pin astroid version?

[skshetry]

Hi there. Would it be possible to pin the astroid version? It looks like in astroid==2.5.7, the inference was improved and started to notify errors that were false negatives before (which is a good thing).

But this makes CI fail in a surprising way. If the new release came with pylint, it'd have been easier to fix the errors. 🙂

[Pierre-Sassoulas]

I think it makes sense, beside, anticipating the effect of new astroids in old pylint is always a difficult exercise. We might have impacted the performance of pylint 2.8.2 because of #4524 and cannot expect everyone to downgrade to 2.5.6 manually if that really become a problem. What do you think @cdce8p ?

[cdce8p]

I agree, the situation isn't ideal. Usually it's not that big of a deal as we also release pylint shortly after, but as you mentioned we have to deal with #4524 first. The only downside to limiting it that releasing becomes a bit more time consuming, since it needs to be >= during the development period.

@Pierre-Sassoulas Do you consider releasing 2.8.3 just with astroid==2.5.7?

Long term we might want to consider vendoring astroid. That way we could update the astroid version pylint uses more frequently without the need for a new release. It would also negate the need for some pylint MRs to wait until the next release before we can merge them.

[stanislavlevin]

With the released Astroid 2.5.7 even Pylint's selftests fail...

[stanislavlevin]

Some mechanism of synced releases is required, IMO. Otherwise, CI is broken again :(

[Pierre-Sassoulas]

With the released Astroid 2.5.7 even Pylint's selftests fail...

Two tests are not passing but only because a false negative is now detected. See #4523

[Pierre-Sassoulas]

I released pylint 2.8.3 with astroid pinned to 2.5.6 and upgraded the release documentation so astroid is pinned for the next release in #4533.

[stanislavlevin]

Thank you!