Ref v2 Spec

[Julian]

This is some random notes on a new version of the ref resolution API. Feel free to leave comments, although some of the below might be overly terse chicken scratch

• It should probably be easier to pre-resolve all refs (maybe that should even be the default) rather than lazy/forced resolution
• Keep in mind async use cases
• Too many methods bruh. Just 2? Resolver().resolve(ref) + Resolver().for_scope(url)?
• Take yarl.URLs, not strs
• What should be the thing for users who want to resolve relative to files on disk?
• Move the resolver argument to extend

[agoose77]

I've just been thinking about the problem of following $id declarations during ref resolution and came across this issue.

I don't think that we should recursively dereference schemas. Whilst it's certainly a simpler approach (at least, conceptually), I think the extra cost should be avoided (especially if there are schemas with lazy reference paths (though I'm not sure if currently jsonschema does this, as even in the case of oneOf we need to ensure only one subschema validates (and hence we have to dereference it anyway)).

I think a better approach would be to follow the $id uri from within the resolution process, such as the approach taken here. I'm not convinced that we could (or should) just use this repository (for a start, it doesn't support the newer $id keyword, but the approach is valid enough (we already use context managers to track the current scope).

I'll give it some thought.