fix(test): rootless fixture should have no roots, not null roots

[rvagg]

I was trying to use this as a fixture and it's causing a bit of a problem. What we want this fixture to test is that the version or 42, so the error is invalid car version: 42. But, because it's got a Null roots it violates the schema in a novel way. The schemas for CARv1 headers dictate that the roots field is strictly necessary, and not nullable, but for the CARv2 pragma, we remove it entirely. So this effectively makes the initial bytes schema something like:

type CarHeader struct {
  version Int
  roots optional [&Any]
}

So it serves both the CARv2 pragma, and the CARv1 header proper.

But this fixture would make it:

type CarHeader struct {
  version Int
  roots nullable optional [&Any]
}

Which introduces additional awkwardness.

So I'd like to hand-wave this problem away entirely.

I'll also note here that it's a bit too easy at the moment to write a CARv1 with a Null in this position, WriteCar() and friends, and ReplaceRootsInFile() in CARv2 just pass the roots value on to cbor.DumpObject() which doesn't care about these nuances, it'll write a nil as a Null. I'm thinking that maybe we should switch from go-ipld-cbor to bindnode and assert a strict schema for the header to make this go away. The alternative is to add a nullable to the CARv1 spec (I started a PR to do this but then thought that maybe it'd be better to sweep this fixture under the carpet and defer this problem till later).

[masih]

Thank you sir 🍻