spec: be clear that rel is a URI (hyper-schema)

[handrews]

This line:

https://github.com/json-schema-org/json-schema-spec/blob/master/jsonschema-hyperschema.xml#L567

describes the rel LDO keyword as:

The value of the "rel" property indicates the name of the relation to the target resource.

While RFC 5988 explains how "rel" works as a concept, the casual phrase "name of the relation" encourages people who might not read RFCs for fun to think this is a free-form namespace. This can be seen as I muddle through this in issue #47 (Note: I do read RFCs for fun, had read RFC 5988, and knew that unregistered link rels should be URIs, and I still confused myself. This may say more about me than about the spec's wording, but it's a data point).

We shouldn't duplicate RFC 5988, but a brief mention of the URI-ness and an example of a proper extension URI would help. Instead, we go on to use an unregistered rel of "children" in a later example, with no indication that it is or should be considered differently from the other rel in the example, "up" (which is registered):

https://github.com/json-schema-org/json-schema-spec/blob/master/jsonschema-hyperschema.xml#L590

[awwright]

Made edits in master, you can close this out if that works for you

[handrews]

Is there a reason for saying

or a normalized URI following the URI production of RFC 3986

instead of just following the bit about RFC 5988 registered relations with a mention of RFC 5988 extension relations? I know that what you are saying with the RFC3986 reference is the same thing (I think), but offloading the whole concept to RFC 5988 feels cleaner.

I am also curious about the change from "SHOULD NOT be media type dependent" to the much weaker "are not normally media type dependent"; are there signifiant accepted relationships that are media type dependent for anything other than grandfathered reasons?

[awwright]

instead of just following the bit about RFC 5988 registered relations with a mention of RFC 5988 extension relations?

I suppose we could just import the production

I am also curious about the change from "SHOULD NOT be media type dependent" to the much weaker "are not normally media type dependent"

Because that's a pre-existing fact, and using "SHOULD NOT" would create a normative statement that isn't ours to make.

[handrews]

Because that's a pre-existing fact, and using "SHOULD NOT" would create a normative statement that isn't ours to make.

Makes sense, thanks! I don't feel at all strongly about the RFC 5988 vs 3986 thing, so I'll close this.