I’m trying to use pydantic together with ABC. I created a base class Datasource and four subclasses:

from pydantic import BaseModel
from typing import Literal, Union, TypeVar
from devtools import debug
from abc import ABC, abstractmethod


class Datasource(ABC, BaseModel):
    source: str
    frequency: str

    @abstractmethod
    def hello():
        pass


class FirstStaticDatasource(Datasource):
    source: Literal["first"] = "first"
    frequency: Literal["static"] = "static"
    a: int

    def hello():
        pass


class FirstDynamicDatasource(Datasource):
    source: Literal["first"] = "first"
    frequency: Literal["daily"] = "daily"
    b: int

    def hello():
        pass


class SecondStaticDatasource(Datasource):
    source: Literal["second"] = "second"
    frequency: Literal["static"] = "static"
    c: int

    def hello():
        pass


class SecondDynamicDatasource(Datasource):
    source: Literal["second"] = "second"
    frequency: Literal["yearly"] = "yearly"
    d: int

    def hello():
        pass


DatasourceInterface = TypeVar('DatasourceInterface', bound=Datasource)
# DatasourceInterface = Union[tuple(Datasource.__subclasses__())]

class Settings(BaseModel):
    datas: list[DatasourceInterface]


if __name__ == "__main__":
    settings = Settings(datas=[
        FirstStaticDatasource(a=1),
        FirstDynamicDatasource(b=2),
        SecondStaticDatasource(c=3),
        SecondDynamicDatasource(d=4)
    ])

    debug(settings)
    json = settings.model_dump_json()
    debug(json)

    new_settings = Settings.model_validate_json(json)
    debug(new_settings)

If I try to run this I got the following error

TypeError: Can't instantiate abstract class Datasource without an implementation for abstract method 'hello'

Now, if I try to remove the abstract method, the code runs, but, in new_settings, datas results to be a list of the base class DataSource (i.e.: I cannot see a, b, c and d class members specified in the derived subclasses).

All the problems seems solved if, instead of

DatasourceInterface = TypeVar('DatasourceInterface', bound=Datasource)

I place

DatasourceInterface = Union[tuple(Datasource.__subclasses__())]

Is there a way to reach the same result without using “metaprogramming”?