> On 26 Jan 2026, at 4:46 pm, Jasdip Singh <[email protected]> wrote:
> 
> Hi Gavin,
> 
> From: Gavin Brown <[email protected]>
> Date: Monday, January 26, 2026 at 10:24 AM
> To: Jasdip Singh <[email protected]>
> Cc: [email protected] <[email protected]>
> Subject: Re: [regext] [Ext] I-D Action: 
> draft-ietf-regext-rdap-referrals-01.txt
> 
> > > “Servers which implement this specification MUST include the string 
> > > "referrals0" in the "rdapConformance" array in all RDAP responses.” … 
> > > Should this exclude search responses given the applicability of this 
> > > extension to just lookups? Including in the help response should be ok.
> >
> > No, I think it needs to be all responses, since it indicates *server* 
> > capability, not just the server's capability to extend search.
> >
> > [JS] I think this extension id should only be included in the 
> > “rdapConformance” array for the help response, since that signals server 
> > capabilities. For non-help responses, an extension id is expected to be 
> > included in the “rdapConformance” array only if the server used that 
> > extension to construct the response, which would be moot for these referral 
> > responses, no?
> 
> I stand corrected. RFC 9083 says "A response to a "help" request will include 
> identifiers for all of the specifications supported by the server. A response 
> to any other request will include only identifiers for the specifications 
> used in the construction of the response." So the next version will correct 
> this statement.
> 
> [JS] Great!
> 
> 
> > -02 feedback:
> >
> > Section 2: Wonder if the “<base path>” introduction is necessary? In the 
> > base spec (RFC 9082), typically a new path segment is introduced as a 
> > relative path and the examples start with the “https” scheme. IMO, might be 
> > good to emulate that.
> 
> There are two reasons:
> 
> 1. naive clients which just concatenate path segments together
> 2. servers which do not "normalise" request paths.
> 
> If you have a base path which is "/", a format for the request path which is 
> "/referrals0_ref/<relation>/<lookup path>", a relation of "related" and a 
> lookup path which is "/domain/example.com <http://example.com/ 
> [example.com]>", and you do naive concatenation, you end up with a request 
> path of "//referrals0_ref/related//domain/example.com".
> 
> For many HTTP servers, those double slashes are normalised, but there are 
> also many HTTP servers which don't do this. So for some servers the above 
> request path works, and for some servers it doesn't. Without explicit 
> instruction in how to construct the request path, you have an 
> interoperability problem.
> 
> [JS] OK. One further observation. The -02 draft currently says: "<base 
> path>referrals0_ref/<relation><lookup path>”. To your point, should it be 
> “<base path>referrals0_ref/<relation>/<lookup path>” instead, and that the 
> “<lookup path>” is relative?

Yes, you're right, as RFC 9082 specifies that <lookup path> does not have a 
trailing slash (so it can be concatenated with a Base URL, which always has a 
trailing slash). This will be fixed in the next version.

> > Section 3: Curious why we are still limiting 3xx responses to section 5.2 
> > of RFC 7480? Granted 308 (Permanent Redirect) happened after RFC 7480 
> > (AFAIK) but it seems a more appropriate response code than 301 (Moved 
> > Permanently) if one wants the HTTP method in the request to not be changed 
> > in the redirect.
> 
> While RFC 7480 says (at the top of Section 5) that "no standard HTTP response 
> code is forbidden", Section 5.2 then seems quite clear on the HTTP status 
> codes that should be used to do redirects, even if it doesn't use a BCP14 
> keyword.
> 
> Would an RDAP client implementation sticking solely to RFC 7480 struggle to 
> process a 308 response, because RFC 7480 does appear to limit redirects to 
> 301, 302, 303 or 307?
> 
> [JS] Since this spec is presently explicit about which 3xx responses 
> (currently limited to those from section 5.2 of RFC 7480) to return, not 
> explicitly including a more recent 308 in that list might confuse some server 
> implementors as to picking a less-appropriate 302 instead. Though, to your 
> question, clients shouldn't struggle processing a 308 even if a server sends 
> it in spite of exclusion from that 3xx list. Good to get other opinions here. 
> :)

Agreed. BTW, I had already planned to get some HTTP people to review this 
draft, but it would be good to hear from client implementers!

G.

--
Gavin Brown
Principal Engineer, Global Domains & Strategy
Internet Corporation for Assigned Names and Numbers (ICANN)

https://www.icann.org

_______________________________________________
regext mailing list -- [email protected]
To unsubscribe send an email to [email protected]

Reply via email to