Re: superimposing the Fielding and TBL architectures

On Wed, Sep 28, 2011 at 8:19 AM, Henry S. Thompson <ht@inf.ed.ac.uk> wrote:
> Jonathan Rees writes:
>
>> . . .
>> In the Fielding architecture the resource is unconstrained. I can give
>> you a bunch of different resources, and then when you challenge me to
>> prove that there is a resource with those Fielding-representations, I
>> can cook up any story I like, post hoc, and you'd have no way to prove
>> me wrong.
>
> Presumably you mean "give you a bunch of different _representations_".
>
> I take it this lack of constraint stems from the Fielding-importance
> of the role of the URI owner -- it's up to them (without constraint)
> to say what a URI identifies, and what is or is not a representation
> of whatever that is.  I think I know where the first as-it-were grant
> of discretion comes from, but what about the second?  Can you point me
> to the relevant bit of 3986?

There is no such grant in 3986 or 2616, only the implicit grant given
by silence (that has been exploited by the Toucanists, no offense
intended). But the HTTPbis editors certainly have in mind that there
ought to be such a grant, as they have written:

   Note that answers to the
   questions of what can be represented, what representations are
   adequate, and what might be a useful description are outside the
   scope of HTTP and thus entirely determined by the URI owner(s).

http://tools.ietf.org/html/draft-ietf-httpbis-p2-semantics-16#section-8.3.4

I objected strenuously to this on the HTTP WG list, and was summarily
shot down. Maybe I didn't marshall the best arguments - like perhaps I
didn't point out that the "thus" is a non sequitur - and it's worth
another try.

Regardless, I think that practically speaking, unless there is some
documented constraint on what constitutes a representation, the lack
of such documentation will be taken as complete freedom. And it is too
late to argue that something in particular was meant all along - the
cat's out of the bag and we're in no better position to interpret the
RFCs than anyone else. The best we can do IMO is to attempt to get a
new consensus (amended httpRange-14) from the RDF and webarch-aware
communities. And I'm doubtful we can get even that.

>> In Tim's architecture the resource is determined, modulo usually we
>> probably don't care about, by what the correct retrieval results would
>> be. Once those results are determined, there's no choice as to what
>> the resource is. Contrariwise, if the server side commits to what the
>> resource is, we can hold them to it by checking any
>> TBL-representations that they deliver.
>
> Sorry to come late to this party, but maybe I'll count as an
> intelligent first-time reader of the Final Report.  Does the above
> para. amount to saying that the way the owner of a 200-responding URI
> establishes what it TBL-identifies is by his/her choice of what is
> served with a 200 in response to GET requests for that URI?

Almost. I would say "potentially served" instead of "served", and I would
say that the potential responses determine the referent well enough to
answer questions like title, author, subject, word containment (when
those apply), but not well enough to answer questions about
circumstance dependencies (resources may have the same potential
responses now, or at one requesting IP address, but differ in what
responses they have tomorrow, or from another IP address) or, should
we decide to go this route, differences in the effects of POSTs (e.g.
you could have the same potential responses in all circumstances, but
a POST to one URI turns on a light, while a POST to another URI starts
a motor; is that to be consistent with the two URIs naming the same
thing?  or consistent with them not naming the same thing?).

Another example is two URIs that both "identify" Moby Dick the novel
(understood in some particular way). The potential responses for both
URIs are the same, even if the actual responses from the two URIs
differ in typeface or other details (that don't matter according to
aforementioned particular way).

> So, for example, serving application/rdf+xml or application/n3 as 200
> in response to a GET establishes that the URI involved identifies an
> RDF graph?

If the wa:Representations are all RDF graphs, then (in my proposal)
the URI would name an RDF graph. I would say they are not, since every
wa:Representation has a length and media type, and no RDF graph has a
length or media type. The wa:Representations are serializations of RDF
graphs and if they're delivered for GET U then (in my proposal) U
would name a wa:GenericResource that has the properties common to
those wa:Representations (and being an RDF graph is not such a
property).

This is being discussed - perhaps resolved already - in the RDF WG,
and my guess is that their answer disagrees with mine, putting another
nail in the coffin of httpRange-14, IMO.

> So to sum up
>
>  * wrt the left-hand (TBL) side of your diagram, the owner's free
>    will is expressible only via his/her control of the server,

Pretty much, yes.

>  * whereas on the right-hand (3986) side, the owner's free will is
>    expressible _both_ wrt his/her out-of-band statements about what a
>    URI identifies _and_ his/her control of the server.

Yes

>> Nothing much new here, pretty much what Pat has said in different
>> words (although I put less stock in "access" and more in social
>> agreement over what would constitute correct access were it to occur).
>
> So this is meant to decouple the 'has TBL:representation' link from
> the vagaries of connectivity, and/or time variation?  Doesn't that
> render it liable to Wittgenstein's counting paradox?  That is, there
> is no guarantee that whatever I may induce about a wa:GenericResource
> from some retrieval history wrt some URI will stand up to the results
> of the next retrieval.

This is true of any prediction you can make about the world. It is
true or not true that your house will still be standing tomorrow. You
can't know or predict the future with certainty. Yet you're still
willing to make statements like "I'll have supper at home tomorrow",
"the bridge will carry the load", "Saturn will rise at 3:20am", "my
student will pass the test", etc.) with confidence. You can make
pretty good bets, which is all science (or character assessment, in
Wittgenstein's case) can ever give you. The question of whether a
particular wa:GenericResource (such as Moby Dick) is served from a
particular URI - or rather whether it's *correct* for it to be served
from the URI - has the same epistemological status. If we're confident
about the latter, then we should be willing to use the URI as a name
for Moby Dick (under this architecture).

If you're not willing - don't use http: URIs to name things! Or reject
the architecture, which would be OK with me.

Jonathan

> ht
> --
>       Henry S. Thompson, School of Informatics, University of Edinburgh
>      10 Crichton Street, Edinburgh EH8 9AB, SCOTLAND -- (44) 131 650-4440
>                Fax: (44) 131 651-1426, e-mail: ht@inf.ed.ac.uk
>                       URL: http://www.ltg.ed.ac.uk/~ht/
>  [mail from me _always_ has a .sig like this -- mail without it is forged spam]
>

Received on Wednesday, 28 September 2011 13:50:38 UTC