Unable to resolve local subschemas through "$defs"

[MatthewNielsen27]

The issue

The json-schema specification outlines the ability to reference local subschemas that are defined in $defs. Currently, a RefResolutionError is raised when attempting to validate an instance of a schema that references local subschemas with $defs.

How to replicate

Running the following python code will raise a RefResolutionError.

import jsonschema

# note: the $id uses a "namespace" for the project instead of a protocol, which is still valid.
schema = {
    "$schema": "https://json-schema.org/draft-07/schema",
    "$id": "project://test.schema.json",
    "type": "object",
    "properties": {
        "foo": {"$ref": "#/$defs/bar"}
    },
    "$defs": {
        "bar": {"type": "integer"}
    }
}

# In reality there is a dependency between multiple schemas in the same directory or namespace.
# I understand that we can pre-populate a RefResolver with a store of schemas.
store = {"project://test.schema.json": schema}
resolver = jsonschema.RefResolver.from_schema(schema, store=store)

instance = {"foo": 12}
jsonschema.Draft7Validator(schema=schema, resolver=resolver).validate(instance)

What I think is going wrong

Subschema references are resolved using the _find_in_subschemas(...) function which calls the following function:

    def _get_subschemas_cache(self):
        cache = {key: [] for key in _SUBSCHEMAS_KEYWORDS}
        for keyword, subschema in _search_schema(
            self.referrer, _match_subschema_keywords,
        ):
            cache[keyword].append(subschema)
        return cache

The issue is that _SUBSCHEMAS_KEYWORDS doesn't contain "$defs" (currently it is only comprised of ("$id", "id", "$anchor", "$dynamicAnchor"))

Proposed fix

Add "$defs" to _SUBSCHEMAS_KEYWORDS.

[Julian]

@MatthewNielsen27 thanks, will have a look.

[Julian]

I didn't fully look at this yet, but just glancing at it again briefly, I don't think this has anything to do with $def; does the issue go away if you use a different scheme? If so it's probably another instance of #544.

[MatthewNielsen27]

I'm suspicious because "$defs" doesn't appear anywhere in the python code. Since "$defs" isn't included in _SUBSCHEMA_KEYWORDS, it isn't considered in subschema resolution.

When the search for the id in the schema_store and subschemas fails, it then tries to pull it from the web. Which is the failure condition seen here and in #544. (in these cases since its a URN, the url is parsed as '').

To put it plainly, the subschema resolution only considers a subset of valid subschema tags.

[Julian]

$defs shouldn't appear in _SUBSCHEMA_KEYWORDS. That constant has to do with which keywords are anchor-like and affect the URI that's being resolved. $defs doesn't, it's just a normal addressible location.

(EDIT: Which isn't to say I know what the solution is yet, but yeah I'm reasonably sure that's not it).

[MatthewNielsen27]

Ok, thanks for clearing that up. I think we can mark this as a duplicate of #544.

[Julian]

Sounds good, will do, appreciate the report!