Support more recent drafts

[adeschamps]

Schemafy currently supports draft 4. What would be involved in updating to support the more recent drafts? Is this something I might be able to take on?

[Marwes]

I don't know how much has changed with the newer revisions but to get started you can try to update schema.json with a newer draft, regenerate the rust module representing a JSON schma according to https://github.com/Marwes/schemafy#development and then chase down the compile errors that occurs.

Supporting multiple, incompatible drafts could be possible but I doubt it is worth it.

[handrews]

@adeschamps @Marwes I would definitely focus on the newly released draft-handrews-json-schema-02 / draft-handrews-json-schema-validation-02, a.k.a. draft 2019-09.

The cumulative changes between draft-04 and this draft (which would have been draft-08 if we hadn't changed the numbering scheme to dates b/c it got confusing) are significant, but we expect 2019-09 to be a stable platform with only modest changes or compatible extensions before finalization.

2019-09 includes mechanisms for interoperable extensions without having to put every extension into the spec. 2019-09 will also be adopted by OpenAPI 3.1, which means that OpenAPI will move beyond its current state of semi-compatability with draft-04 for the first time. We expect that to drive further adoption and development, as it is probably the largest complex application of JSON Schema.

[adeschamps]

In order to get started on this, #3 needs to be implemented first; the new draft is basically an allOf clause that refers to other schemas.

[handrews]

@adeschamps yeah, that has to do with modularizing the keyword vocabularies.

Several implementations that I know of require the user to pass in all of the filenames (or parsed data structures, or whatever the interface is) up front. And then when they hit a $ref they just look at the set of schemas (specifically their $id values) that are already loaded to resolve it.

That's a lot easier than actually implementing file or http or whatever retrieval. It's also more flexible if your $ids are https:// but you're actually working with local file copies.

(my apologies if you already knew this- it was a point we've tried to clarify so people understand what they are and aren't responsible for to make a conforming implementation)

[adeschamps]

No, that's actually very helpful. It didn't occur to me that you could pass all the files upfront and that making http calls is not a requirement. That drastically simplifies things.