Re: Diagram of it all

On Fri, Mar 4, 2011 at 9:54 PM, Pat Hayes <phayes@ihmc.us> wrote:
>
> On Mar 4, 2011, at 7:52 PM, Jonathan Rees wrote:
>
>> Did you spot the contradiction, in one of your diagrams, to my axioms?
>> In my little world, if a resource has only one representation, then
>> much of what you say about the representation has to also be true of
>> the resource - for example, whether its content contains the letter
>> 'x'.
>
> Hmmm. I confess to not being entirely uptodate with your axioms, but this sounds like it would rule out a lot more than just RDF graphs.

Yes, I'm very sorry, I sent that message in error. It was wrong. I
wasted a bit of your time and lost credibility. Mea culpa.

A simple IR, one that has only one 'representation', would be subject
to what I said, but a graph is not one of those since it has many of
'representations'.

Your suggestion, that a property shared by all serializations of a
graph is their being a serialization of a graph, is technically
correct, but what I'm looking for is something *informative* - that
is, something that will distinguish serializations of graph A from
serializations of graph B.  That itself is such a property, and I list
this in the latest version of the document, so *that* would be the
contradiction between the axioms and the idea that a g-snap is an IR
with its serializations as 'readings' (since a g-snap is not a
serialization of any graph, not even itself).

That leaves the question of whether a changing g-box could have its
g-snaps' serializations as 'readings'.  If they were diverse enough,
and really had nothing at all in common, then yes, it could be
(according to the axioms) an IR with no metadata properties. But as
you and Nathan say these diverse serializations *might* have something
in common, such as being serializations (not any particular one) or
having lengths (not any particular one), and that would be enough to
raise the question.

A more serious problem, I think, is ruling out spam inside comments.
When you use a dereferenceabe URI to name a g-box, you would probably
prefer *not* to authorize arbitrary serializations since some of those
will contain comments you don't agree with. Put another way, you could
have distinct IRs related to a single g-box, differing only in the
comments that occur in serializations.

So I still fear there are only three options
1. Give up on this idea (P(IR) iff {P(SIR) for all SIR specializing
IR)}  => maybe give up on httprange-14 rule
2. Persuade RDF WG that g-boxes and/or g-snaps are not good IR candidates
3. Live with a wart

Jonathan

Received on Saturday, 5 March 2011 14:46:21 UTC