RE: Comments on the new RDF Test Cases draft

> -----Original Message-----
> From: Jan Grant [mailto:Jan.Grant@bristol.ac.uk]
> Sent: Friday, May 31, 2002 12:58 PM
> To: Brian McBride
> Cc: Massimo Marchiori; phayes; www-rdf-comments
> Subject: RE: Comments on the new RDF Test Cases draft
>
>
> On Fri, 31 May 2002, Brian McBride wrote:
>
> > At 10:17 31/05/2002 +0100, Jan Grant wrote:
> > [...]
> >
> > >I can make it as wooly as you like.
> >
> > Oh dear, I wasn't trying to be woolly, I was trying to be precise :(
> >
> > [...]
> >
> >
> > >The reason for the wording is that the author of a popular parser (that
> > >is, Dave Beckett) said that he wanted some explicit words in the
> > >document to tell him exactly what he (as a parser writer) was supposed
> > >to do to see if his parser agreed with the expected output
> from the test
> > >cases. If you want to call that "conformance", fine, I suppose it fits
> > >that definition.
> >
> > Tricky to do norminatively since the specs don't define a
> processing model.
> >
> > The spec could provide guidance though.
>
> We don't need a processing model. We just say, if you parse the input
> document(s), then the result should be isomorphic to the rdf graph
> described by the ntriples result document.
Yes, but that's precisely the kind of things that lead to the
conformance problems, and that Brian (I think, Brian correct
me if I'm wrong) is trying to avoid, because they raise
the related issue that made this thread start:
"an RDF parser should do this" ==> conformance ==>
definition of RDF parser (-> process model Brian was talking
about) ==> rigorous definitions ==> syntactic/semantic
equivalence questions.

Just talking about "expected outputs" and similar mild wording,
instead, gets where we all want to go, without the conformance
(and formal definition) troublesome waves.
And, more important for RDF-Core, saves time and cycles too :)
No? ;)

>
> The preamble to the test case descriptions _does_ say that all test
> case results should be interpreted by a person (ie, that we're not into
> exhaustive conformance testing).
Blub.... you mean the test:description?
This awfully reminds me of the rdfs:comment recent thread...
What's the normative value of an rdf:description inside a test case?
If instead it's in the main text, I am not sure about the reference.

> Dave's comment was the only input I received at the time from anyone who
> is actually writing a parser that's in production use; so I complied
> with his request.
>
> Might I suggest we wait for Jeremy's document describing exactly what we
> mean by "an RDF graph" and point to that for appropriate definitions of
> equality/isomorphism/equivalence? It will (I hope) make this clear,
> accurate, and moot.
Yes, but modulo the bigger conformance point above. Once settled that, in
the non-conformancy option we might even not be formal at all
and not even lose cycles to precisely define graph isomorphism ;)

-M

Received on Friday, 31 May 2002 13:28:34 UTC