RFC 1123, "Requirements for Internet Hosts - Application and Support", October 1989Source of RFC: Legacy
Errata ID: 1353
Publication Format(s) : TEXT
Reported By: John Klensin
Date Reported: 2008-03-08
Rejected by: Lisa Dusseault
Date Rejected: 2008-07-14
Section 2.1, ID 1081 says:
Errata ID 1081, reported 2007-11-20, identifies a problem with the evolution of naming of top-level domains and the text of RFC 1123. It reads: Section 2.1 says: However, a valid host name can never have the dotted-decimal form #.#.#.#, since at least the highest-level component label will be alphabetic. It should say: However, a valid host name can never have the dotted-decimal form #.#.#.#, since at least the highest-level component label will be not all-numeric. Notes: RFC 3696 section 2 states: "There is an additional rule that essentially requires that top-level domain names not be all-numeric." The eleven IDN test TLDs created in September 2007 contain hyphen-minus as specified in the IDNA RFCs.
It should say:
However, a valid host name can never have the dotted-decimal form #.#.#.#, since this change does not permit the highest-level component label to start with a digit even if it is not all-numeric.
This is a correct identification of the problem, but the wrong fix. RFC 3696, which ID 1081 cites, is an informational document that is deliberately relaxed about the fine details and says so. It is not relevant to determination of the text that should have been (with perfect knowledge of the future) in 1123.
Based on discussions when we were doing RFC 1591 and subsequently, the expectation then (and presumably when 1123 was written) was that the name of any new TLD would follow the rules for the existing ones, i.e., that they would be exactly two or three characters long and be all-alphabetic (which is exactly what 1123 says). The slightly-odd "will be" language in 1123 was, I believe, because that restriction was expected to be enforced by IANA, rather than being a protocol issue. ICANN, with a different set of assignment policies, effectively eliminated the length rule with the TLDs allocated in 2000. IDNA (RFC 3490) uses a syntax for IDNs that requires embedded hyphens in TLDs if there were ever to be an actual IDN TLD (hence the comment in ID 1081 about the IANA IDN testbed).
While the proposed correction in Errata ID 1081 would fix the problem by imposing the narrowest possible restriction ("not all-numeric"), the original host name rule and the original statement in 1123 both assume the possibility of a minimal check to differentiate between domain names and IP addresses, i.e., checking the first digit only. Because I believe that there are probably implementations that depend on such minimal parsing --some probably ancient and embedded-- it would appear to be wise to relax the rule as little as possible and, in particular, to restrict the "leading digit" exception to domains below the top-level, as 1123 effectively does.
The suggested text above reflects that reasoning. Because of the possible consequences of this issue, I would hope that it would be discussed with the relevant DNS-related WGs, the Root Server Advisory Committee (RSAC), and with IANA for comment and as a heads-up. This issue is substantive enough that it should probably be dealt with by a document that explicitly updates 1123 and that is processed on the Standards Track, but an accurate statement in the errata is the next-best option until that can be done. In the interim and while this suggestion is being discussed, Errata ID 1081 should probably be taken out of "validated" status.
This errata is in conflict with errata 1081. A new I-D and community consensus are probably needed.