- From: Karen R. Sollins <sollins@lcs.mit.edu>
- Date: Mon, 26 Jun 1995 12:40:49 -0400
- To: peterd@bunyip.com
- Cc: uri@bunyip.com
Hi Peter, From: Peter Deutsch <peterd@bunyip.com> Date: Sun, 25 Jun 1995 12:02:38 -0400 X-Mailer: Mail User's Shell (7.2.4 2/2/92) Cc: uri@bunyip.com Sender: owner-uri@bunyip.com Precedence: bulk Hi Karen! [ You wrote: ] } Peter, } } We've been through some of these arguments about the degree of } user-friendliness before (I'm sure you were there - didn't we do some } of this in Houston?) <sigh> Yup... } . . . I believe that } we've perpetuated a slight mistake in the URI WG by calling these } things URNs ("names") because too people assume that a name has to be } human friendly. These things should be as human unfriendly as we can } get away with, to discourage their direct use by humans. They are } serving a different purpose at a different level of abstraction than } human friendly names. I agree, we're definitely talking about two sets of functionality here, although I'm not yet convinced there's something wrong with calling these things "names". We just need to be careful about what we think we can do with them. For me, the requirements for comparision and dereferencing remain the defining elements. I agree. The reason I shy away from the term "name" is that I hate having arguments with people about terminology. Too many people have seen "name" in 'URN" and not read carefully enough to realize that these things may not match exactly all their previous assumptions about what a "name" is or is not. Then they proceed to have long, pointless arguments with me about something they think I said that I didn't say. I'm just a little gunshy at this point. Anyway, it's too late now. Depending upon how we do it, I think dereferencing does mean we need a human friendly part for when humans must select these things. For dereferencing, we obviously want to focus on the mechanical processing capabilities and might be willing to sacrifice readibility here if it makes it work. But more to the point, comparison and dereferencing are indeed the important issues. If we have some other human friendly scheme(s) for giving things nmemonic names, then why should people need to see URNs if they get involved in dereferencing? Maybe I'm misunderstanding something here. If there is other information about particular instances that the human needs in order to help in dereferencing a URN, that information should come from the URC not the URN. It is meta-information, and I would like very much to see us stay away from putting meta-information into the URN. Taken together this probably means that a URN probably needs two components, and that we need to be able to process/handle the two separately. Still, I think they also need to remain bound together as in gopher menu items. So, perhaps this means a URN should look like: <naming authority>:[<optional human readible string]>:<unique ID> AARRGGHH!! Remember the discussions about putting mutable information about a resource into the name? I still think we shouldn't do that. So far, other than a content free identifier, the only other fact I found so far that is intended not to change across the board for resources is the fact that they are immutable, if they are immutable. (A mutable resource might become immutable, but the intention is that this won't happen the other way around.) But I don't think that's enough semantics to make interesting human readable names. Note! I'm not proposing a syntax, I'm proposing a set of components here... Yeah, I know, but that's not the issue...(see my comments above) - peterd Someone also pointed out that I was contradicting the requirements doc for URNs by suggesting that URNs should not be human friendly. Unfortunately, we had lengthy (measured in years) arguments on such topics. The intention was that "human transcribable" not be the same as "human friendly". "Human transcribable" was supposed to capture the idea that if you really needed to, you could type it on your keyboard; more likely you might cut and paste it. It was not intended to suggest that you ought to WANT to do these things. In fact, rather the opposite. (Just as an example, in our project, the ids we've been generating use only 16 letters, none of them vowels. You can type them, but you probably won't get a lot of meaning out of them.) Just as a piece of history, there were very similar discussions as the DNS was being designed. Although the hierarchy was to exist, early on it was intended first that those names would NOT be the user friendly names, but that there would be another scheme on top. Second, people became attached to places in the DNS hierarchy. It is important that as a company you have a top level domain under .com or co.uk or whatever. If your company had some name 4 or 5 levels down, it (or rather the people running it) would probably be pretty annoyed. Given that a namespace was used to which people could attach meaning, they did and they never got beyond that and its consequences. Are we following the same path? Karen
Received on Monday, 26 June 1995 12:40:15 UTC