Feedback: #rdfms-identity-anon-resources

My proposal has gratifyingly generated some responses.

(I would point out:  this is isn't a clear-cut issue for me -- I am seeing 
some potent arguments for explicit support for some form of anonymous 
resources in the model/abstract syntax.  But, I feel I must continue to 
play devil's advocate here to push down to the minimum requirements that 
must be supported/specified.  Otherwise, I fear that we could end up with a 
whole load of baggage that really doesn't belong in the RDF core.)

I include below a summary of the responses I have noted to date.

In summary, the arguments _for_ anonymous resources in the model/abstract 
syntax are:
(a) difficulty of creating globally unique genids
(b) capturing semantics needed to express queries

Are there more?

...

To my mind, the most potent reason raised in response to my proposal for 
supporting anonymous resources (or similar) in the model is the difficulty 
and complexity of reliably generating a unique identifier, and the 
implications this has when combining RDF graphs.  This represents a 
difficulty in dealing with M&S as currently defined and used.

The other case that has been strongly argued is the requirement for 
variables in a query.  While I can see the utility of what is being 
proposed, I have two problems with this:

(a) I don't see anything in M&S that sanctions the idea of query 
variables:  as such, this proposal seems to be a significant extension 
rather than clarification of the current M&S that we are chartered to 
do.  (I'm not claiming this means that the query issue should not be 
addressed, but that we must be clear why and how far we might pursue it.)

(b) I have a concern that anonymous resources are not exactly the right 
hook for attaching the query variable problem.  A query requires that one 
can distinguish between known and unknown values in the query expression, 
and anonymous resources can provide that distinction.  But when we take it 
to a different context, it seems that a very different interpretation of 
the anonymity may be desired.

I find myself wondering if support for some kind of scoped identifier 
wouldn't address both of the goals raised.  This would be in contrast to 
URIs which are, by design, globally scoped and bound to specific 
resources.  I don't know if the semantics could play out simply yet 
sufficiently flexible to handle these goals.  I think this, too, would be a 
clear extension to M&S.

#g
--

Summary of responses to anonymous resource proposal at 
http://lists.w3.org/Archives/Public/w3c-rdfcore-wg/2001Jul/0082.html.

Aaron:  [11-Jul-2001 12:47:33-0500]
(a) Is OK that triples can contain anonymous entries, but not with a new 
anonymous resource type.
(b) thinks genID algorithm should be specified.

Sergey:  [11-Jul-2001 14:26:07-0800]
Agrees with Aaron about genID algorithm.

Jos:  [12-Jul-2001 12:17:24+0100]
Thinks that anaonymous resources have NO name;  cites empty circles.

Ron: [11-Jul-2001 20:33:47-0700]
Agrees with Aaron about (a), but "predefined" genIDs would be nice, not 
essential:

[[[
I'd be happy with a non-normative convention suggesting how
the IDs should be generated. I don't want it to be a MUST
in the clarified spec.
]]]

DanC: [12-Jul-2001 02:02:51-0500]
Strongly feels anonymous resources must be represented in the model.

Jan: [12-Jul-2001 10:36:54+0100]
Anon resources are needed in the model.

Jos: [12-Jul-2001 12:28:51+0100]
Agrees with DanC.

DanBri: [12-Jul-2001 07:29:33-0400]
Warning about possible problems of generating same URI for same text in 
different context.  Gives an example with a relative URI.

StefanK [non-WG comment, posted to RDF-IG, 12-Jul-2001 09:38:28+0200]
Thinks that forcing a system to invent names will cause problems.

[[[
1. An important application of RDF is the join of RDF graphs:
By now named resources (with identical URI) are glued together,
but anonymous are not. Please don't change this.
]]]

and

[[[
2.1 I don't think there exists an algorithm that fulfills
the requirement of your 'essential property of a unique URI'
you mentioned above.
]]]

Jan: [12 Jul 2001 14:19:28 +0100]
Cites an earlier analysis, claiming it shows why anonymous resources are 
needed in the model.  http://ioctl.org/rdf/discuss/anonymous#Complications




------------
Graham Klyne
(GK@ACM.ORG)