Scott, I include feedback embedded below.
-- JG James Gould Fellow Engineer jgo...@verisign.com <applewebdata://13890C55-AAE8-4BF3-A6CE-B4BA42740803/jgo...@verisign.com> 703-948-3271 12061 Bluemont Way Reston, VA 20190 Verisign.com <http://verisigninc.com/> On 7/18/22, 9:11 AM, "Hollenbeck, Scott" <shollenbeck=40verisign....@dmarc.ietf.org> wrote: > -----Original Message----- > From: Mario Loffredo <mario.loffr...@iit.cnr.it> > Sent: Monday, July 18, 2022 4:40 AM > To: Gould, James <jgould=40verisign....@dmarc.ietf.org>; a...@hxr.us > Cc: Hollenbeck, Scott <shollenb...@verisign.com>; regext@ietf.org > Subject: [EXTERNAL] Re: [regext] OK, What Next? (was RDAP Extensions > Approach Analysis v2) > > Caution: This email originated from outside the organization. Do not click > links > or open attachments unless you recognize the sender and know the content > is safe. > > I agree with James. > > The drawback of Approach A is that even an additive change to an existing > extension would result in a breaking change to the RDAP service. As a > consequence, servers should always manage the transition from two > subsequent versions of an extension. Please explain how there's a breaking change. Let's assume that we have an extension named "foo version 1" identified by the prefix "foov1". "foov1" is registered with IANA, returned in the server's rdapConformance data structure, and used to prefix extension elements. Now assume that a second version of the extension (foo version 2) exists, identified by the prefix "foov2". "foov2" is also registered with IANA, returned in the server's rdapConformance data structure, and used to prefix extension elements. JG - There is an additional case of pointed versioning with an Internet Draft ("0.1", "0.2", "0.N" in EPP or "O_1", "0_2", "0_N" in RDAP), where the pointed versioning will change more frequently than what is reflected in the IANA registry for RFCs. If changing the hinted version in the RDAP Conformance also requires changing all the extension elements (path segments, query parameters, response members, and objectClassName values), even with backward compatible updates (e.g., inclusion of new optional feature), it will impact all client implementations, or it will discourage the use of pointed versioning to reduce the impact. Implementation to draft versions does come with risk, but it's important for drafts to get implementation experience, and we want to reduce the impact of making version updates. Using pointed versioning has become a best practice in the creation of the EPP extensions and should be used as well for RDAP extensions. If the server supports only "foov1" or "foov2", it returns only one of those values in the rdapConformance data structure, accepts only extension elements prefixed with the supported value, and returns only extension elements prefixed with the supported value. If a server supports both "foov1" and "foov2", it returns both values in the rdapConformance data structure, accepts extension elements prefixed with either value, and returns extension elements prefixed with the value that matches the requested value. So how does this transition scenario not work? Server supports "foov1" and returns that value in the rdapConformance data structure. The server accepts requests and returns responses prefixed with "foov1". The client sends requests and receives responses prefixed with "foov1". At some point in the future a new version of "foo", identified by "foov2", exists. The server enters a transition period and announces support for both extensions by returning both values in the rdapConformance data structure. It accepts extension elements prefixed with either value and returns extension elements prefixed with the value that matches the requested value. The client sends requests and receives responses prefixed with either "foov1" or "foov2" depending on which value of the extension they support. Time passes, and the transition period ends. The server deprecates support for "foov1" and announces support for only "foov2" by returning only that value in the rdapConformance data structure. The server accepts requests and returns responses prefixed with "foov2". The client sends requests and receives responses prefixed with "foov2". Where's the breakage here? In both cases, the client and server can identify extension elements by doing a simple pattern match for "foov1" or "foov2". JG - There is no true version negotiation of extensions in RDAP like there is for EPP, where some extensions only include response members and do not include extensions of the path segments or query parameters. A server would need to remove inclusion of the "foov1" response members and only use the "foov2" response members after some time of overlap. What is the advantage with potentially breaking the clients that have not been updated to look for "foov2"? The updated extension could be completely backward compatible. Support for version 1 and version 2 can be included in the RDAP conformance with the values ("foo_level_1" and "foo_level_2") for clients that are interested, but the response members with "foo" can stay as is. The same prefix "foo" can be used for the RDAP conformance values and the extension elements for consistency and guaranteed uniqueness. Non-backward compatible changes should be discouraged, but if needed the signaling in the RDAP conformance should provide the client with versioning information. The RDAP conformance member is meant for signaling extension support and is well suited to support versioning. Cascading versioning down to the extension elements can cause interoperability issues with little to no benefit to the clients, such as returning only "foov2" instead of "foov1" or "foov1" with "foov2". Scott _______________________________________________ regext mailing list regext@ietf.org https://www.ietf.org/mailman/listinfo/regext
