Re: A parable about RFC 3986.

On Thu, Jan 26, 2012 at 11:19 AM, David Booth <david@dbooth.org> wrote:
> I would paraphrase these architectures as:
>
>>   A. The Ian Davis architecture - a GET/200 URI can identify anything.
>
>  "By fiat, a GET/200 response means that the URI identifies that
>  web resource, UNLESS the response content is an RDF/XML document,
>  in which case the URI identifies whatever the RDF/XML document
>  says it identifies."

This is not what I meant - I was referring to the "take at face value"
architecture which should more properly be attributed to Harry Halpin
and Manu Sporny - Ian's is similar and I wasn't referring to the
conneg details. So replace 'Ian Davis' with 'Harry Haplin' in what I
wrote before.

>>   B. The Roy Fielding architecture - a GET/200 URI identifies a "fiat"
>> (REST) resource.
>
>  "By fiat, a GET/200 response means that the URI identifies that
>  web resource."

I don't know what a 'web resource' is and Fielding doesn't use that
term. He doesn't in fact use any term other than 'resource'. I
sometimes use 'REST resource' although it's not obvious to me that RFC
2616 and 3986 are talking about the same abstraction as the subject of
the REST writing.

When I say 'Fielding architecture' here I mean what 2616 and 3986, and
especially HTTPbis, describe (to the extent they describe anything).
My conjecture is that my idea of 'fiat resource' coincides with what
2616 and 3986 would imply for resources permitting retrieval that are
identified by http: URIs. But the details are subtle.

>>   C. The TimBL architecture - a GET/200 URI identifies a "generic" resource.
>
>  "By fiat, a GET/200 response means that the URI identifies that
>  web resource.  And don't mint a URI that ambiguously identifies
>  both a web resource and a toucan, because toucans are not web
>  resources!"

I don't know where you get this; TimBL's view is entirely about what
properties these things have, not what their type is; it's about
titles and authors and subjects and licenses. If someone wants to say
that a toucan has a title, author, etc. that's not reasonable but it's
their prerogative (in my view). Maybe the right moniker is the JAR
architecture, but I think Tim and I are mostly on the same wavelength
so I prefer to attribute it to him.

And we've had the ambiguity/identity discussion before on the list
without resolution, so I don't know why you bring it up, without
bringing new information to the table.

I said that 'generic resources' are also 'fiat resources' but now I'm
not so sure - if TimBL thinks Moby Dick is a generic resource, and
he's right, then not all generic resources are fiat, since Moby Dick
is not fiat. There may be a tighter class 'fiat generic resource' that
is of interest, making yet another architecture.

Enumerating these classes has nothing to do with disjointness, it is
about keeping track of what axioms you want to consider. There may be
consistent models in which there exist things not in any given class
(fiat resource etc.), but there may also be consistent models in which
there are not.

I believe I have a moderately rigorous treatment that works and will
send it under separate cover after I can get it independently checked.
The quarter-baked 'parable' email was just to get some feedback, and I
suppose I deserved what I got.

Jonathan

Received on Friday, 27 January 2012 20:28:00 UTC