language detection error

[adam-moss]

Just had the following output from commitlint, where it has incorrectly detected the language:

input: feat(commitlint-config-default): add default commitlint rules
⚠   commit must be in languague "eng", was one of: fra, cat, src, spa, sco [lang]
⚠   found 0 problems, 1 warnings

Running npx franc-cli "feat(commitlint-config-default): add default commitlint rules" yields cat as the result 😖

(Also note the minor typo in the error message)

[marionebl]

Yeah, all the rules based on natural language processing are rather flaky. I see the following options:

• Experiment with remote API solutions and see if those yield better results
• Deprecate, then remove the NLP rules: body-tense, footer-tense, lang, subject-tense

[adam-moss]

So for my use-case the first option would be a no-go because of an air gapped environment.

To be honest I'd leave them as is, it doesn't matter too much if they're in warning mode, but might be worth updating the docs to say you might get problems in error mode, if the text sample is relatively short ?

[marionebl]

Thanks for bringing the unknown constraints of development environments to the table, that rules out remote API usage.

My goal for commitlint is being as seamless as eslint, possibly adding autofix capabilities and commit message suggestions in the very long run. Keeping around flaky rules will prevent the tool from achieving that.

That being said I'd do this by deprecating them in some of the 2.x releases by warning on stderr if they are used and remove them with the 3.0 release the earliest.

[marionebl]

Typo fixed at marionebl/commitlint@fd56dff, will ship with next release

[morewry]

I ran into this too immediately after initially setting up.

⧗   input: chore(*): suggest project setup
⚠   ...
⚠   commit must be in language "eng", was one of: src, fra, spa, por, bos [lang]
⚠   found 0 problems, 2 warnings

You'd said back in July you planned to address this in 3.0, did that not happen? Every commitlint dependency I have is > 3, I'm using these:

"@commitlint/cli": "^4.1.1",
 "@commitlint/config-angular": "^3.1.1",
"@commitlint/config-lerna-scopes": "^4.1.1",

Update: If there is a way to suppress warnings and only output errors that would be very useful to me for both this issue and for https://github.com/marionebl/commitlint/issues/78. It would hide the noise from the rest of the team. I didn't see anything in the docs that looked like such an option...is there one?

[marionebl]

Due to other breaking changes I bumped to 3.0 earlier - the plan is to remove the functionality because the false positives are to frequent for texts as short as commits.

The removal will happen with one of the next major bumps, one of the releases before then will contain new deprecation warnings.

Edit: Currently there is no easy way to filter warnings from the output. As we are discussing commitlint bugs here I think we should not hide the symptom (the warnings) but fix the actual issue. For #78 there is a fix underway, and the nlp rules will be removed in the near future.

[lostintangent]

I just recently adopted commitlint, and the lang and subject-tense rules result in false negative warnings for every commit. I'd definitely love to see them removed/improved, so that my commit lint runs were "clean". Ideally, any error or warning would be actionable, so that they're worth paying attention to.