Re: [Cbor] I-D: draft-rundgren-cote-00

[Carsten Bormann <[e-mail]>]

On 2022-07-19, at 15:52, Christian Amsüss <[e-mail]> wrote:
> 
> Signed PGP part
> Hello Anders,
> 
> On Tue, Jul 19, 2022 at 02:40:37PM +0200, Anders Rundgren wrote:
>> Since I did neither make it to Philadelphia nor to the cut-off date,
>> here is a private publishing:
>> https://cyberphone.github.io/Internet-Drafts/draft-rundgren-cote.html
> 
> Sounds reasonable to me, two comments:

Agreeing with the below comments, but if I had my designated expert hat handy, I would say:

This is underspecified as the specification does not really spend any effort ensuring that the type identifiers are unique to a specific usage.  This is likely to lead to interoperability problems.  I would think a few more things could be said about the type identifiers that would reduce these hazards.  Re the example given: URLs in general are surprisingly bad as stable identifiers, while a few specific forms exist that are much better.

> * Given that there are full URIs in there, compactness is barely a
>  concern. Thus, I'd appreciate if you asked for a tag from the range
>  above 256.

Indeed.  Going above 32768 would also take the designated expert out of the loop, so the format can be as foolish as one wants :-)

> * Please consider what of RFC 6648 applies here. When would a protocol
>  author pick a type extension over a tag, and what would happen if such
>  a type extension becomes popular outside the community or application,
>  and maybe users push for registering an equivalent registered tag?

This is part of a fundamental problem in a totally open type identifier system: Why would the generator spend effort fitting into an existing type when they can simply create another identifier?  Such a system should instead nudge its users towards behavior that is in favor of more interoperability.

I think we found a pretty good way to handle this problem in https://www.ietf.org/archive/id/draft-ietf-core-problem-details-08.html — yes, we do allow URIs in certain places, but we are even proposing a good class of URIs in the example.

Grüße, Carsten