Pydantic 2.10 handling for dataclasses that are fields of base models does not propagate allow_arbitrary_fields

[DiamondJoseph]

Pydantic 2.10 fixed a "bug" that dataclasses propagated ConfigDicts from classes that held them. This behaviour was used when we had plans that used dataclass Composite classes with devices as parameters.

Steps To Reproduce

Minimal example:

from collections.abc import Callable
from dataclasses import dataclass

from inspect import signature
from typing import Any, get_type_hints
from pydantic import create_model
from pydantic.config import ConfigDict
from pydantic.fields import FieldInfo

ModelConfig = ConfigDict(
    arbitrary_types_allowed=True,
)


class Foo:
    ...

@dataclass(config=ModelConfig)  # fails
class PlanParam:
    foo: Foo 

def plan(plan_param: PlanParam) -> int:
    return 0

def _type_spec_for_function(func: Callable[..., Any]) -> dict[str, tuple[type, FieldInfo]]:
    types = get_type_hints(func)
    return {
        name: (types.get(name), FieldInfo())
        for name in signature(func).parameters
    }  # type: ignore


create_model(
    plan.__name__,
    __config__=ModelConfig,
    **_type_spec_for_function(plan)  # type: ignore
)

Acceptance Criteria

• The above code works
• OR
• A supported alternative is documented

e.g.

from pydantic.dataclasses import dataclass

CompositeConfig = ConfigDict(arbitrary_types_allowed=True)

@dataclass(config=ModelConfig)  # works
class PlanParam:
    foo: Foo 

There may be additional config that is desired for Composite, e.g. frozen=True, extra=forbid, strict=True

[DiamondJoseph]

This behaviour is restored in Pydantic 2.10.1. However documenting how to use a Composite should still be done.