Support serving de-referenced / bundled schemas by default (unless specified otherwise)

[jviotti]

Instead of requiring ?bundle=1, configure the registry to always serve them bundled, unless otherwise specified

[jviotti]

Might be even worth always serving the bundled one. Optionally with i.e. ?bundle=0 otherwise. In theory, a bundled one is always better?

[jviotti]

Also see Apicurio/apicurio-registry#7264

[fetta]

While I agree that it would be great, @EricWittmann raised compatibility concerns in #7264 about compatibility - changing default behaviour is a breaking change and would probably need a wide consensus. It also would be less intuitive for devs.

But that's just my assumption, would need to get some community feedback on which behaviour is intuitive.

[jviotti]

Yeah, might be worth digging into. The thing is that for a compliant consumer of JSON Schema, getting it bundled should never break compatibility. It should be the exact same result the consumer would get if they would go about bundling themselves.

The only point would be if somebody does really need to make use of the unbundled one WITHOUT ever seeing the external references in there. But not sure I can think of such use case. And offer an option to fetch that might be enough?

[fetta]

Haha sorry I thought this was the Apicurio project. This would break apicurio compatibility, not schema compatibility. If you don't have dereferencing yet, the comment isn't valid.

[jviotti]

I gave this a shot today. The issue is a bit more subtle: for OpenAPI specs referencing pre-bundled schemas, seems like most OpenAPI editors out there or popular bundling tooling like redocly bundle get really confused with pre-bundled schemas and $id de-duplication on multiple references... My own implementation handles this fine, but a lot of the other ones don't.

So maybe we can provide special paths to get the schemas pre-bundled, so its all still at the URI path, and people can refer to those when needed.

I'm also planning to eventually implement int => URI internal associations for compatibility with Kafka, etc and maybe the pre-bundling can happen there?

[fetta]

I'd love to learn the results of this analysis but I'm afraid I can't help much here - I don't know this ecosystem well enough.