chore: pin eslint-plugin-diff to workaround git ref issue

[acolombier]

Fix an issue where the remote ref is wrongly assumed to always be the PR base, instead of pre-commit from arg, regression introduce by eslint-plugin-diff@v2.1.1 (see paleite/eslint-plugin-diff@e1fd847)

[daschuer]

Is there an upstream bug report for this?

[acolombier]

I couldn't see one yet. Do you want to create one? Are you happy to merge this in the interim to unblock #16037 ?

[daschuer]

Could you file this issue? I am not so into it to put a good description.

[daschuer]

LGTM, Thank you for the additional documentation. I will merge this to unblock the depending PR.

It looks like this is an intended change form upstream. The situation that we are cut off from updates is not so good. Do we have a plan to change that?

[acolombier]

Good question. Quite frankly, I cannot tell. I guess the easy option would be to add a git fetch in the pipeline to make it work as they intend it to, not like pre-commit does.

One issue we have to work with is that pre-commit isn't really the leading linter for git pre commit checks and thus it is unlikely they have or would consider its behaviour. One question becomes what side effect are to be expected (e.g running locally, falling back to origin/main is just wrong as for most of us, it should at least fallback to upstream/main)

Not entirely sure what is the best option, probably trying to see if they would consider keeping this behaviour as opt out could be good?

[daschuer]

Yes. At least they should understand our issue.

[acolombier]

Ah I didn't realise we were passing the branch explicitly in the pipeline: 8a9b040
We need to test if we can pass HEAD^1