Subclass to a class inheriting from GenericModel - no validation at instantiation

[Esadruhn]

Bug

There is a bug, or maybe a feature request, with inheritance and Generics.

If I define a class that inherits from GenericModel and a subclass, then when I instanciate the subclass there is no check on the generic type.

From what I saw in the code, it is because the checks are done in the metaclass __new__ method (generics are treated as Optional[Any] at that point), and the generic type is defined afterwards, in the __getitem__ method.

Output of python -c "import pydantic.utils; print(pydantic.utils.version_info())":

             pydantic version: 1.4
            pydantic compiled: False
                 install path: $HOME/envs/thenv/lib/python3.7/site-packages/pydantic
               python version: 3.7.3 (v3.7.3:ef4ec6ed12, Mar 25 2019, 16:52:21)  [Clang 6.0 (clang-600.0.57)]
                     platform: Darwin-19.3.0-x86_64-i386-64bit
     optional deps. installed: ['typing-extensions']

This test passes - no inheritance

def test_case_no_inheritance():
    TypeX = TypeVar('TypeX')

    SampleTypeX = list

    class MyBaseClass(GenericModel, Generic[TypeX]):
        X: TypeX

    with pytest.raises(pydantic.error_wrappers.ValidationError):
        MyBaseClass[SampleTypeX](X=1)

Failed test case - inheritance

import pydantic
from pydantic.main import ModelMetaclass
from pydantic.generics import GenericModel
import pytest

def test_case():
    TypeX = TypeVar('TypeX')

    SampleTypeX = list

    class MyBaseClass(GenericModel, Generic[TypeX]):
        X: TypeX

    class ExampleMeta(ModelMetaclass):
        # This metaclass is not necessary for the test, it illustrates that we get the 'generic_type' in __getitem__
        def __getitem__(self, generic_type):
            return type('Example', (Example,), {'_my_types': (SampleTypeX, generic_type)})

    class Example(MyBaseClass[TypeX], metaclass=ExampleMeta):
        pass

    Example[SampleTypeX](X=[0, 1, 2])

    with pytest.raises(pydantic.error_wrappers.ValidationError):
        Example[SampleTypeY](X=1)

[samuelcolvin]

humm, not sure. @dmontagu any idea whether we want to fix this?

[dmontagu]

It works all right if you repeat the Generic[TypeX] (and drop the metaclass):

from typing import TypeVar, Generic

import pydantic
from pydantic.main import ModelMetaclass
from pydantic.generics import GenericModel
import pytest

def test_case():
    TypeX = TypeVar('TypeX')

    SampleTypeX = list

    class MyBaseClass(GenericModel, Generic[TypeX]):
        X: TypeX

    class Example(MyBaseClass[TypeX], Generic[TypeX]):
        pass

    Example[SampleTypeX](X=[0, 1, 2])

    with pytest.raises(pydantic.error_wrappers.ValidationError):
        Example[SampleTypeX](X=1)

test_case()

@Esadruhn does this help?

I think it might make sense to refactor this more fully at some point for better compatibility with the typing library, though there are challenges with that as noted in other issues (namely, the typing package has been rather aggressive about keeping certain functionality private/internal use only).

[Esadruhn]

Thank you for the answer, it works fine ! Pending a refactor, would it be possible to document this ?

[dmontagu]

Yes, I think adding docs about it is a good idea. If you wanted to submit a PR adding the docs that would be awesome, otherwise I'll get to it eventually.

[Esadruhn]

Thanks, I'll see if I can do it in the next few days.

I really like the inclusion of actual Python scripts in the doc !

make docs-serve works only with Python 3.8 as it has to build examples that use 3.8 features:

error in validation_decorator_parameter_types.py: Traceback (most recent call last):

pydantic/docs/examples/validation_decorator_parameter_types.py", line 20

`def pos_only(a: int, b: int = 2, /) -> str:  # python 3.8 only`

[samuelcolvin]

Yes, that's expected, I should be able to build in a check that avoids executing that example for earlier versions.

[PrettyWood]

Documentation added in #1249 to explain how to properly inherit from a GenericModel