Can we define the behaviour for a missing reference? about wg-reference-types HOT 7 CLOSED

I'm aware this may be too late, or that this might need to be raised elsewhere now, so feel free to say that!

from wg-reference-types.

For a refers field to a manifest that doesn't exist, the call we made was that the registry MAY return an empty list before that reference exists, but once the reference exists, the referrers API MUST include previously pushed entries. That lets us, for example, push a signature before pushing the image. This wording is going in the next PR to the oci-playground/distribution-spec.

from wg-reference-types.

See oci-playground/distribution-spec#11 in spec.md under Listing Referrers:

If <ref> does not exist, a registry MAY return an empty manifest list.
After <ref> is pushed, the registry MUST include previously pushed entries in the manifest list.

from wg-reference-types.

Thanks, just went and had a look on the playground.

What I didn't find was any guidance on what the registry is allowed to do when someone attempts to upload an artifact that points to a digest that doesn't exist. Has that been discussed too?

from wg-reference-types.

Good point, I'm adding this to the PR:
https://github.com/oci-playground/distribution-spec/pull/11/files#diff-bc6661da34ecae62fbe724bb93fd69b91a7f81143f2683a81163231de7e3b545R204

A registry SHOULD accept a manifest with a refers field that references a manifest that does not exist.

from wg-reference-types.

That works for me, mainly because I'm thinking about image indexes and that weak references should be a supported thing, also because I think being able to push a graph of references in parallel makes sense.

Going to tag @milosgajdos and @deleteriousEffect because we had a similar discussion a while ago.

from wg-reference-types.

This is in the PR to distribution-spec, so I'll mark this as closed over here, but feel free to keep the discussion going.

from wg-reference-types.