如何排除Pydantic允许空字典字段的冗余验证错误？

我定义了如下Pydantic模型，通过member字段作为鉴别器区分ModelX和ModelY，同时允许model字段为空字典匹配Empty模型：

from typing import Union

from pydantic import BaseModel, Extra, Field
from typing_extensions import Annotated, Literal


class Empty(BaseModel):
    class Config:
        extra = Extra.forbid


class ModelX(BaseModel):
    member: Literal["x"]
    q: str


class ModelY(BaseModel):
    member: Literal["y"]
    q: None


class MainModel(BaseModel):
    model: Union[Annotated[Union[ModelX, ModelY], Field(discriminator="member")], Empty]


MainModel.parse_obj({"model": {}})
MainModel.parse_obj({"model": {"member": "x", "q": None}})

当传入{"model": {"member": "x", "q": None}}时，因为ModelX要求q为字符串不能为None，触发了校验错误，但返回的错误信息里包含两条多余的extra fields not permitted错误：

pydantic.error_wrappers.ValidationError: 3 validation errors for MainModel
model -> ModelX -> q
  none is not an allowed value (type=type_error.none.not_allowed)
model -> member
  extra fields not permitted (type=value_error.extra)
model -> q
  extra fields not permitted (type=value_error.extra)

这两条额外错误并非预期，我只需要保留q字段的类型错误。

问题根源是Pydantic处理联合类型时，会尝试匹配所有分支模型。当前定义下，传入带字段的model时，既触发了ModelX的校验错误，又触发了Empty的额外字段错误。

方案1：调整联合类型匹配顺序（推荐） #

将带鉴别器的ModelX/ModelY分支放在联合类型的前面，让Pydantic优先处理鉴别器匹配，只有当匹配失败（比如没有member字段）时才尝试匹配Empty模型：

from typing import Union

from pydantic import BaseModel, Extra, Field
from typing_extensions import Annotated, Literal


class Empty(BaseModel):
    class Config:
        extra = Extra.forbid


class ModelX(BaseModel):
    member: Literal["x"]
    q: str


class ModelY(BaseModel):
    member: Literal["y"]
    q: None


class MainModel(BaseModel):
    # 优先匹配带鉴别器的ModelX/ModelY，再匹配Empty
    model: Union[Annotated[Union[ModelX, ModelY], Field(discriminator="member")], Empty]


# 测试验证
try:
    MainModel.parse_obj({"model": {"member": "x", "q": None}})
except Exception as e:
    print(e)

运行后错误信息会简化为：

1 validation error for MainModel
model -> ModelX -> q
  none is not an allowed value (type=type_error.none.not_allowed)

方案2：让Empty参与鉴别器匹配 #

给Empty模型添加一个可选的member字段，让它加入鉴别器的匹配逻辑，这样Pydantic只会根据member值匹配对应的模型，不会触发Empty的额外字段错误：

from typing import Union

from pydantic import BaseModel, Extra, Field
from typing_extensions import Annotated, Literal


class Empty(BaseModel):
    member: Literal[None] = None
    class Config:
        extra = Extra.forbid


class ModelX(BaseModel):
    member: Literal["x"]
    q: str


class ModelY(BaseModel):
    member: Literal["y"]
    q: None


class MainModel(BaseModel):
    model: Annotated[Union[ModelX, ModelY, Empty], Field(discriminator="member")]


# 测试空字典匹配
MainModel.parse_obj({"model": {}})  # 正常匹配Empty
# 测试错误输入
try:
    MainModel.parse_obj({"model": {"member": "x", "q": None}})
except Exception as e:
    print(e)

这种方式下，空字典会因为member默认是None匹配Empty模型，而带member: "x"的输入只会匹配ModelX，不会触发Empty的校验错误。

内容的提问来源于stack exchange，提问作者xi¹⁸⁶