Passing a type as a variable for dynamically created fastapi endpoints

[Sanchoyzer]

Gist URL: https://gist.github.com/mypy-play/a0067f2bc095dbdb03fc9e1933cc5ad8
Playground URL: https://mypy-play.net/?mypy=latest&python=3.12&gist=a0067f2bc095dbdb03fc9e1933cc5ad8

I'm trying to dynamically create a fastapi endpoint. For endpoints with a body I need to pass a schema. In this solution the code works fine, but I don't know how to fix a mypy problem.

class CreateBaseSchema(BaseModel):
    pass

class CreateUserSchema(CreateBaseSchema):
    login: str
    password: str

def create_entity_endpoint(
    business_logic_func: Callable,
    response_model: type[CreateBaseSchema],
) -> Callable:
    def endpoint(data: Annotated[response_model, Body()]) -> dict:
        return business_logic_func(data=data)
    return endpoint

create_users_endpoint = create_entity_endpoint(
    business_logic_func=lambda data: {'id': 1, 'login': data.login},
    response_model=CreateUserSchema,
)

my_router.add_api_route(
    path='/users/',
    endpoint=create_users_endpoint,
    methods=['POST'],
    response_model=UserResponse,
)

In function "create_entity_endpoint":
error: Variable "response_model" is not valid as a type  [valid-type]
note: See https://mypy.readthedocs.io/en/stable/common_issues.html#variables-vs-type-aliases

I've read that variables-vs-type-aliases link, but I'm not sure that I can use TypeAlias in my case, because I need to pass a type as a variable

[Daverball]

You can just ignore mypy's error here using a # type: ignore[valid-type] comment on that line, since the annotation isn't really used for static type checking in this context, you're just trying to pass dynamic type information to FastAPI via the annotation. All you really lose by ignoring the type error is the ability for mypy to be able to infer the type of data inside endpoint, so it will just fall back to Any.

You could also do this in a more dynamic way by directly manipulating __annotations__. Rather than defining a wrapper function with the correct annotations (although unless you're dealing with an instance of FunctionType, you may need to create a wrapper anyways).

The way it's written currently is not very type safe however, since you're not validating that the business logic function can handle the model you're passing it, so data in the lambda will be inferred as Any, so you will not get code completion in the lambda or know when you've made a mistake, like accessing a property that doesn't exist.

[Sanchoyzer]

First of all, thanks for your answer

The way it's written currently is not very type safe however, since you're not validating that the business logic function can handle the model you're passing it, so data in the lambda will be inferred as Any, so you will not get code completion in the lambda or know when you've made a mistake, like accessing a property that doesn't exist.

A lambda function was just for an example, I try to use a "common" function like:

def create_user_handler(body: CreateUserSchema) -> dict:
    return {'id': 1, 'login': data.login}

create_users_endpoint = create_entity_endpoint(
    business_logic_func=create_user_handler,
    response_model=CreateUserSchema,
)

[Daverball]

You'd still get no verification that the function is compatible with the model this way though. You can pass in an arbitrary callable and the type checker will accept it.