model_validate_json() doesn't handle pydantic model fields

[SonOfLilit]

First Check

• I added a very descriptive title here.
• I used the GitHub search to find a similar question and didn't find it.
• I searched the SQLModel documentation, with the integrated search.
• I already searched in Google "How to X in SQLModel" and didn't find any information.
• I already read and followed all the tutorial in the docs and didn't find an answer.
• I already checked if it is not related to SQLModel but to Pydantic.
• I already checked if it is not related to SQLModel but to SQLAlchemy.

Commit to Help

• I commit to help with one of those options 👆

Example Code

# test_pydantic/test_model_in_model.py

import json
from typing import Optional

import pytest
from pydantic import BaseModel, ValidationError
from sqlmodel import Field, SQLModel, Column, String


class A(BaseModel):
    field: str


class Model(SQLModel):
    a: A


class Table(SQLModel, table=True):
    id: Optional[int] = Field(primary_key=True)
    a: A = Field(sa_column=Column(type_=String))


def test_model_in_model() -> None:
    Model.model_validate(Model(a=A(field="f")))
    with pytest.raises(ValidationError):
        Model.model_validate(Model(a=A(field=1)))  # type: ignore
    Model.model_validate_json(json.dumps({"a": {"field": "f"}}))
    with pytest.raises(ValidationError):
        Model.model_validate_json(json.dumps({"a": {"field": 1}}))

    Table(a=A(field="f"))
    with pytest.raises(ValidationError):
        Table(a=A(field=1))  # type: ignore

    # double-validate works
    Table.model_validate(Table.model_validate_json(json.dumps({"a": {"field": "f"}})))
    with pytest.raises(ValidationError):
        Table.model_validate(Table.model_validate_json(json.dumps({"a": {"field": 1}})))

    # normal validate doesn't (as of writing this, I have a fix)
    Table.model_validate_json(json.dumps({"a": {"field": "f"}}))
    with pytest.raises(ValidationError):
        Table.model_validate_json(json.dumps({"a": {"field": 1}}))

Description

Without a fix e.g. like below, SQLModel.model_validate_json() doesn't recursively parse fields of type pydantic.ModelBase. I have a branch with a test and this simple fix and @tiangolo if you're interested I'd love to contribute it.

The core issue is that we're not doing the with partial_init(): in sqlmodel_validate():

        if not is_table_model_class(cls):
            new_obj: _TSQLModel = cls.__new__(cls)
        else:
            # If table, create the new instance normally to make SQLAlchemy create
            # the _sa_instance_state attribute
            # The wrapper of this function should use with _partial_init()
            with partial_init():
                new_obj = cls()

A more performant fix would be to try and duplicate what's been done in sqlmodel_validate(). I can try to do that instead if you ask.

This is related to this comment #453 (comment), but I hope my description sheds more light.

It should also in theory be related to #818, but I checked and his code fails both his test and my test, and my code fails his test, so I'm not 100% sure what's going on with it.

    @classmethod
    def model_validate_json(
        cls: Type[_TSQLModel],
        json_data: str | bytes | bytearray,
        *,
        strict: bool | None = None,
        context: dict[str, Any] | None = None,
    ) -> _TSQLModel:
        obj = super().model_validate_json(
            json_data=json_data, strict=strict, context=context
        )
        # Without this, SQLModel.model_validate_json() doesn't recursively parse fields of type pydantic.ModelBase.
        # This has to do with (how we handle) the issue explained in a comment to __init__.
        # Since we already go to great lengths to fix this for model_validate, lets just use that (at some perf cost).
        return cls.model_validate(obj=obj, strict=strict, context=context)

Operating System

Windows, macOS

Operating System Details

No response

SQLModel Version

0.0.18, 5a421ad

Python Version

3.9.4, 3.11.3

Additional Context

No response