Replace types-attrs with attrs

[cdce8p]

Blocked by python/mypy#14137

[hauntsaninja]

We already install types-attrs. I think the bug is a cache bug, so it won't repro in mypy_primer (which only does from scratch checking). Or at least, I wasn't able to repro with mypy_primer -k home-assistant — are you able to?

[AlexWaygood]

It might be worth removing the types-attrs import and replacing it with attrs regardless of the mypy bug, since types-attrs was deleted from typeshed in June 2021 after attrs introduced inline types, and home-assistant no longer appears to use types-attrs.

[cdce8p]

We already install types-attrs.

🤦🏻 I should have looked more carefully. I just assumed that it wasn't installed since I was able to crash mypy with just attrs or types-attrs installed and the primer didn't catch it.

I think the bug is a cache bug, so it won't repro in mypy_primer (which only does from scratch checking). Or at least, I wasn't able to repro with mypy_primer -k home-assistant — are you able to?

It is a cache bug but one that occurs on the first write / serialize. I'm able to reproduce it with --no-incremental. Interestingly though setting --cache-dir=/dev/null (like the primer does) hides it 🤔

It might be worth removing the types-attrs import and replacing it with attrs

Will do that.

[hauntsaninja]

It looks like there are several projects that use types-attrs instead of attrs, want to upgrade all of them?

[cdce8p]

I think the bug is a cache bug, so it won't repro in mypy_primer (which only does from scratch checking). Or at least, I wasn't able to repro with mypy_primer -k home-assistant — are you able to?

It is a cache bug but one that occurs on the first write / serialize. I'm able to reproduce it with --no-incremental. Interestingly though setting --cache-dir=/dev/null (like the primer does) hides it 🤔

Does it make sense to look for ways to remove the --cache-dir=/dev/null argument? Maybe if we replace it with --no-incremental?

[hauntsaninja]

Thank you, this looks good!

Good question. primer sets both --no-incremental and --cache-dir=/dev/null. I guess we could set some --cache-dir, we might just want to ensure that all the different concurrent mypys write to different caches. The cost of dropping --cache-dir=/dev/null is basically figuring that out, some extra CI slowness, and some personal dev workflows that I can adjust.

I do confess to being surprised at the existence of an issue this would have caught that wasn't caught by the regular mypy self check. So updating my priors, but still don't think it's critical if we don't do it :-)

[cdce8p]

It is a cache bug but one that occurs on the first write / serialize. I'm able to reproduce it with --no-incremental. Interestingly though setting --cache-dir=/dev/null (like the primer does) hides it 🤔

I've opened a PR to run serialize even with --cache-dir=/dev/null. Might be useful to have.
python/mypy#14155