[Mulgara-dev] System Model and Canonical Aliases

[Paul Gearon]

On Wed, Sep 9, 2009 at 10:40 AM, Alex Hall <alexhall at revelytix.com> wrote:
> Looking in the StringPoolSession code, I see that if you use a URI that
> is the database URI and include a "graph" query parameter, then that
> graph parameter is decoded, unwrapped and used as the "real" URI that
> gets stored.  Since the wrapped URI is not subject to canonicalization,
> this could be a short-term workaround for your problem.

(I don't know if this will help James, but it would nice if it does)

I introduced that last year when I allowed the system to use any URI
for a graph. It allows a client (or a Mulgara server doing federated
queries) to do autoconnections to other servers now that URIs don't
necessarily identify the server for you.

If I could throw it all away and start from scratch, I'd create a new
interactive command called "connect" exactly like mysql uses (I've
been thinking of this anyway, but it would be nice if it were the ONLY
option). As for federated queries, the location should be specified in
the query. There is a proposal for this (it includes the location of
the SPARQL endpoint to federate to), though it is only a
"time-permitting" feature:
  http://www.w3.org/2009/sparql/wiki/Feature:BasicFederatedQuery
  http://www.w3.org/TR/2009/WD-sparql-features-20090702/#Basic_federated_query

So ultimately, we have, or will have, alternatives for the legacy
rmi://hostname/servername#modelname structure of graph URIs, so we
ought to be able to drop them. The only issue is the legacy code.

Paul