You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
[Feature Request / Contribution Proposal] working with internet drafts #112
Hi, and first of all thanks for the crate, it great, love it.
When working with internet drafts there are very often a lot of pending/requested IANA codepoints that are not within the private range and hence cannot be used within the ClaimSetBuilder which is a very convenient way to build a Cose object.
One have to resort to do the CBOR serialization by hand on the side to bypass this line.
If the draft is reasonably stable it might be easiest just to update the enum values to include the draft values, albeit with a risk that a future change in the value would necessitate a breaking change to the crate. For example, we're likely to add (0dd34d7) the draft values for ML-DSA.
The other alternative is to change the RegisteredLabel types as you suggest. I had a go in #114, and it does result in a fair amount of non-back-compatible change.
OIDC claims, WGLC should happen fairly soon as well. 19 claims this time 😅
The datatracker for that one still seems early/unstable – IESG state is just "I-D Exists"?
This one is mine, I can assure you WGLC is close ;)
Anyway, my issue is more about the next time I'll be implementing a draft, the issue will still be there. If adding an Unassigned variant and merging that in the next major release of the crate could be done that would be awesome 😁
Activity
Allow unassigned values for labels
daviddrysdale commentedon Sep 15, 2025
Which draft IANA codepoints do you need?
If the draft is reasonably stable it might be easiest just to update the
enumvalues to include the draft values, albeit with a risk that a future change in the value would necessitate a breaking change to the crate. For example, we're likely to add (0dd34d7) the draft values for ML-DSA.The other alternative is to change the
RegisteredLabeltypes as you suggest. I had a go in #114, and it does result in a fair amount of non-back-compatible change.beltram commentedon Sep 15, 2025
Hi @daviddrysdale thanks for having a look.
I have 2 drafts on top of my mind for which I required this:
daviddrysdale commentedon Sep 16, 2025
Looks like there's 3 now?
The datatracker for that one still seems early/unstable – IESG state is just "I-D Exists"?
beltram commentedon Sep 16, 2025
Yes right, 3 indeed.
This one is mine, I can assure you WGLC is close ;)
Anyway, my issue is more about the next time I'll be implementing a draft, the issue will still be there. If adding an
Unassignedvariant and merging that in the next major release of the crate could be done that would be awesome 😁