Re: “RDF systems” and empty graphs (was: Re: Making progress on graphs)

On Tue, 2012-05-15 at 14:45 +0100, Richard Cyganiak wrote:
> On 15 May 2012, at 14:11, Sandro Hawke wrote:
> > Is there something wrong with an RDF system that doesn't implement blank
> > nodes?  
> 
> Well, that is a topic better left for another day.
> 
> > Or wont store triples which use literals with datatypes?
> 
> (All literals have datatypes in RDF 1.1.)
> 
> I wouldn't find it very useful. It wouldn't be able to store a good number of interesting RDF graphs. It wouldn't interoperate well with other RDF systems. Users may well complain and call it broken.
> 
> > If so, how would you characterize what is wrong with it, in terms of
> > conformance?
> 
> Artefacts conform to specifications. In the absence of a specification that claims to define “RDF systems” we cannot talk about RDF systems in terms of conformance.
> 
> We could point out that it cannot store the RDF graphs serialized in many conforming RDF/XML documents, and that it doesn't pass the RDF test cases.
> 
> RDF Concepts, the spec that defines what blank nodes are, doesn't define conformance criteria for “systems” but only for other specs. RDF/XML, SPARQL, RDF Semantics and so on all are specs that conform to RDF Concepts (modulo 2004 vs. 1.1 issues).
> 
> The closest we have to a spec that defines an “RDF system” is SPARQL Update, which normatively defines “graph stores”. The definition requires implementation of blank nodes and literals, so our RDF system could not be a conforming graph store.
> 
> Returning to the topic: The definition of “graph store” does not require, but allows, implementation of empty graphs. So, a system that implements empty graphs is a conforming graph store (assuming it meets the other criteria in SPARQL Update). A system that doesn't implement them is a conforming graph store too (assuming it meets the other criteria). This is of course not so good for interoperability between graph stores, but it is a compromise the SPARQL WG made in order to accommodate different kinds of implementations.

Given this framework (which seems quite reasonable), it seems clear to
me that SPARQL Update 1.1, as a spec, would not conform to RDF, if RDF
Concepts defined Dataset as you propose. 

I mean: wouldn't RDF Concepts be setting the expectation that specs
using "RDF datasets" would mandate systems support all the features of
datasets?

For clarity on this conformance perspective, I don't think the RDF/XML
spec will conform to RDF Concepts going forward.   I assume we're
granting it an exemption for historical reasons, but I think we would be
quite resistant to any new RDF syntax had those same limitations.

   -- Sandro

Received on Tuesday, 15 May 2012 14:53:54 UTC