Pydantic：验证嵌套模型

Question

我在 Pydantic 中有一个嵌套模型。我在外部模型中有一个root_validator 函数。我的印象是，如果调用外部根验证器，则内部模型是有效的。但显然不是。在下面的 MWE 中，我给内部模型提供了错误的字段名称，但外部验证器失败了：

from typing import List
from pydantic import BaseModel
from pydantic.class_validators import root_validator

class CarList(BaseModel):
    cars: List[str]
    colors: List[str]

class CarDealership(BaseModel):
    name: str
    cars: CarList

    @root_validator()
    def check_length(cls, v):
        cars_list = v.get("cars")

        if len(cars_list.cars) != len(cars_list.colors):
            raise ValueError("wrong length")
        
        return v


car_deal = {
    "cars": {
        "cars1": ["toyota", "tesla"], 
        "colors": ["white", "red"]
    }
}

CarDealership.parse_obj(car_deal)

我得到的错误是：

  File "test.py", line 17, in check_length
    if len(cars_list.cars) != len(cars_list.colors):
AttributeError: 'NoneType' object has no attribute 'cars'

我期待的是这样的：

pydantic.error_wrappers.ValidationError: 1 validation error for CarList
cars
  field required (type=value_error.missing)

如何确保首先验证内部模型？

Answer 1

更新：外部类版本的验证

from typing import List
from pydantic import BaseModel
from pydantic.class_validators import root_validator


class CarList(BaseModel):
    cars: List[str]
    colors: List[str]


class CarDealership(BaseModel):
    name: str
    cars_list: CarList

    @root_validator(pre=True)
    def check_length(cls, v):
        cars_list = v.get("cars_list")

        if not cars_list:
            return v
        cars = cars_list.get("cars")
        colors = cars_list.get("colors")
        if not isinstance(cars, list) or not isinstance(colors, list):
            return v
        if cars is not None and colors is not None and len(cars) != len(colors):
            raise ValueError("wrong length")

        return v

---

root_validator默认pre=False，内部模型已经验证，所以你得到v == {}。

您可以在CarList中创建check_length，并检查cars和colors是否存在（已经验证，如果失败则为None）。

from typing import List
from pydantic import BaseModel
from pydantic.class_validators import root_validator


class CarList(BaseModel):
    cars: List[str]
    colors: List[str]

    @root_validator
    def check_length(cls, v):
        cars = v.get("cars")
        colors = v.get("colors")

        if cars is not None and colors is not None and len(cars) != len(colors):
            raise ValueError("wrong length")

        return v


class CarDealership(BaseModel):
    name: str
    car_list: CarList


car_deal_1 = {
    "name": "value",
    "car_list": {
        "cars1": ["toyota", "tesla"],
        "colors": ["white", "red"]
    }
}

CarDealership.parse_obj(car_deal_1)
# pydantic.error_wrappers.ValidationError: 1 validation error for CarDealership
# car_list -> cars
#   field required (type=value_error.missing)

car_deal_2 = {
    "name": "value",
    "car_list": {
        "cars": ["toyota", "tesla"],
        "colors": ["white", "red", "bule"]
    }
}

CarDealership.parse_obj(car_deal_2)
# pydantic.error_wrappers.ValidationError: 1 validation error for CarDealership
# car_list -> __root__
#   wrong length (type=value_error)

Answer 2

基于@YeJun 响应，但假设您对响应的评论是您需要将内部类用于其他目的，您可以创建一个带有验证的中间类，同时保留原始 CarList 类用于其他用途：

from typing import List
from pydantic import BaseModel
from pydantic.class_validators import root_validator


class CarList(BaseModel):
    cars: List[str]
    colors: List[str]


# this class will have the same fields from CarList, but with validation
class CarListWithValidator(CarList):
    @root_validator
    def check_length(cls, v):
        cars = v.get("cars")
        colors = v.get("colors")

        if cars is not None and colors is not None and len(cars) != len(colors):
            raise ValueError("wrong length")

        return v


class CarDealership(BaseModel):
    name: str
    car_list: CarListWithValidator


car_deal_2 = {
    "name": "value",
    "car_list": {
        "cars": ["toyota", "tesla"],
        "colors": ["white", "red", "bule"]
    }
}

CarDealership.parse_obj(car_deal_2)
# pydantic.error_wrappers.ValidationError: 1 validation error for CarDealership
# car_list -> __root__
#   wrong length (type=value_error)