如何从 json 中解析带有“Type”类型字段的 pydantic 模型？

Question

How to make the following work with pydantic?

from typing import Type

import pydantic


class InputField(pydantic.BaseModel):
    name: str
    type: Type

InputField.parse_raw('{"name": "myfancyfield", "type": "str"}')

It fails with

pydantic.error_wrappers.ValidationError: 1 validation error for InputField
type
  a class is expected (type=type_error.class)

But I need to parse this from json, so I don't have the option to directly pass the Type object to the __init__ method.

Answer 1

A custom validator with pre=True will allow you to attempt to find a class with the provided name. Here is a working example first trying to grab a built-in and failing that assuming the class is in global namespace:

from pydantic import BaseModel, validator


class InputField(BaseModel):
    name: str
    type_: type

    @validator("type_", pre=True)
    def parse_cls(cls, value: object) -> type:
        name = str(value)
        try:
            obj = getattr(__builtins__, name)
        except AttributeError:
            try:
                obj = globals()[name]
            except KeyError:
                raise ValueError(f"{value} is not a valid name")
        if not isinstance(obj, type):
            raise TypeError(f"{value} is not a class")
        return obj


class Foo:
    pass


if __name__ == "__main__":
    print(InputField.parse_raw('{"name": "a", "type_": "str"}'))
    print(InputField.parse_raw('{"name": "b", "type_": "Foo"}'))

Output:

name='a' type_=<class 'str'>
name='b' type_=<class '__main__.Foo'>

If you want to support dynamic imports as well, that is possible too. See here or here for pointers.