Greedy (?) object typing leads to semantically wrong error message

jsonschema.validate を使って issue のスキーマとデータを再現し、その後 validators.py の検証パスを調べます。anyOf の各分岐がどのように ValidationError を生成するかを比較します。done は、無効なエンティティデータが entity_property なしで BREAK_ENTITY を報告することを意味し、実証されたオブジェクト順序のケースをカバーする必要があります。

Hi, thanks for a great project!

We've stumbled across an issue with misleading error messages, although I'm not sure if there is a useful fix.

• jsonschema version: 3.2.0
• Python: 3.8.5

When you define two objects (e.g., person and entity), which share a number of properties but define others which are mutually exclusive across the two, and one object (say, entity) also requires fields where the other doesn't, properties are reported in a ValidationError as unexpected when from the user perspective, they aren't.

Expected behaviour

Given the definitions

• person (properties: person_property (optional), alias (optional))
• entity (properties: entity_property (required), alias (optional))

and data that introduces an additional property BREAK_ENTITY (which is disallowed via "additionalProperties": false) to an otherwise valid entity data object, I expect the error message to raise the additional key, but not any other (semantically valid) keys.

Actual behaviour

The error messages raises two unexpected keys: the additional one (correct), and entity_property (incorrect).

Without having had a look at the source code, I assume that the additional property trips the error, and then trying to type the object and collecting unexpected properties is greedy, and in this case assumes person whereas we're looking at entity really, which would be valid if only the additional property was removed. I guess changing the heuristic to find the best match may come at a performance penalty though, as you'd have to look at all potential best matches?

This also happens independently from which of the two objects is defined first, but depends on the order of objects in anyOf.

Reproducing the issue

(Minimalized (well, more or less) schema

{
    "$schema": "http://json-schema.org/draft-07/schema",
    "additionalProperties": false,
    "type": "object",
    "properties": {
        "parties": {
            "items": {
                "anyOf": [
                    {
                        "$ref": "#/definitions/person"
                    },
                    {
                        "$ref": "#/definitions/entity"
                    }
                ]
            },
            "minItems": 1,
            "type": "array",
            "uniqueItems": true
        }
    },
    "required": [
        "parties"
    ],
    "definitions": {
        "person": {
            "type": "object",
            "additionalProperties": false,
            "properties": {
                "person_property": {
                    "type": "string",
                    "minLength": 1
                },
                "alias": {
                    "$ref": "#/definitions/alias"
                }
            }
        },
        "entity": {
            "additionalProperties": false,
            "properties": {
                "entity_property": {
                    "type": "string",
                    "minLength": 1
                },
                "alias": {
                    "$ref": "#/definitions/alias"
                }
            },
            "required": [
                "entity_property"
            ],
            "type": "object"
        },
        "alias": {
            "type": "string",
            "minLength": 1
        }
    }
}

Valid data

{
    "parties": [
        {
            "entity_property": "Entity",
            "alias": "Entity alias"
        },
        {
            "person_property": "Person",
            "alias": "Person alias"
        }
    ]
}

Invalid person data (yields expected message)

The following invalid person data breaks and yields the expected error message: jsonschema.exceptions.ValidationError: Additional properties are not allowed ('BREAK_PERSON' was unexpected).

{
    "parties": [
        {
            "entity_property": "Entity",
            "alias": "Entity alias"
        },
        {
            "person_property": "Person",
            "alias": "Person alias",
            "BREAK_PERSON": "BREAK"
        }
    ]
}

ISSUE: Invalid entity data (yields unexpected message)

The following invalid entity data yields the expected error, albeit with an unexpected error message jsonschema.exceptions.ValidationError: Additional properties are not allowed ('BREAK_ENTITY', 'entity_property' were unexpected) (, 'entity_property' being the unexpected part, full trace further below)

{
    "parties": [
        {
            "entity_property": "Entity",
            "alias": "Entity alias",
            "BREAK_ENTITY": "BREAK"
        },
        {
            "person_property": "Person",
            "alias": "Person alias"
        }
    ]
}

Trace

Linebreaks included for legibility.

Traceback (most recent call last):
  File "repr_bug.py", line 27, in <module>
    validate(args.data, args.schema)
  File "repr_bug.py", line 14, in validate
    jsonschema.validate(instance=data, schema=schema_data)
  File "~/venv/lib/python3.8/site-packages/jsonschema/validators.py", line 934, in validate
    raise error
jsonschema.exceptions.ValidationError: 
  Additional properties are not allowed ('BREAK_ENTITY', 'entity_property' were unexpected)

Failed validating 'additionalProperties' in schema[0]:
    {'additionalProperties': False,
     'properties': {'alias': {'$ref': '#/definitions/alias'},
                    'person_property': {'minLength': 1, 'type': 'string'}},
     'type': 'object'}

On instance:
    {'BREAK_ENTITY': 'BREAK',
     'alias': 'Entity alias',
     'entity_property': 'Entity'}