Confusing handling of badly defined model_validator

[TimChild]

Initial Checks

• I confirm that I'm using Pydantic V2

Description

If there is a bad model_validator that forgets to return self then SomeModel.model_validate(dict_obj) returns None while SomeModel(**dict_obj) continues to return the validated model.

This makes it seem like there is an issue with the .model_validate method, but obviously it's because of the bad validator...

This might have been the cause of #7055 (that's what I found when I first had the issue).

So, not quite a bug given that the code all works correctly if the user doesn't do something stupid... But, it would have been nice to be told I was stupid via a meaningful error or warning :)

Example Code

from pydantic import BaseModel, model_validator, ValidationError


class SomeModel(BaseModel):
    a: int

    @model_validator(mode="after")
    def validate_a(self):
        if self.a == 10:
            raise ValueError("Bad")
        # Should return `self` here... but "accidentally" ommitting it to show confusing behaviour that results from this.

if __name__ == "__main__":
    # Confusing behaviour is that model_validate returns None while direct instantiation does not
    gets_none = SomeModel.model_validate({"a": 5})
    assert gets_none is None, "This is somewhat surprising (although obviously the result of a bad validator)"

    gets_value = SomeModel(**{"a": 5})
    assert isinstance(gets_value, SomeModel), "This confuses the matter by still returning a value..."

    # Both correctly raise validation errors still
    try:
        SomeModel.model_validate({"a": 10})
        print("Does not get to here")
    except ValidationError as e:
        print("Correctly raises validation")

    try:
        SomeModel(**{"a": 10})
        print("Does not get to here")
    except ValidationError as e:
        print("Correctly raises validation")

Python, Pydantic & OS Version

pydantic version: 2.7.1
        pydantic-core version: 2.18.2
          pydantic-core build: profile=release pgo=true
                 install path: /home/tim/.cache/pypoetry/virtualenvs/backend-dvWTEZau-py3.12/lib/python3.12/site-packages/pydantic
               python version: 3.12.1 | packaged by Anaconda, Inc. | (main, Jan 19 2024, 15:51:05) [GCC 11.2.0]
                     platform: Linux-5.15.146.1-microsoft-standard-WSL2-x86_64-with-glibc2.35
             related packages: typing_extensions-4.11.0 pydantic-settings-2.2.1 fastapi-0.109.2
                       commit: unknown

[sydney-runkle]

Agreed, a more helpful warning message would be helpful here. PR welcome with added support for this :).

[Viicos]

I'll probably work on this!

[Viicos]

Taking a look at it, we could error here on the pydantic-core side:

impl FunctionAfterValidator {
    fn _validate<'py, I: Input<'py> + ?Sized>(
        &self,
        call: impl FnOnce(&I, &mut ValidationState<'_, 'py>) -> ValResult<PyObject>,
        py: Python<'py>,
        input: &I,
        state: &mut ValidationState<'_, 'py>,
    ) -> ValResult<PyObject> {
        let v = call(input, state)?;
        let r = if self.info_arg {
            let info = ValidationInfo::new(py, state.extra(), &self.config, self.field_name.clone());
            self.func.call1(py, (v.to_object(py), info))
        } else {
            self.func.call1(py, (v.to_object(py),))
        };
        r.map_err(|e| convert_err(py, e, input))
        // Check if the value is `None`
    }
}

However, I'm unsure if the FunctionAfterValidator is also used in other places, e.g. with field validators where None would be a valid value returned by the validator?

As a side note, this can be prevented by using a type checker (and I'll implement a rule in flake8-pydantic soon), plus you will notice during development (as opposed to during runtime if adding an explicit check as suggested)