I have a model defined like this;

class SomeModel(BaseModel):
    name: str

class SomeOtherModel(BaseModel):
    name: int

class MyModel(BaseModel):
    items: List[Union[SomeModel, SomeOtherModel]]

    @validator("items", always=True)
    def validate(cls, value):
        by_type = list(filter(lambda v: isinstance(v, SomeModel), value))
        if len(by_type) < 1:
            raise ValueError("we need at least one SomeModel")

The submodels are unimportant, essentially I need a list of different sub-models, and the list must contain at least one of the first type. All well and good.

Elsewhere in my code I am referring to this model (context: for the purposes of saving user settings, should be irrelevant for this question).

class ComposerModels(BaseModel):
    user: List[MyModel] = []
    system: List[MyModel] = []

class ComposerSettings(BaseModel):
    models: ComposerModels

class UserSettings(BaseModel):
    composer: ComposerSettings

My program needs to be able to save new models into the UserSettings model, something like this;

my_model = MyModel(items=[SomeModel(name="a"), SomeOtherModel(name=1)])

user_settings = UserSettings(
    composer=ComposerSettings(
        models=ComposerModels(
            user=[my_model]
        )
    )
)

However, this throws an error;

Traceback (most recent call last):
  File ".../pydantic/shenanigans.py", line 38, in <module>
    user=[my_model]
  File "pydantic/main.py", line 342, in pydantic.main.BaseModel.__init__
pydantic.error_wrappers.ValidationError: 1 validation error for ComposerModels
user -> 0
  we need at least one SomeModel (type=value_error)

This is where it gets weird.

For some reason, when trying to instantiate UserSettings, it goes through validation of the items field inside MyModel, but instead of having been passed a list of sub-models, as expected, somehow it’s getting a list of MyModel instances instead. This obviously fails validation, and raises the error above. However, if I comment out the validation code, it works. No error, and the user settings model contains the MyModel we’d expect.

I can’t just disable that validation, I need it elsewhere in my program… Any ideas on what’s going on here? I’m stumped…

I’m running python 3.7 with pydantic 1.10 on centos 7. I can’t easily upgrade any versions, because my company doesn’t believe in devops, so if this is a known bug with pydantic at that version I’ll have to think of something else.