(Let's just remove the refurls. I'm not sure what they're useful for.)

yeah, seen this too. The documentation for $id doesn't absolutely state that the URI has to be valid, but I think what we're running into here is this bit:

"But be aware of the second purpose of the $id property: that it declares a base URL for relative $ref URLs elsewhere in the file. For example, if you had:

{ "$ref": "person.json" }

in the same file, a JSON schema validation library that supported network fetching would fetch person.json from http://foo.bar/schemas/person.json, even if address.json was loaded from somewhere else, such as the local filesystem."

okay, so the doc is slightly confusing about what is allowed here because I think it is using incorrect terminology. Specifically, it keeps saying that $ids and $refs should be "URIs", but it gives examples which are not strictly speaking URIs, like here:

"$ref can also be a relative or absolute URI, so if you prefer to include your definitions in separate files, you can also do that. For example:

{ "$ref": "definitions.json#/address" }

would load the address schema from another file residing alongside this one."

But...definitions.json#/address is not strictly a URI, because a URI requires a scheme. I believe the correct terminology here would be that this is a "relative URI reference", per this bit of the spec.

It seems pretty definite from the examples and context in the doc that it is valid for $refs to be relative references. It's also clear that $ids for lower-level schemas can be relative references. It's not entirely clear whether it's considered valid for top-level schema $ids to be relative references, such that there is no absolute URI for anything anywhere in the whole set of schemas. However, I tried it, and it seems like doing that does work with Python jsonschema at least, so I guess that's what we should do at least for now...

Ah, OK, so if I look at the actual JSON Schema IETF draft, this is a little clearer. Section 8.2 covers $id and states "If present, the value for this keyword MUST be a string, and MUST represent a valid URI-reference [RFC3986]." (note "URI-reference" not "URI"). Section 8.3 covers $ref and states "The value of the "$ref" property MUST be a URI Reference." So I think we're good. I sent a PR for the json-schema.org docs.

Sections 8.3.1-8.3.2 are also interesting (though hard to parse) and may suggest that jsonschema is doing the wrong thing here: I think it's trying to say that jsonschema should realize the schema is available locally and work, even though the URI doesn't resolve. I need more coffee and the author of the draft to yell at in order to be sure, though...

Should be closed with #75.