Python JSON Schema中allOf结合$ref验证失效问题

问题情况 #

我为Resource字典编写了JSON Schema，在$defs中定义PartialResource并通过allOf字段引用该定义。使用datamodel-codegen可正确生成对应的两个TypedDict类，但使用jsonschema的Draft202012Validator验证符合结构的数据时，却抛出ValidationError，提示PartialResource中的resource_kwargs、resource_path等属性为不允许的额外属性，仿佛未识别到PartialResource的定义。

相关代码与数据 #

JSON Schema代码 #

{
    "title": "resource",
    "type": "object",
    "allOf": [
        { "$ref": "#/$defs/PartialResource" }
    ],
    "$defs": {
        "PartialResource": {
            "title": "partial_resource",
            "properties": {
                "spec": {
                    "title": "spec",
                    "type": "string"
                },
                "resource_kwargs": {
                    "title": "resource_kwargs",
                    "type": "object"
                },
                "resource_path": {
                    "title": "resource_path",
                    "type": "string"
                },
                "root": {
                    "title": "root",
                    "type": "string"
                },
                "uid": {
                    "title": "uid",
                    "type": "string"
                }
            },
            "required": [
                "resource_kwargs",
                "resource_path",
                "root",
                "spec",
                "uid"
            ]
        }
    },
    "properties": {
        "path_semantics": {
            "title": "path_semantics",
            "type": "string",
            "enum": [
                "posix",
                "windows"
            ]
        },
        "run_start": {
            "title": "run_start",
            "type": "string"
        }
    },
    "additionalProperties": false
}

生成的TypedDict代码 #

from typing import TypedDict, Dict, Any, NotRequired, Literal

class PartialResource(TypedDict):
    spec: str
    resource_kwargs: Dict[str, Any]
    resource_path: str
    root: str
    uid: str


class Resource(PartialResource):
    path_semantics: NotRequired[Literal["posix", "windows"]]
    run_start: NotRequired[str]

待验证数据 #

{
    'path_semantics': 'posix',
    'uid': 'e0fc15fc-3dd3-4822-848c-76f84585202d',
    'spec': 'DUMMY',
    'root': '/placeholder/path',
    'resource_kwargs': {'a': 1, 'b': 2},
    'resource_path': 'stack.tiff',
    'run_start': '8172473c-fec5-4baf-b28a-73bc4ada3bc5'
}

错误信息 #

jsonschema.exceptions.ValidationError: Additional properties are not allowed ('resource_kwargs', 'resource_path', 'root', 'spec', 'uid' were unexpected)

问题原因与解决方案 #

原因 #

additionalProperties: false的作用范围仅局限于当前Schema的properties字段，不会考虑allOf引用的子Schema中定义的属性。因此验证器会把PartialResource里的属性判定为额外属性。

解决方案 #

使用unevaluatedProperties: false替代additionalProperties: false（JSON Schema Draft 2019-09及以上版本支持）。unevaluatedProperties会检查所有未被当前Schema或任何组合关键字（如allOf）引用的Schema评估过的属性，这样就能认可PartialResource中定义的属性。

修改后的Schema如下：

{
    "title": "resource",
    "type": "object",
    "allOf": [
        { "$ref": "#/$defs/PartialResource" }
    ],
    "$defs": {
        "PartialResource": {
            "title": "partial_resource",
            "properties": {
                "spec": { "title": "spec", "type": "string" },
                "resource_kwargs": { "title": "resource_kwargs", "type": "object" },
                "resource_path": { "title": "resource_path", "type": "string" },
                "root": { "title": "root", "type": "string" },
                "uid": { "title": "uid", "type": "string" }
            },
            "required": ["resource_kwargs", "resource_path", "root", "spec", "uid"]
        }
    },
    "properties": {
        "path_semantics": {
            "title": "path_semantics",
            "type": "string",
            "enum": ["posix", "windows"]
        },
        "run_start": { "title": "run_start", "type": "string" }
    },
    "unevaluatedProperties": false
}

另一种可选方案：将additionalProperties: false移到PartialResource定义中，主Schema不设置该字段，但这种方式无法限制主Schema中出现未定义的额外属性，需根据实际需求选择。

内容的提问来源于stack exchange，提问作者Eva Lott