SQLModel does not execute type checks in pydantic when table=True set

[monchin]

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

from pydantic import BaseModel
from sqlmodel import SQLModel, Field

class TestPD(BaseModel):
    id: int
    name: str
    age: int

class TestSM(SQLModel, table=True):
    id: int | None = Field(default=None, primary_key=True)
    name: str
    age: int

def test_pd():
    test = TestPD(id=1, name="test", age="test")
    print(test)

def test_sm():
    test = TestSM(id=1, name="test", age="test")
    print(test)

# test_pd() # raise ValidationError here
test_sm() # doesn't raise

Description

Hi, I just found that althogh SQLModel derives from pydantic.BaseModel, it doesn't really execute type checks in pydantic. Please read the code above. If we uncomment test_pd() and comment test_sm(), it would raise ValidationError saying age is not a valid integer; but if we execute test_sm(), everything goes well, output would be

id=1 name='test' age='test'

although it should raise a ValidationError. And I found if I delete table=True, it behaves just like BaseModel, but in my condition table=True must be set.

I'd like to know why pydantic type checking doesn't work when I set table=True? I have found some simillar discussions for example this but it remained unanswered. Is there a way to solve it?

Operating System

Windows

Operating System Details

No response

SQLModel Version

0.0.21

Python Version

3.10.12

Additional Context

No response

[monchin]

I read the source code, and found if table=True is set, original pydantic validators will not executed, instead sqlmodel_table_construct will be excuted. That is the reason why pydantic doesn't work.

One reason I'm attrachted by sqlmodel is that it combines pydantic and orm, so I think it would be great if I can benefit all pydantic traits, not only for type checking, but also for example field_validator to change values automatically. Thus I tried to modify some code like this

# source code
    def sqlmodel_init(*, self: "SQLModel", data: Dict[str, Any]) -> None:
        old_dict = self.__dict__.copy()
        if not is_table_model_class(self.__class__):
            self.__pydantic_validator__.validate_python(
                data,
                self_instance=self,
            )
        else:
            sqlmodel_table_construct(
                self_instance=self,
                values=data,
            )
# modified code
    def sqlmodel_init(*, self: "SQLModel", data: Dict[str, Any]) -> None:
        old_dict = self.__dict__.copy()
        raw_self = self.copy()
        pydantic_validated_model = self.__pydantic_validator__.validate_python(
            data,
            self_instance=self,
        )
        if is_table_model_class(self.__class__):
            sqlmodel_table_construct(
                self_instance=raw_self,
                values=pydantic_validated_model.model_dump(),
            )

In this way pydantic also works even though table=True is set, including type checking and all validators to validate or change values.
If you think it is a good idea and the code is OK, I'm glad to contribute!

[monchin]

Now I have fixed my code to pass tests

    def sqlmodel_init(*, self: "SQLModel", data: Dict[str, Any]) -> None:
        # old_dict = self.__dict__.copy()
        # if not is_table_model_class(self.__class__):
        #     self.__pydantic_validator__.validate_python(
        #         data,
        #         self_instance=self,
        #     )
        # else:
        #     sqlmodel_table_construct(
        #         self_instance=self,
        #         values=data,
        #     )
        old_dict = self.__dict__.copy()
        if not is_table_model_class(self.__class__):
            self.__pydantic_validator__.validate_python(
                data,
                self_instance=self,
            )
        else:
            raw_self = deepcopy(self)
            pydantic_validated_model = self.__pydantic_validator__.validate_python(
                data,
                self_instance=raw_self,
            )
            pydantic_dict = pydantic_validated_model.model_dump()
            for k in pydantic_dict.keys():
                if k not in data.keys():
                    continue
                data[k] = pydantic_dict[k]
            sqlmodel_table_construct(
                self_instance=self,
                values=data,
            )
        object.__setattr__(
            self,
            "__dict__",
            {**old_dict, **self.__dict__},
        )

Not all tests passed, but tests do not pass also fails in original code, so I think my code works well. But for one test tests/test_instance_no_args/test_allow_instantiation_without_arguments, the original code passed but mine failed. In this test, I notice that original code deliberately initiate an instance of SQLModel without neccesary non-nullable parameters, and set the neccesary parameters after initiation. So I'd like to know if table=True disabling pydantic validators is a feature or a bug. I hope I can regard it as a bug because I hope to make full use of pydantic in sqlmodel, of course I can accept it if it is a feature.