John C Klensin wrote: [RFC 3461, 3464, and 4409 references only informative] > its appearance in that particular SHOULD does not make the > reference normative -- one does not need to understand RFC 4409 > in order to make a competent and conforming implementation of > 2821/ 2821bis.
Implementing 2821bis is one thing, but many readers might be only interested in understanding how SMTP is supposed to work today - admins, postmasters, users, and email-arch authors. Here RFC xxxx about X says that users SHOULD do Y in RFC yyyy. For that they need to read and understand RFC yyyy, otherwise they could end up not doing Y without a good excuse. In the case yyyy = 4409 it is also relevant for implementors, if what they implement is all in one box, MX, MSA, and final delivery MTA. > The cast that 3461 and 3464 are normative is a little > stronger but, if you read the relevant sentence, what is > it basically saying is that, if you find another standard > that is applicable to NDNs and the particular situation, > then you should use it. The statement in 3.6.3 says "for example", but admittedly I'm not aware of other "examples". RFC 3461 [35] is also needed in an ABNF comment: | the "xtext" syntax [35] SHOULD be used. That is as normative as it can get, implementors won't know when and why they violate this SHOULD without first looking in RFC 3461. RFC 3463 appears in the "for example" SHOULD statement, but also in another SHOULD about extended error codes: | the system described in RFC 3463 [36] SHOULD be used in | preference to the invention of new codes. That is normative for anybody considering new error codes. Chapters 2.3.7 and 3.5.1 contain further pointers to RFC 3463. The RFC 3464 [37] "for example" may be informative, but RFC 3461 [35] and RFC 3463 [36] are IMO normative. OTOH I don't see why RFC 1123 is "normative": It broke a clear concept of responsibility in RFC 821, its attempt to simplify RFC 974 wasn't too good after all, and 2821bis replaces everything in the relevant RFC 1123 chapter 5. IMO RFC 1123 belongs to the set of informative references (974, 1047, 1652, 1869, 1870, 2821) obsoleted by 2821bis. Frank
