- From: Martin Bryan <mtbryan@sgml.u-net.com>
- Date: Thu, 28 Nov 1996 13:01:37 +0000
- To: "Deborah A. Lapeyre" <dlapeyre@mulberrytech.com>, Paul Prescod <papresco@calum.csclub.uwaterloo.ca>
- Cc: w3c-sgml-wg@w3.org
At 13:29 27/11/96 -0800, Deborah A. Lapeyre wrote: > >On Wed, 27 Nov 1996, Paul Prescod wrote: > >> But PUBLIC identifiers put that redirection under the control of end-users. >> There are numerous techniques for doing redirection on the server side. > >And that is why they are critical to me. Not that there may be lookup >but that there must be and on the client side. I can't agree that lookup should be on the client side. >But perhaps that is exactly what we are trying to avoid. >A URL is all on the front end. An FPI is >by definition NOT on the front end and guarantees indirection. >Are we trying to forbid indirection and force front-endedness? >Is that the objection? Hopefully not. An interesting statement that was made at the W3C Internationalization and Multilingualism Symposium last week was that persistent URLs (PURLs) needed to be stored in a "library" that guaranteed long term access to the PURL. The idea is that the library would redirect the PURL to the current storage location for the relevant file. Consider what this means. A library can be looked at as a "set of catalogued references to objects stored in a set of local storage units". The catalogue provides both the indirection and the means of identifying where multiple copies of an object can be found. It also allows the same object to be identified in more than one way, e.g. by author name or by title. Consider if everyone wanting to create a reference to a paper on SGML decided, instead of referencing the file directly, to reference the entry for the paper in Robin Cover's bibliography. This would then point to the server(s) currently containing copies of this paper. If one wanted to create a link to anything about the subject you could do so by pointing to the public catalogue, whose URL would be much more persistant than any maintained at the individual sources. What does this tell us about SGML catalogs? Firstly it suggests they are currently stored in the wrong place. They are currently stored in the relocatable servers or in clients: I contend that they should be stored in an SGML Open Catalogue Library or some similar form of long-term digital library. Secondly it suggests that the FPIs used to identify the objects to be indirected do not need to be unique. There could be more than one way of identifying the same file using an FPI (e.g. an unregistered name when the file was first created, and a set of registered ones used for later references to the same file). It also implies that different catalogues in the same library can provide different ways of referencing the same object. >P.S. >Because of ownership, accidental name duplication is very easy to avoid. >Both registered and unregistered (the 99% case) provide simple but very >real duplication protection. (It is easy to screw up your own stuff but >very hard to screw up someone elses if you follow the rules at all.) > Funny, I now seemed to have covered Debbie's last point as well! You don't need ownership of 'your catalog', and you don't need to avoid name duplication in FPIs. What you actually want to do is to decouple the catalogue entry from its owner and encourage identifier duplication. What you need to do first, however, in encourage the use of indirection, which is the whole point of URNs, URCs, etc. They all need a directory/catalogue mechanism in some form of "library" to be able to work (as does the DNS service needed to resolve URLs). Martin Bryan ---- Martin Bryan, The SGML Centre, Churchdown, Glos. GL3 2PU, UK Phone/Fax: +44 1452 714029 WWW home page: http://www.u-net.com/~sgml/
Received on Thursday, 28 November 1996 08:01:33 UTC