Pydantic V1 doesn't support mypy V1.1.1

[samuelcolvin]

We are aware that Pydantic doesn't support V1.1.1 of mypy.

This is a tracking progress of fixes for this

Issues so far:

• Mypy plugin crashes when encountering from_orm method on user's classes #5187
• Mypy regression - Missing named argument #5190
• Mypy plugin crashes with explicit_package_bases set as true #5191

AFAIK v1.0.1 is pretty well supported by pydantic v1.10.6, if you find a problem with v1.0.1 and pydantic, please report it here.

Please don't create more issues about mypy >v1.0.1, instead comment here if you think the above issues don't cover the problem

Fix timeline

We currently working very hard on pydantic V2 and the significant complexity of pydantic V2 AND the pydantic mypy plugin means that switching back and forth between the two has a significant overhead.

Realistically we won't work on fixing the above for around a month, we will of course try to review community fixes.

My recommendation is to pin mypy==1.0.1 or earlier.

Partially related, Pydantic V2 will probably support mypy>=1.2, and we'll likely add a optional dependency group pydantic[mypy] just to constrain mypy, see python/typeshed#9771 (comment).

[hatem-goc]

In #5190 I found that version 1.10.2 of pydantic works well with mypy 1.1.1 YMMV

[dmontagu]

I believe all reported issues with using mypy 1.1.1 with pydantic v1 have been resolved, and fixes are included in the 1.10.7 release today. However, we don't currently run tests in CI againt mypy 1.1.1, so I'm not totally sure this is the case.

Please continue to report any issues you hit with mypy 1.1.1 (whether using pydantic v1 or v2), and I will try to address them as I can. Many of the issues tracked here were actually also issues with earlier versions of mypy and/or required fixes in v2 as well, so it's very helpful.

Also: I don't think we should "close" this issue until we have tests against mypy 1.1.1+ running in CI.

[trim21]

I'm running mypy 1.1.1 with pydantic 1.10.7 in ci， it still failed because of pydantic, I'll create a issue with details later.

[markedwards]

This code is still not working with mypy 1.1.1 and pydantic 1.10.7:

from pydantic import BaseSettings, Field


class Settings(BaseSettings):
    foo: str = Field(..., env='foo')


settings = Settings()  # error: Missing named argument "foo" for "Settings"

[dmontagu]

@markedwards I assume you meant to inherit from BaseSettings rather than BaseModel.

I think this is going to be impossible to resolve without using the pydantic mypy plugin due to the fact that mypy now respects the @dataclass_transform decorator (and this is important for IDE integrations), and this applies to any class that inherits from BaseModel (and therefore any class that inherits from BaseSettings).

It does appear to be working correctly if you use the pydantic mypy plugin though.

[dmontagu]

Okay, actually, it looks like there may be a good solution: you can do

class Settings(BaseModel):
    foo: str = Field(default=..., env='foo')

this has the same runtime semantics but the @dataclass_transform will think there is a default value. In particular, I don't get mypy errors with the code written like this.

[markedwards]

@dmontagu yes, corrected the example. Actually, you're right it's working okay with the plugin, and with the modification you suggest. I had thought the plugin was enabled in my environment but it was not.

Thanks.

[dmontagu]

Once #5167 is merged, I think we will have addressed all reported issues with pydantic and mypy 1.1.1 (in both v1 and v2).

@trim21 if you are hitting a new bug I'm definitely interested to resolve it, please create an issue and I'll do it as soon as I can.

[trim21]

@dmontagu I've create a new issue #5259

[dmontagu]

I'm going to close this now, though will still note that I won't be surprised if there are issues.

In general, for new versions of mypy, our goal will be to support them in v2 and backport fixes as best we can, but make no promises that we will maintain the integration of pydantic v1 with future mypy versions.