Pydantic与含oneOf的JSON Schema兼容问题：缺失field_1字段

我尝试基于包含oneOf关键字的JSON Schema生成Pydantic模型，Schema内容如下：

{
    "$schema": "https://json-schema.org/draft/2019-09/schema",
    "type": "object",
    "title": "v2_test",
    "additionalProperties": true,
    "oneOf": [
        {
            "type": "object",
            "properties": {
                "field_1": {
                    "enum": [
                        "response_1"
                    ]
                }
            },
            "additionalProperties": true,
            "oneOf": [
                {
                    "type": "object",
                    "properties": {
                        "field_2": {
                            "enum": [
                                "response_a"
                            ]
                        }
                    },
                    "additionalProperties": true,
                    "required": [
                        "field_2"
                    ]
                }
            ],
            "required": [
                "field_1"
            ]
        },
        {
            "type": "object",
            "properties": {
                "field_1": {
                    "enum": [
                        "response_2"
                    ]
                }
            },
            "additionalProperties": true,
            "oneOf": [
                {
                    "type": "object",
                    "properties": {
                        "field_2": {
                            "enum": [
                                "response_b"
                            ]
                        }
                    },
                    "additionalProperties": true,
                    "required": [
                        "field_2"
                    ]
                },
                {
                    "type": "object",
                    "properties": {
                        "field_2": {
                            "enum": [
                                "response_c"
                            ]
                        }
                    },
                    "additionalProperties": true,
                    "required": [
                        "field_2"
                    ]
                }
            ],
            "required": [
                "field_1"
            ]
        }
    ]
}

使用命令datamodel-codegen --input schema.json --input-file-type jsonschema --output model.py生成后，模型代码丢失了Schema要求的field_1字段（枚举值为response_1或response_2），生成的代码如下：

# generated by datamodel-codegen:
#   filename:  v2_differnet_names_schema.json
#   timestamp: 2022-08-21T09:12:26+00:00
from __future__ import annotations
from enum import Enum
from typing import Union
from pydantic import BaseModel, Extra, Field

class Field2(Enum):
    response_a = 'response_a'

class V2TestItem(BaseModel):
    class Config:
        extra = Extra.allow
    field_2: Field2

class Field21(Enum):
    response_b = 'response_b'

class V2TestItem1(BaseModel):
    class Config:
        extra = Extra.allow
    field_2: Field21

class Field22(Enum):
    response_c = 'response_c'

class V2TestItem2(BaseModel):
    class Config:
        extra = Extra.allow
    field_2: Field22

class V2Test(BaseModel):
    class Config:
        extra = Extra.allow
    __root__: Union[V2TestItem, Union[V2TestItem1, V2TestItem2]] = Field(
        ..., title='v2_test'
    )

解决方案 #

方法1：手动修正生成的模型代码 #

直接在生成的各个Item模型中添加field_1字段及对应枚举，确保field_1的枚举值与field_2匹配：

from __future__ import annotations
from enum import Enum
from typing import Union
from pydantic import BaseModel, Extra, Field

class Field1(Enum):
    response_1 = 'response_1'
    response_2 = 'response_2'

class Field2(Enum):
    response_a = 'response_a'

class V2TestItem(BaseModel):
    class Config:
        extra = Extra.allow
    field_1: Field1 = Field(Field1.response_1)
    field_2: Field2

class Field21(Enum):
    response_b = 'response_b'

class V2TestItem1(BaseModel):
    class Config:
        extra = Extra.allow
    field_1: Field1 = Field(Field1.response_2)
    field_2: Field21

class Field22(Enum):
    response_c = 'response_c'

class V2TestItem2(BaseModel):
    class Config:
        extra = Extra.allow
    field_1: Field1 = Field(Field1.response_2)
    field_2: Field22

class V2Test(BaseModel):
    class Config:
        extra = Extra.allow
    __root__: Union[V2TestItem, Union[V2TestItem1, V2TestItem2]] = Field(
        ..., title='v2_test'
    )

方法2：调整JSON Schema结构，优化兼容性 #

datamodel-codegen对嵌套oneOf的解析存在局限性，可将Schema中的嵌套oneOf合并为顶层的oneOf，明确每个分支的完整字段要求：

{
    "$schema": "https://json-schema.org/draft/2019-09/schema",
    "type": "object",
    "title": "v2_test",
    "additionalProperties": true,
    "oneOf": [
        {
            "type": "object",
            "properties": {
                "field_1": {
                    "enum": ["response_1"]
                },
                "field_2": {
                    "enum": ["response_a"]
                }
            },
            "additionalProperties": true,
            "required": ["field_1", "field_2"]
        },
        {
            "type": "object",
            "properties": {
                "field_1": {
                    "enum": ["response_2"]
                },
                "field_2": {
                    "enum": ["response_b"]
                }
            },
            "additionalProperties": true,
            "required": ["field_1", "field_2"]
        },
        {
            "type": "object",
            "properties": {
                "field_1": {
                    "enum": ["response_2"]
                },
                "field_2": {
                    "enum": ["response_c"]
                }
            },
            "additionalProperties": true,
            "required": ["field_1", "field_2"]
        }
    ]
}

使用同样的datamodel-codegen命令重新生成，此时生成的模型会包含完整的field_1和field_2字段。

方法3：升级工具或启用特定参数 #

如果使用的是旧版本的datamodel-codegen，先升级到最新版本：

pip install --upgrade datamodel-codegen

或者添加--use-subclass参数，帮助工具更好地处理嵌套的oneOf结构：

datamodel-codegen --input schema.json --input-file-type jsonschema --output model.py --use-subclass

内容的提问来源于stack exchange，提问作者Cole