`Pydantic` `self.model_dump()` doesn't pick up subclass instances passed to a field

Question

I am trying to use a custom repr method to make a readable output of the object. I want to type a field of the super class, so all subclass instances passed to the field are validated correctly. Although, the self.model_dump() in the repr method doesnt seem to work unless I have correctly and explicitly typed the same class in the definition of the field.

WORKING CODE (explicitly typed field)

from pydantic import BaseModel, Field
from pydantic.config import ConfigDict


class QueryParams(BaseModel):
    pass


class subQueryParams(QueryParams):
    test: str = "test"


class YourModel(BaseModel):
    model_config = ConfigDict(arbitrary_types_allowed=True)
    command_params: subQueryParams = Field()

    def __repr__(self) -> str:
        """Human readable representation of the object."""
        items = [
            f"{k}: {v}"[:83] + ("..." if len(f"{k}: {v}") > 83 else "")
            for k, v in self.model_dump().items()
        ]
        return f"{self.__class__.__name__}\n\n" + "\n".join(items)

YourModel(command_params=subQueryParams())

Returns:

YourModel

command_params: {'test': 'test'}

BUG CODE (changed the field type to a super class)

from pydantic import BaseModel, Field
from pydantic.config import ConfigDict


class QueryParams(BaseModel):
    pass


class subQueryParams(QueryParams):
    test: str = "test"


class YourModel(BaseModel):
    model_config = ConfigDict(arbitrary_types_allowed=True)
    command_params: QueryParams = Field()

    def __repr__(self) -> str:
        """Human readable representation of the object."""
        items = [
            f"{k}: {v}"[:83] + ("..." if len(f"{k}: {v}") > 83 else "")
            for k, v in self.model_dump().items()
        ]
        return f"{self.__class__.__name__}\n\n" + "\n".join(items)


YourModel(command_params=subQueryParams())

Returns

YourModel

command_params: {}

How can I make use of typing a super class, while having the desired outut form the first code example?

Hacky Solution:

    def __repr__(self) -> str:
        """Human readable representation of the object."""
        items = [
            f"{k}: {v}"[:83] + ("..." if len(f"{k}: {v}") > 83 else "")
            for k, v in self.model_dump().items()
        ]

        # Needed to extract subclass items
        if self.command_params:
            add_item = self.command_params.model_dump()
        for i, item in enumerate(items):
            if item.startswith('command_params:'):
                items[i] = f'command_params: {add_item}'
                break  # Assuming only one item with 'command_params:', stop after updating


        return f"{self.__class__.__name__}\n\n" + "\n".join(items)

Answer 1

This does not work in the simple way you suggested. How should Pydantic know which class to instantiate if only the base class is annotated? However you can work with a Union:

from pydantic import BaseModel, Field
from pydantic.config import ConfigDict
from typing import Union


class QueryParams(BaseModel):
    pass


class subQueryParams(QueryParams):
    test: str = "test"


class YourModel(BaseModel):
    model_config = ConfigDict(arbitrary_types_allowed=True)
    command_params: Union[QueryParams, subQueryParams] = Field()

    def __repr__(self) -> str:
        """Human readable representation of the object."""
        items = [
            f"{k}: {v}"[:83] + ("..." if len(f"{k}: {v}") > 83 else "")
            for k, v in self.model_dump().items()
        ]
        return f"{self.__class__.__name__}\n\n" + "\n".join(items)


YourModel(command_params=subQueryParams())

Which prints:

YourModel

command_params: {'test': 'test'}

Or even better you could work with a discriminated union. See https://docs.pydantic.dev/latest/concepts/unions/#discriminated-unions-with-str-discriminators

I hope this helps!