Sub class validator behaviour loses type.

There’s some non-obvious behaviour that loses types in pydantic in the following case.

    import pydantic
    
    class IntSub(int):
        pass

    class IntConf(pydantic.BaseModel):
        count: IntSub
    
    c = IntSub(42)
    x = IntConf(count=c)
    assert type(x.count) is IntSub

This is because find_validators() checks whether count is a subclass of int and if it is returns a (int, number_size_validator) pair of validators. So upon creation x.count is set to int©.

This could be fixed by replacing the int() validator with the following:

def new_int_validator(x):
    try:
        int(x)
    except ValueError:
        raise SomeDescriptiveError("blah blah")
    return x

This is more strongly typed and since we appear to be caring about typing if we’re using pydantic, better … I believe. But this is a breaking change since things like IntSub(3.14) and IntSub(“2”) will stop working.

Same goes for float subtypes … I assume.

An alternative is to require a cast constructor for subclasses and use that to check the type instead, e.g. consider what pydantic does without the explicit c=IntSub(42) step above. So perhaps a better check is that the input is already of the expected type… if not try a cast constructor, if that fails we tried.

[samuelcolvin]

Humm, I see the problem, but I definitely want 3.14 and "2" to work with int fields.

It looks like it’s not properly documented by the solution is to use get_validators like constr does.

You’ll want something like

class IntSub(int):
    @classmethod
    def get_validators(cls):
        yield cls.myvalidator

    @classmethod
    def myvalidator(cls, v):
        return cls(v)

Let me know if that works.

[samuelcolvin]

fix via #157