Re: re-opening the DESCRIBE issue

On Wed, 2006-03-29 at 22:36 -0800, Pat Hayes wrote:
> >and concluded that this issue merits considerable further work.
> >We talked about the possibility of postponing the issue
> >and marking the DESCRIBE syntax "reserved for future use"
> >and he was supportive of that.
> Phew. I'm for that. Let me formally propose that we postpone this 
> issue, on the grounds
> (1) there are no currently available techniques for deciding how to 
> give a rational response to a DESCRIBE request, in general;
> (2) Although we could follow Graham's idea, there are in fact no 
> currently proposals for any such URI-named schedules for determining 
> such responses, and nobody (?) has given any signs of suggesting any;

Indeed, point (2) is an observation that came up in review
of our request for CR: the design is interesting, but not mature.
I would want to have in hand two or three actual URIs being used
in this role, and actual interoperability experience, before
I'd be comfortable putting the question to add that to the spec.

> (3) The motivation for DESCRIBE is the case where a query user does 
> not know what information might be available. However, this issue can 
> be approached within the current SPARQL framework by query service 
> providers including meta-data in the RDF stores to support less 
> pointed queries. I suggest that the possibility of techniques such as 
> this serving the primary purpose is more likely, since information 
> providers are more motivated than SPARQL engine implementors to 
> provide for useful query user experience.

I find this rationale appealing. I also agree with Kendall
that there's a connection to serviceDescription, which is
also postponed.

But to be complete, the proposal needs more details about how
the text of the spec changes.

In particular:

 -- does the SPARQL grammar change?
Andy suggests:
"I would like to see the keyword reserved and the current syntax
maintained as 
it is already out there."

I find that appealing/expedient, since, unless we re-open
the syntaxExtensionProtocol issue, if we take DESCRIBE out
of the grammar completely, all conforming services must not give
results for queries that use it.

 -- what text do we put in the spec to give appropriate guidance

Hmm... maybe take the DESCRIBE examples out, mark the syntax
"reserved for future use", and add an appendix near the Change Log
that says something like...

  DESCRIBE is reserved for future use; for some discussion
  of possible designs, see _section xyz of previous drafts_
  and the postponed _DESCRIBE issue_.

 -- do any of our approved tests change?

hmm... can I do a SPARQL query for approved tests with DESCRIBE
in the input query? This is a serious question.

Here's what I get using find/grep; I dunno how many of these are

connolly@dirk:~/w3ccvs/WWW/2001/sw/DataAccess$ find tests -type f -print
| xargs grep DESCRIBE
<u> ?u WHERE { <x> <q> ?u . }
tests/data/extracted-examples/query-10.4.2.rq:DESCRIBE ?x
tests/data/extracted-examples/query-10.4.3.rq:DESCRIBE ?x WHERE { ?x
ent:employeeId "1234" }
tests/data/extracted-examples/query-10.4.2-q1.rq:DESCRIBE ?x
tests/data/extracted-examples/query-10.4.2-q2.rq:DESCRIBE ?x ?y

> >The issue is hereby re-opened.
> >

Dan Connolly, W3C
D3C2 887B 0F92 6005 C541  0875 0F91 96DE 6E52 C29E

Received on Thursday, 30 March 2006 16:14:56 UTC