v6 hyper-schema: extended templating (vars+template) about json-schema-spec HOT 2 CLOSED

Opinions on various points:

vars should allow schemas in addition to relative JSON Pointers

The vars section should not only describe how to resolve the variables from an instance, but also support validation of user-supplied data when resolving the templates in other ways. This should work as follows, taken from Riverbed Technology's service definition format which made extensive use of this mechanism:

An entry in the "vars" field can either adjust the starting point for that resolution through a relative JSON pointer, or provide a schema, indicating that the value must be provided from an external source. Any schema, whether implied through the resource's schema or given explicitly, must describe a type that can be resolved through the rules for that variable as described in the URI templates standard (RFC 6570). Otherwise the resulting behavior is undefined.

While we found it to be a best practice to always be able to fully resolve the self link from the instance data, it doesn't seem to be something that should be strictly required.

Legacy pre-processing rules

Toss them. They're ugly and confusing and not worth the burden of carrying them forwards. It's also not horribly difficult to migrate from them to vars. People who don't want to update them can keep using the Draft 04 meta-schema until they are ready to move forwards.

Templating $ref

No use case is given for this, and it introduces considerable complication. I move that this be dropped unless/until compelling use cases are identified. It clearly could do a lot of things, but it's not clear that those things are desirable.

from json-schema-spec.

Having split the most urgent portion of this out into #142, I am going to close this. It's an omnibus that is hard to discuss all at once.

Templating $ref is not something I have heard anyone else request. If someone wants that, please open it as its own issue. The question in that issue should just be whether to template $ref, yes or no. The question of whether templates use pre-processing, mapping, or some other mechanism should be decided separately and applied to all template uses (currently href and base, possibly also $ref and rel).

My comment tacked on more proposals about user input and validation. That's a complex topic, the requirements for which are being explored in #108 which is a better vehicle for the discussion (it's at least garnered a few comments, while this issue has not).

from json-schema-spec.