Re: Issue-57

On Thu, Jun 16, 2011 at 7:35 PM, Xiaoshu Wang <xiao@renci.org> wrote:
> For me, I don't care because if there is no ontology that classify
> everything in the world into IR and if there is no reasoner that implements
> code to check every URI's response code and triggers subsequence inference,
> following httpRange-14 (even if I can) is simply a waste of time. Hence, I
> doubt I will ever do it. But for those who cares, 2xx code is cheaper than
> 303.
> Xiaoshu Wang

The reason for the httpRange-14 rule is that it lets us write metadata
for things on the web in a direct way, and be understood. The
inference rule you are looking for is: If U is a dereferenceable
absolute URI, and M(<U>) for some metadata M, and a retrieval of U
yields a representation Z, then M(Z). E.g. if an information resource
has dc:title "Little mouse", then its associated representations do,
too. Conversely, if M(Z) consistently for Z retrieved from U, then
M(<U>).

I agree with you: if nothing important follows about <U> from
retrievals of U, then there is absolutely no reason to assume anything
about the relation between <U> and U. But metadata is important and it
does follow. Maybe my logic is a bit too simplistic, but I think
metadata is the previously unarticulated justification for the rule.
And it is very consequential - statements of license and authorship
can have quite serious ramifications. There is no metaphysics
involved.

The rule is misstated and that has confused everyone; it seems to be
about an "information resource" type distinction but it makes a whole
lot more sense if it is about the relation between a URI and the
particular metadata subject (information resource, document, whatever)
*served from that URI* - not to just any old information resource, but
that particular one. I think the creators of the resolution took this
as a given.

I've written this up here:   http://www.w3.org/2001/tag/awwsw/ir/20110517/

Now, if you have a better way to write this kind of metadata, then you
don't need the httpRange-14 rule, but we absolutely need *some* way to
write metadata and I have yet to see anyone propose a better one. All
the alternatives I've come up with are very ugly as they involve an
occurrence of the URI as a string.

You ask what we gain from this; it's the ability to talk easily,
naturally, and consequentially about what's on the web. Practically
every rdfs:seeAlso, owl:imports, foaf:homePage, dc:creator,
xhtml:license depends on this convention. There's a ton of metadata
out there, often doing its job inconspicuously. People usually follow
the httpRange-14 rule without even thinking about it. That's why it
has so few champions. Without the rule, there is huge FUD about every
use of all of these properties, very serious ambiguity when a page
could be about a page, and we may as well just stop writing metadata
the way we do now. This is the direction we're headed if we can't get
stronger agreement.

(And by the way a lot of people think 303 means not-an-IR. There is no
justification for this idea and it is false in practice. A good
example is the 303 responses served from dx.doi.org, where the URIs
name documents without any commitment as to whether they are or aren't
on the web.)

I'm not trying to persuade you it's a good design (although if you
think otherwise I'd like to see what you'd substitute as a metadata
notation), I'm just trying to get you to stop saying the rule is
inconsequential or without serious purpose.

Jonathan

Received on Thursday, 23 June 2011 00:07:10 UTC