issue might be related to how Jinx handles mixed input systems (Latin + Cyrillic alphabets).

I provide a Ukrainian sentence with two intentional spelling mistakes for testing:
"Сьогодні я бачив прекрастне небо над містом, воно було надвичайно яскраве"

System Info

• Emacs 30.2
• Jinx 20251013.1018
• OS Arch Linux
• Hunspell 1.7.2 (with en and uk dictionary)

[minad]

Thanks for the report. I can reproduce the problem, but I don't know how to fix it. The problem is that jinx--word-valid-p returns t for the Ukrainian dictionary for non-Cyrillic words. jinx--word-valid-p calls the native jinx_check, which calls enchant_dict_check, which finally calls into Hunspell. Maybe Hunspell Cyrillic dictionary ignores all non-Cyrillic words by default (which it should not do imo)? Do you have further debugging insights?

[minad]

It could be that it is somehow necessary to configure Hunspell to handle mixed alphabets differently. Alternatively Jinx could be modified to dispatch the word checking to "correct" dictionaries only, instead of trying all of them. But I don't immediately see how to implement this given the limitations of the Enchant API, see the file enchant.h.

[judaew]

I found the solution. The Aspell backend works perfectly. To use Aspell (without any configuration) you need to set the short two-character language codes. However, nuspell and Hunspell don't work correctly.