Pydantic: GenericModel does not support generic subclasses

[DrPyser]

Bug

It seems the implementation of pydantic.generics.GenericModel does not support generic subclasses. Example:

import pydantic.generics
import typing

T = typing.TypeVar("T")

class ResponseModel(pydantic.generics.GenericModel, typing.Generic[T]):
    data: List[T]

class PaginatedResponseModel(ResponseModel[T]):
    total_count: int

ConcreteModel = PaginatedReponseModel[dict] # TypeError: Cannot parameterize a concrete instantiation of a generic model

PaginatedReponseModel is erroneously considered "concrete" even though type variables are used as parameters, and not concrete types.
It seems GenericModel.__class_getitem__ does not check if any parameters are TypeVar, like typing.Generic does.

To support generic subclasses, __class_getitem__ should check if all of the parameters are fully concrete type(i.e. any of them are typing.TypeVars, or any of them are generic types that have non-concrete type parameters). If not, a concrete model should not be created.

I would consider this a bug since it does not follow the behavior of other typing.Generic usages, but it could also be a feature request.

Please complete:

• OS: Ubuntu 18.04.1
• Python version import sys; print(sys.version): 3.7.1 (default, Oct 22 2018, 11:21:55)
  [GCC 8.2.0]
• Pydantic version import pydantic; print(pydantic.VERSION): 1.0

[dmontagu]

Yep this is definitely a bug.

In general, I think it should be possible to subclass even concrete generic models. I'm hopeful it's not too hard to implement.