On Tue, Jan 07, 2014 at 11:29:04AM -0500, Phillip Hallam-Baker wrote: > It didn't have to be a tree, it could have been something else, we could > still change it. Gee, Phill, I love it when people elide the exact point I made when selectively quoting, and then chide me for not having made that point. Yes, it could be something else. In the logically-possible world in which I am a mud puddle, I spend less time responding to emails about the nature of naming on the Internet. But both the naming system on the Internet and I (perhaps unhappily for one of us) exist in this world. The Internet as it stands uses the DNS, a tree-structured name space. A tree-structured name space has, by definition, a unique root, due to the math in this universe.[1],[2] > Describing this as having a single root conflates a large number of issues > and essentially commits to a particular conclusion. A uniform namespace is > a requirement, a single 'root' is not. I wasn't claiming a single root is a requirement for a uniform name space; I don't care about that question. I was suggesting that, given that we have the DNS and it is the name system that is actually deployed, we have right now a requirement for a unique root in that name system. If you want to make up a new non-DNS naming system and deploy it and get the world to use it, please go nuts. I have plenty of complaints about the DNS and the way it works. I suggest that the history of IPv6 deployment prior to the actual exhaustion of the IPv4 number space gives us a pretty good lesson in the likelihood of non-DNS global naming taking off without some killer feature. In my opinion, "Doesn't make Vladimir Putin grumpy," is unlikely to be the killer feature that will cause people to change whatever invisible naming system they use to bootstrap the applications on their tablets or find Google or Bing. Best regards, A [1] To acknowledge Avri's point elsewhere in this thread, yes, it is possible to have a different CLASS with different RDATA at the name owner name and RRTYPE. The name space is still tree shaped with a unique root, in that the owner names persist across all CLASSes. This is perhaps not quite as obvious as one would like in RFCs 1034, 1035, and maybe 2181. DNSEXT attempted to do something about this a few years ago, but we didn't get any traction so we gave up. It is undoubtedly the case that there was some reason people didn't do something about the difficulties with CLASSes. One of the explanations, of course, is that if we were going to fix the DNS at that basic a level, we might as well replace it completely with something else that doesn't have the properties we'd be trying to repair. [2] I refuse to have a position on whether there is a logically possible world in which mathematical tree structures do not by definition have a single root. Possible-worlds semantics over mathematical proofs was probably too hard for me when I actually did possible-worlds semantics work, and by now I've forgotten most of that Latin. -- Andrew Sullivan ajs@xxxxxxxxxxxxxxxxxx