- From: Tim Kindberg <timothy@hpl.hp.com>
- Date: Mon, 21 Jan 2002 08:42:55 -0800
- To: Patrick Stickler <patrick.stickler@nokia.com>, "ext Sean B. Palmer" <sean@mysterylights.com>, sandro@w3.org
- Cc: URI <uri@w3.org>, URN <urn-ietf@lists.netsol.com>
At 09:29 AM 1/21/02 +0200, Patrick Stickler wrote: >On 2002-01-21 5:37, "ext Sean B. Palmer" <sean@mysterylights.com> wrote: > > > Hi Patrick, > > > > Sorry for taking so long to go through your recent URx publications; > > here are some fairly innane questions, and some general comments. > > > > My first Q is about the hierarchial URN scheme. Michael pointed out > > that they can have domain names as authority components, which isn't > > all that persistent. UUIDs would be alright, but they're fairly > > difficult to generate. The other option is a tag-esque "domain,date" > > component - which brings me to the question: how does "hrn:" differ > > from "tag:"? As one of the people who devised tags -- http://www.ietf.org/internet-drafts/draft-kindberg-tag-uri-01.txt -- I'm having trouble understanding Patrick's hrn: design and his comments below. One of the rationales for tag: has always been to separate minting from binding and make minting convenient for humans, so we seem to share that. An important aspect of minting, obviously, is a guarantee of uniqueness. Email addresses and domain names belong to at most one entity at a time and are thererfore suitable 'seeds' for uniqueness -- hence we use them in tags. But they're not uniquely assigned over time. Therefore we qualify them with a date on which they're assigned: tag:timothy@hpl.hp.com,2002-01-21:ikaika, which is a name that I have just assigned for my dog, is guaranteed to be unique now and forever, because I own that email address on today's date. No-one -- in particular, a timothy@hpl.hp.com at HP in the year 2034 -- is allowed to mint a tag for any date other than one on which they own the email address/domain name, so they can never legitimately choose the same name. In contrast, the person who holds abc.com in 2034 might 're-mint' hrn://abc.com/288918293/3/en/global/docbook. At least, there may be no records of what previous owners of abc.com have minted, so the new minter can't be sure. As for the hierarchical nature of hrns -- sure, hierarchies are nice. But why mandate a syntax to that extent for the 'local' part of the identifier? It's nobody else's business; it just has to be unique. By the way, tag: is in the RFC editor's queue. It's also somewhere in the urn nid registration process. Tim. >As I pointed out in an earlier response, the use of the web >authority does not make an 'hrn:' non-persistent, as the >web authority in an 'hrn:' represents only the minting >authority, and not the resolution authority. In an 'http:' >URL, a web authority is both, and thus if it changes or >becomes inactive, the resolution authority becomes invalid >and thus persistence is impacted. > >'hrn:' URNs do not have that problem, even though web authorities >may be used, because the web authority remains historically valid >and even if it becomes inactive, that has no impact on the >persistent validity of the 'hrn:' URN. > >See section 3 of draft-pstickler-uri-taxonomy-00 regarding >these distinctions. > >The benefit of allowing a web authority as the minting authority >is that it provides a far less centralized method of "namespace" >than 'urn:' NIDs or purchasing of URIs from NID owners (e.g. DOI, >ISBN, etc.). With 'hrn:'s, anyone can mint a mnemonic URN >based on a web authority, or a non-mnemonic (or partially >mnemonic) URN based on a UUID easily and without fear of >collision. And those URNs can be hierarchically defined to boot. > >'hrn:' = "URN Power to the People" ;-) > > > I presume that the hierarchial aspect is what you're > > after, although I'm not sure what the relationship between the > > segments is. > >Hierarchy was one desired characteristic. Minimially-centrallized >minting was another (i.e. based on any web authority, not having >to register a namespace or pay some agency that has a namespace). > >Per above. > > > Why not use ":" as a hierarchial segment delimiter in > > "tag:"? > >Because there already is a standard syntax for hierarchical URIs and >many APIs and libraries exist for parsing such hierarchical URIs into >their components. > >Why introduce another hierarchical syntax if the present standard >does the job? > > > Note that "tag:" was going to be registered as a URI, and URN > > NID. > >I understand that. But it appears to me that it is only the >contemporary view that forces this redundancy. You should be >able to register it simply as a URI with a classification of URN. > >Or, you can register it as an NID of the 'urn:' scheme. > >Why do both? > >Having both 'xxx:foo' and 'urn:xxx:foo' is sure to result in >alot of needless overhead. > > > Next Q: I can't work out what "voc:" is for, if anything. The draft > > states: "This provides a more robust and safe treatment of unqualified > > names than the 'online:' or 'genid:' treatments employed by most RDF > > systems to date.", so it sounds as if they're meant to be replacements > > for anonymous nodes... but the structure of the URIs suggests > > otherwise. > >The use of 'voc:' in RDF for providing non-collisive identifiers >for local (not anonymous) resources is very much a fringe use. > >The 'voc:' URT scheme is intended for vocabularies. See the >examples in the I-D. That should clarify its intended usage. > >In short, for the URIs of all element and attribute names of all >XML content models, for the URIs of all resource names of all >RDF schemas, for the URIs of all controlled vocabularies >and taxonomies such as ISO language and country codes, TGN >geographic names, etc. etc. > >I.e. for all abstract identifiers. > >I think that the 'voc:' URT scheme is the most important of >all of the newly proposed schemes, and has the widest application. > > > I've also been wondering about the taxonomy in general. A lot of > > people will tell you that an HTTP URI is just as good a persistent > > identifier as any URN - it's the social contract that matters, and > > HTTP URIS are widely deployed. The URN/URL/URP taxonomy feels rather > > artificial to me, and I fear that creators of new schemes will have to > > beg to you as the arbiter of where a new scheme belongs. > >I agree that before such a taxonomy would reach the maturity >of a standard that it would need more precise definition. I think >though that the criteria distinguishing the proposed classes is >fairly straightforward. > >If your URI scheme denotes a point of access, it is a URL. If it >denotes an indirect access key, it is a URN, if it does not resolve >(is self contained) it is a URP. > >I.e.: > > URL direct resolution > URN indirect resolution > URP no resolution > > > [BTW, I'm not > > sure I would have chosen the acronym "URP". Every time I write it, I > > feel like excusing myself afterwards]. > >It does seem to give a few folks indigestion ;-) > > > For example, you've listed ESL as a URV. I can see the motivation > > behind that, and I would agree - if not for the fact that ESL could > > easily have been submitted as a URN NID. > >But why? Is the primary and fundamental purpose of an 'esl:' URI to >denote some other web resource independent of its location? > > > At the moment, I am one of > > those who feel that the boundary between URP and URN is not all that > > solid. [In fact, I wonder why there isn't a "uri-x:" alternative of > > "urn:urn-x:".] > > > > I'm not sure what the utility of the "qname:" scheme is. In fact, many > > of the drafts are lacking in describing the utility of the schemes > > themselves. Whilst this seems to be common practise, it's something > > that I battle against. All new schemes should have a detailed space > > describing their purpose and motivation, because it obviates arguments > > later on. If you could prepare a summary of the aims of each scheme, > > that would be rather useful. > >All of the schemes have specific utility but I tried to avoid being >too specific with regards to all possible applications, keeping each >I-D focused on the essential technical details regarding the URI >scheme in a generic fashion that would be as future-flexible as >possible. > >I am actually working on a more descriptive account of where/how each >is used, which, if I get it done in time, I will likely submit for >the SW track at WWW12. > >Cheers, > >Patrick > > >-- > >Patrick Stickler Phone: +358 50 483 9453 >Senior Research Scientist Fax: +358 7180 35409 >Nokia Research Center Email: patrick.stickler@nokia.com Tim Kindberg mobile systems and services lab hewlett-packard laboratories 1501 page mill road, ms 1u-17 palo alto ca 94304-1126 usa www.champignon.net/TimKindberg/ timothy@hpl.hp.com voice +1 650 857 5609 fax +1 650 857 2358
Received on Monday, 21 January 2002 11:36:40 UTC