Returning instance from @model_validate(mode="before") raises model_type validation error

[nilsso]

Initial Checks

• I confirm that I'm using Pydantic V2

Description

Returning a model instance from a @model_validate(mode="before") decorated method raises ValidationError.

Example Code

from typing import Any
from pydantic import BaseModel, model_validator

class Foo(BaseModel):
    a: int
    b: int

    @model_validator(mode="before")
    @classmethod
    def _validate(cls, v: Any) -> Any:
        if isinstance(v, str):
            a, b = v.split("-")
            return Foo(a=int(a), b=int(b))
        return v

print(Foo.model_validate("1-2"))

Traceback (most recent call last):
  File "/Users/.../test.py", line 19, in <module>
    print(Foo.model_validate("1-2"))
          ^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/Users/.../.pyenv/versions/3.12.1/lib/python3.12/site-packages/pydantic/main.py", line 509, in model_validate
    return cls.__pydantic_validator__.validate_python(
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
pydantic_core._pydantic_core.ValidationError: 1 validation error for Foo
  Input should be a valid dictionary or instance of Foo [type=model_type, input_value=Foo(a=1, b=2), input_type=Foo]
    For further information visit https://errors.pydantic.dev/2.6/v/model_type

As a work around, taking from #7292 (comment)

class Foo(BaseModel):
    a: int
    b: int

    @model_validator(mode="before")
    @classmethod
    def _validate(cls, v: Any) -> Any:
        if isinstance(v, str):
            a, b = v.split("-")
            return Foo(a=int(a), b=int(b)).model_dump()
            # or return {"a": int(a), "b": int(b)}
        return v

Python, Pydantic & OS Version

pydantic version: 2.6.4
        pydantic-core version: 2.16.3
          pydantic-core build: profile=release pgo=true
                 install path: /Users/.../.pyenv/versions/3.12.1/lib/python3.12/site-packages/pydantic
               python version: 3.12.1 (main, Jan  3 2024, 13:12:52) [Clang 15.0.0 (clang-1500.1.0.2.5)]
                     platform: macOS-14.0-arm64-arm-64bit
             related packages: pyright-1.1.350 typing_extensions-4.10.0
                       commit: unknown

[sydney-runkle]

@nilsso,

Thanks very much for reporting this. Definitely a bug we want to fix for v2.8.0.

Here's an odd reproduction:

from typing import Any
from pydantic import BaseModel, ConfigDict, model_validator


class Foo(BaseModel):
    a: int
    b: int

    @model_validator(mode="before")
    @classmethod
    def _validate(cls, v: Any) -> Any:
        if isinstance(v, str):
            a, b = v.split("-")
            return Foo(a=int(a), b=int(b))
        return v

print(repr(Foo.model_validate({"a": 1, "b": 2})))
#> Foo(a=1, b=2)
print(repr(Foo.model_validate(Foo(a=1, b=2))))
#> Foo(a=1, b=2)
print(repr(Foo.model_validate("1-2", from_attributes=True)))
#> Foo(a=1, b=2)
print(repr(Foo.model_validate("1-2", from_attributes=False)))
"""
pydantic_core._pydantic_core.ValidationError: 1 validation error for Foo
  Input should be a valid dictionary or instance of Foo [type=model_type, input_value=Foo(a=1, b=2), input_type=Foo]
    For further information visit https://errors.pydantic.dev/2.7/v/model_type
"""

[davidhewitt]

Moving over from #7292 -

Looking at the "bug" here, the problem is that the "before" validator is running as a wrapper around the model-fields schema, not around the outer model schema. So the "before" validator is required to return a set of fields (i.e. a dictionary), not a complete model.

Potentially the schema needs to be changed so that the before validator runs before the whole "model" schema. I'm not sure what other stuff breaks if we do that. Needs further investigation.