[Mulgara-general] Model-URI/URL Use-cases and Requirements and Proposal

David Wood dwood at softwarememetics.com
Wed Oct 10 14:21:17 UTC 2007


On 20 Sep2007, at 22:41, Paul Gearon wrote:
> As far as I can tell we just need 2 things:
>
> 1) A way to reliably map from a URL to a server on a host.
> 2) A way to reliably map from a URL to a node ID on a host.
>
> The first point is required to allow the current paradigm, where a  
> client library automatically makes connections to the server.  I  
> don't like this at all (particularly with federated queries) but I  
> don't want to break what we already have.  Discussing it with  
> Andrae, we're expecting an initial attempt to use URLs that  
> identify the host and service.  Ultimately, we expect to have  
> registry support.
>
> I fully expect some scenarios will result in the possibility of  
> connecting to the wrong graph, due to an incorrect match while  
> falling back through the different identification mechanisms.   
> While undesirable, I believe it is mitigated because this mechanism  
> will be deprecated.  Also, if we introduce security again, then  
> security domains will prevent unauthorized data from appearing.   
> (There's also an argument to be made around the universality of RDF  
> assertions, and the open world assumption - and while correct it's  
> not really pragmatic).  That all said, it should work 99% of the time.
>
> The second point will be done by mapping to a URI, which by  
> definition is unique, and has a unique node ID.  The form of this  
> URI is what is keeping Andrae up at night.  :-)
>
> Neither point proscribes how to treat the path of the URLs, nor  
> whether or not fragments exist.  So I'm OK with dropping fragments.

I'd like to be clear that in:

   rmi://pneuma.netymon.com/rdfDatabase#addressBook

#addressBook is a URL fragment.  Dropping that would be bad :)

Regards,
Dave






More information about the Mulgara-general mailing list