Re: [core] CRI Extension for multiple schemes

[Carsten Bormann <[e-mail]>]

On 2020-06-17, at 22:38, Christian Amsüss <[e-mail]> wrote:
> 
> Having a shortcut in CRIs sounds tempting, but I dread the
> translatability to the rest of the web ecosystem where those are
> unavailable, and the need to deal with aliased URIs in general.

Right.

So the question really is:

If we do our own URI variant, do we go all the way and exploit the benefits of having our own variant, even as a superset of URIs, or do we remain stuck with the URIs as they have been for a quarter century?

I’m maybe a bit more on the audacious side here, as long as we can keep the subset of CRIs that maps 1:1 to URIs large enough.  Apart from transport (really: plumbing) hints in URIs, we have always missed security hints (beyond the coaps bit) there.  I think if we could solve this problem, the divergence from URIs might be worth it — that is much less certain just for the plumbing.

The way these security hints could look like is of course way out in research land; if we want to explore this, we should have an extensible base architecture for CRIs and layer on that.

On the plumbing side, I think we already have seen that we need to associate at least a port number with a transport alternative, so we would not just enumerate URI schemes.

In the end, this could solve the aliasing issue as well, albeit only on the CRI side.

Grüße, Carsten