Pydantic: Feature Request: generic/parameterized models

[DrPyser]

Feature Request

It would be nice to be able construct models by composition, instead of just static descriptions and inheritance-based extension. What I mean by "composition", is being able to define a partial, parameterized model, which can yield new, complete models/validators by combining with other models/validators. This could be done by exploiting something like typing.Generic, e.g.:

import typing
import pydantic

T = typing.TypeVar("T")

class Response(pydantic.GenericModel[T]):
    data: T = ...
    total_count: int = None

and then, one could use Response[MyModel] or Response[List[MyModel]] as validators.

I think support for generic is mentioned in https://github.com/samuelcolvin/pydantic/issues/380.

[samuelcolvin]

looks like this would be extremely difficult to accomplish.

You could achieve roughly the same thing with create_model it would be much uglier but more powerful.

Unless lots of people want this, I'd rather not.

Comment here if you really need this.

[samuelcolvin]

You could achieve roughly the same thing with create_model it would be much uglier but more powerful.

Or of course by simple inheritance if you only need a few/defined variants of Response.

[DrPyser]

Yes, I'm taking the approach of using a function to define that kind of model dynamically, e.g.

import pydantic
from typing import Union, Type
def response(data_type: Union[type, Type[pydantic.BaseModel]]) -> Type[pydantic.BaseModel]:
    return pydantic.create_model(getattr(data_type, "__name__", str(data_type))+"Response", data=(data_type, ...), ...)

I guess this is fine.

[samuelcolvin]

Too much work for me unless someone can find an easy way of doing this or wants a big job...

[samuelcolvin]

As a result of @dmontagu's magic it looks like this should be possible.

[DrPyser]

Wow, that's awesome. Thanks @samuelcolvin, @dmontagu .