W3C home > Mailing lists > Public > public-awwsw@w3.org > June 2009

Re: Are generic resources intentional?

From: Jonathan Rees <jar@creativecommons.org>
Date: Wed, 10 Jun 2009 12:22:06 -0400
Message-ID: <760bcb2a0906100922t35b6c96bm76cf9c2b4927ede9@mail.gmail.com>
To: Pat Hayes <phayes@ihmc.us>
Cc: "Williams, Stuart (HP Labs, Bristol)" <skw@hp.com>, David Booth <david@dbooth.org>, Alan Ruttenberg <alanruttenberg@gmail.com>, "noah_mendelsohn@us.ibm.com" <noah_mendelsohn@us.ibm.com>, AWWSW TF <public-awwsw@w3.org>
On Wed, Jun 10, 2009 at 10:55 AM, Pat Hayes<phayes@ihmc.us> wrote:
>> Are images agents?
> No, but I think the key phrase in the above is "which can be named", which
> seems to be playing the same role that "has an identity" was once used for.
> This notion of "named" or "identified" doesn't mean 'referred to': it means,
> in this context, "can be accessed by using its name", which when you unpack
> it carefully means something like "is being handled by some active system or
> agent which can deal effectively with transfer protocols for appropriately
> syntaxed names".
> But OK, I concede that this quote does muddy my water. I am puzzled. Roy
> also says that a resource is conceptually a function from times to
> representations (and in 2000, that a solid, physical book on a shelf is also
> a resource, a position which *really* does not make sense.)

He says the resource *is* this function but I don't think he means
that, since a database is not a function (last I checked). In the same
paragraph he also says "the semantics of the mapping distinguishes one
resource from another" which is either redundant or nonsensical. I
think the bit about functions is just some formalism he introduced to
try to impress someone, or at best is an incomplete model. (As you
know, a model of a Boeing 747 is not itself a Boeing 747.) When I read
this paper I ignore all of the quasi-formalism.

But you're right, this does seem to conflict with RFC 2396 (1998). I
would explain it by saying that "resource" in the ICSE paper is being
defined for the purposes of REST, so you can think of it as shorthand
for "REST resource." This would allow for the existence of other kinds
of resources. A similar practice is followed in RFC 2616 which also
has a local definition of "resource" at odds with 2396 - not
surprisingly since REST laid the foundations for RFC 2616.

> But you do know that SOMETHING is active, or you wouldn't have got a
> response. Maybe you don't know exactly what, and maybe you don't care. But
> you can infer that some active thing *exists*, and you can tell yourself
> that the URI denotes *that* (whatever it is).

Yes, that's fine. The question is whether that something is the
resource, or some other entity such as a server or The Web.

I'm not going to reject this view, as it seems beneficial in ways, and
one could do worse. I just think that while it's natural in some
cases, it's not very natural in others, in particular the document
case. In writing folk metadata (e.g. DC, FOAF) one talks about the
image or recording or music or document, and uses an appropriate URI
to mean *that*; the URI is not meant to name an agent, and often is
not even meant to name its presence on the web ... you say the
document is about potatoes, not that it will send you information
about potatoes over the Internet.

Maybe it's that the URI is used in two ways - as a hypertext link or
as a request-URI, it names the sort of agent-resource you're talking
about, but in metadata of this sort, it's being taking to name a
document, which is OK since RDF doesn't insist on models that agree
with HTTP. I'd be OK with that, but others might complain this goes
against web architecture.

>> (If I do a POST and the robot arm moves, did the
>> resource move it, or did the server?)
> I'd say in that case that the arm could be the resource, conceptually. Of
> course, the arm there includes a great deal of electronics and software, but
> then robot arms do.

Ok, maybe not a good example, since it's not an "information
resource". I'll set this aside for now.

> It is always OK to choose some 'larger' system to be the resource, note.

ok, but not if that larger system is responsible for more than one
URI.  The view that seems natural to me is that one agent (a computer
running Apache) is responsible for many documents (the files in its
web directory). To make the agents-all-the-way down model fit, you
have to be willing to say that the server has been virtually chopped
up into one little serverlet per document, like the sorcerer's
apprentice's broom. I tried doing this on this discussion list a long
time ago, and I think Stuart was right to correct me.

Hmm, another solution is to have lots of synonymous URIs for a single
resource that is a server, and then the server dispatches the correct
content based on the request-URI? No, that would be an awful
perversion. It would prevent you from using the URIs differentially in

>> Certainly this is a big difference in the way you and Tim talk about
>> the web, whether consequential or not.
> Hmm, seems to me I learned it from Tim, but maybe my exposure has been
> limited.

Look at  http://esw.w3.org/topic/AwwswGenericResource ... as far as I
can see, any agency in Tim's IR model belongs to the server, not to
the resource. Something that can process a POST request *cannot* be
"essentially information". So the POST would have to be handled by the
server, not the resource. The resource is passive.

I guess I'm looking for ontological pluralism here; I think sometimes
you want these URIs to name agents, and sometimes you want them to
name documents, and we should welcome other models in the interest of
encouraging metadata generation. Just not sure how to get pluralism
without reverting to the rocks-have-representations position.

Thanks for joining us, Pat! Always a pleasure to have you.

Received on Wednesday, 10 June 2009 16:22:42 UTC

This archive was generated by hypermail 2.3.1 : Tuesday, 6 January 2015 20:21:07 UTC