On Fri, Jul 17, 2020 at 5:39 PM Mark Nottingham <mnot@xxxxxxxx> wrote:
On 18 Jul 2020, at 4:31 am, Larry Masinter <LMM@xxxxxxx> wrote:
>
> RFC 7595, "Guidelines and Registration Procedures for URI Schemes", June 2015
>
> as modified by https://www.rfc-editor.org/errata/rfc7595 was a mistake.
>
> The idea behind RFC 7595 was to change https://www.rfc-editor.org/info/bcp35 to make “provisional” registration FCFS.
> And the errata was correct in its identification of the problem that FCFS is incompatible with having requirements that need to be checked.
> But the proper remedy (as this case shows) should have been to change it back to “expert review” as it was in RFC 4395.
>
> If “expert review” could optionally be accomplished by (dis)approving a Pull Request in an IANA-maintained Git repository (as was recently done for .well-known) so much the better.
I was wondering the same thing. I know we had extensive discussions about that draft, and IIRC at the time several folks (including me) pointed out that FCFS is incompatible with the notion that new schemes should be rare.[1]
Looking at the current registry and the large number of provisional registrations without a specification, I'd be sorely tempted to make it Specification Required, so that there's at least a chance of interoperability.. AIUI the counterargument is that since many vendors have used the scheme as an escape valve to trigger proprietary functions, making it spec req'd would disqualify too many registrations. It might be worth looking at what encourages them to use it in that way (perhaps with the help of WHATWG folks?) and see if there's a better solution.
Also, I note that there doesn't appear to be any mechanism for removing a provisional registration that's fallen into disuse; for example, several appear to reference old Internet-Drafts that have never made it to RFC.. If provisional registrations can't be removed, is there any practical difference between provisional and permanent?
To add some color to this, there has been discussion of using this as input to the question of whether browsers would safelist schemes for registerProtocolHandler:
-Ekr
