Allow defining type for dynamically generated types.

[albertferras-vrf]

Feature
Add a way to allow defining or overriding types for dynamically generated types, such as pydantic's create_model.

Pitch
Pydantic provides a way to create models dynamically (create_model).
The problem is that mypy always complain about these models:

sample code (a.py)

from pydantic import create_model, BaseModel
from pydantic.fields import Field

GeneratedModel = create_model(
    "MyModel",
    some_field=(int, Field(default=0))
)


class MyModel(BaseModel):
    a: int
    model: GeneratedModel

mypy a.py

a.py:12: error: Variable "a.GeneratedModel" is not valid as a type  [valid-type]
a.py:12: note: See https://mypy.readthedocs.io/en/stable/common_issues.html#variables-vs-type-aliases
Found 1 error in 1 file (checked 1 source file)

Currently, you are forced to add # type: ignore[valid-type] on each line where this model is used, which is a problem because it adds a lot of noise but also disables the check for all variables in the same line as this type: ignore.

I've tried things like these, with no luck:

# attempt 1
GeneratedModel: type[Any] = create_model(...)

# attempt 2
GeneratedModel = create_model(
    "MyModel",
    some_field=(int, Field(default=0))
)

GeneratedModel = cast(type[Any], GeneratedModel)

# attempt 3
def is_pydantic_model(model: Any) -> TypeGuard[BaseModel]:
    return isinstance(model, BaseModel)

GeneratedModel = create_model(
    "MyModel",
    some_field=(int, Field(default=0))
)
is_pydantic_model(GeneratedModel)

But I always get the same [valid-type] warning everywhere where this type is used.

The actual real workaround today is to define the type in a different file and import from somewhere else, but that is already marked as bug .

edit: Fixed FieldInfo->Field

[Viicos]

Just a note: FieldInfo shouldn't be instantiated directly, you should use the Field() function instead.

Regarding the issue, mypy is compliant to the spec, and so are other type checkers (e.g. pyright raises an error as well). As such, it is probably best to open a discussion at https://discuss.python.org/c/typing/32.

[brianschubert]

I haven't looked closely at your use case, but I believe the behavior you're looking for is what mypy's get_dynamic_class_hook() plugin hook provides: https://mypy.readthedocs.io/en/stable/extending_mypy.html#:~:text=get_dynamic_class_hook

[Viicos]

Ah thanks @brianschubert I didn't know about this. @albertferras-vrf, if you are willing to, you can open a feature request in Pydantic to ask for get_dynamic_class_hook() to be implemented by the Mypy plugin. There's a high chance that we won't pick it up soon, but at least it will be tracked.

[albertferras-vrf]

Thanks for the quick responses. I will create feature request on Pydantic.

We were also exploring to dynamically generate stub files .pyi (stubgen?) because all our mymodel = create_model(...) do not change. However, in my case it'd be easier if we could set some mypy flag/setting and just treat those cases as type[Any] or trust the type we manually assign like mymodel: type[SuperClassOfMyNewType] = create_model(...)