Use JSON Schema instead of prose descriptions.

[duesee]

We could replace the test case descriptions with JSON Schema and check the test vectors in CI. This way, it cannot happen that things get out-of-sync. I could give it a try when we want it :-)

[bifurcation]

It's a nice idea, but I don't think anyone will actually use this in practice. I would note that there are also semantics here that will not be captured by JSON Schema, e.g., the fact that a string represents a hex-encoded, TLS-serialized struct of a certain type.

Might be useful as a finishing touch once we have an agreed-upon set of test vectors, but probably not during development.

[duesee]

Yeah, maybe. The idea was also to agree on how to document the test vectors in a consistent way. I saw JSON Schema used for this in Project Wycheproof and found it to be very helpful. But I agree that it's probably too early to think about this. Edit: Unless, we use JSON Scheme to discuss the structure of new vectors or so.

Regarding the hex-semantics: Wycheproof uses, e.g., ...

        "key": {
          "type": "string",
          "format": "HexBytes",
          "description": "the key"
        },

... to define a hex-encoded key. There are some pre-defined formats in JSON Scheme but AFAIK "HexBytes" is none of them. So I'm not sure a validation tool could validate that.