I am trying to use pydantic frozen dataclasses to create stable and validated data structures. In the process, it seemed like my code would benefit from using inheritance because a lot of methods were shared between multiple objects.

My question is the following, is there a way to use inheritance to define a parent class that has default fields, while the child class includes non-default ones. Using dataclass(frozen=True) is important to ensure easy serialisation.

The example is the following, I’m creating a parent class A and a child B. A has both default and non-default arguments. The main interest here is having the field A.y that is initialised by the validator using other fields that already passed validation. B requires initialisation of non-default fields.

Note that there are in reality multiple child classes that also share methods with A, that part not causing any issue with pydantic.

from pydantic import field_validator, Field, ValidationInfo
from pydantic.dataclasses import dataclass

@dataclass(frozen=True)
class A:
    x: int
    y: int = Field(default=0, init=False, validate_default=True, repr=False)

    @field_validator("x")
    def validate_x(cls, x):
        if x < 0:
            raise ValueError("x must be non-negative")
        return x

    @field_validator("y", mode="after")
    def validate_y(cls, y, info=ValidationInfo):
        y = info.data["x"] + 1
        return y
    
@dataclass(frozen=True)
class B(A):
    name: str
    z: int = Field(default=0, init=False, validate_default=True, repr=False)

    @field_validator("z")
    def validate_z(cls, z, info=ValidationInfo):
        z = info.data["x"] + info.data["y"]
        return z
    
b = B(x=1, name="name")
print(b)

This returns the following error,

Traceback (most recent call last):
  File "/Users/.../new_file.py", line 21, in <module>
    class B(A):
  File "/opt/miniconda3/envs/qcd/lib/python3.10/site-packages/pydantic/dataclasses.py", line 225, in create_dataclass
    cls = dataclasses.dataclass(  # type: ignore[call-overload]
  File "/opt/miniconda3/envs/qcd/lib/python3.10/dataclasses.py", line 1184, in dataclass
    return wrap(cls)
  File "/opt/miniconda3/envs/qcd/lib/python3.10/dataclasses.py", line 1175, in wrap
    return _process_class(cls, init, repr, eq, order, unsafe_hash,
  File "/opt/miniconda3/envs/qcd/lib/python3.10/dataclasses.py", line 1024, in _process_class
    _init_fn(all_init_fields,
  File "/opt/miniconda3/envs/qcd/lib/python3.10/dataclasses.py", line 544, in _init_fn
    raise TypeError(f'non-default argument {f.name!r} '
TypeError: non-default argument 'name' follows default argument

which I guess is a default python error raised because pydantic automatic __init__ puts arguments in order and a default argument is written before a non-default one.

My main issue until now is that I can’t assign new fields after validation, and hopefully I can also avoid writing validators for each child.

Is there a way to circumvent this issue?