xml-dev - Re: URI concerns continue

[Eric Bohlman <ebohlman@netcom.com>]

On Mon, 10 Jul 2000, Simon St.Laurent wrote:

> It seems like the W3C is pushing hard to use URIs to identify everything
> possible, without giving much concern to the difficult questions that have
> bedeviled the xml-uri list for the past two months.  While the URI
> community seems happy to have identifiers that are processed using a wide
> variety of different rules, I'm not sure that approach is acceptable for
> XML processing.  (It could become acceptable were URI processing and
> comparison, at least in an XML context, defined more clearly and explicitly.)
> 
> While I welcome the effort to reuse existing parts, and could be persuaded
> that URIs are a good answer to all of these problems, I don't think that
> URIs are anywhere near mature or specified enough to carry the burdens that
> various XML specifications are placing on them.  There is no reason to
> expect URI processing to be as predictable as XML processing (even with the
> many possibilities in XML processing), and I'm very worried that URIs are a
> threat to XML processing in the absence of constraints that would make URI
> processing and usage predictable.

I think that one of the problems here is that we've got a "terminological
hijack" going on, similar to the one I mentioned a while back (though I'm
not sure whether on xml-dev, xml-uri, or both) involving the American and
British uses of the phrase "learning disabilities."  In the decade or so
that URIs (almost entirely in the form of URLs) have been in use, they've
been the identifiers (in reality, addresses) of *concrete* things, namely
entity bodies, and a generation of developers and users have built a
mental model of URIs based on this use.

But it's plain that the W3C's vision of URIs is heavily based on using
them to name *abstractions*, and that a good part of the URI community is
having a hard time fitting this into their mental models.  I'm starting to
wonder whether overloading the notion of a URI (particularly in the URL
form) to encompass both addressing sequences of bits and identifying
abstract statements in a form of higher-order-logic is really a good idea.  
Maybe the distinction between floor waxes and dessert toppings actually
serves a useful purpose in practice if not in theory and shouldn't be
blurred solely in the name of mathematical elegance.  Imagine a
programming language in which all arithmetic had to be done in terms of
set-theoretic primitives.  It would be extremely elegant, but not very
useful.

In particular, I have a really hard time with the use of a
retrieval-protocol component (like http or ftp) in something that's
supposed to name an abstraction rather than specify how to retrieve
something concrete.  I'd be a lot happier if we were to reserve URLs for
naming concrete things and keep the names of abstract things to a subset
of URNs or to some sort of public identifier.  To do otherwise is, IMHO,
to invite a great deal of map-territory confusion, and to leave people
wondering whether particular maps actually correspond to any territory.