Re: Using bnode identifiers for predicates, graph names

On 06/02/13 10:11, William Waites wrote:
> On Wed, 06 Feb 2013 09:09:02 +0000, Andy Seaborne <andy.seaborne@epimorphics.com> said:
>
>      andy> An identifer for a graph-not-labeled by the data publisher
>      andy> can still be an IRI.
>
> To reiterate, this is true of all things that are not
> named/labelled. So why bother having blank nodes at all?  What makes
> graphs so special that we can't talk about them without giving them
> IRI-labels?

In theory, nothing, although there isn't one agree set of semantics of 
datasets let alone of bnodes outside and labeling graphs.  bnodes are 
not names.

Presumably, bnode labelled graphs would be partial so that does not for 
for JSON-LD anyway.


In practice, this WG has to finish.

RDF datasets have been around since approx 2006 using IRIs.  TriG has 
been around for longer.  It was a compromise then,it is a compromise now.

(3Store had bnodes for graphs although they were different in usage to 
what is proposed here)

This WG has talked about it since the start.  There is no agreement.

Using IRIs is tested technology and a conservative choice.

As with any WG, it's a compromise.  And now it's time to finish RDF 1.1.

If there are to be structural changes, it's not 1.1.

The reverse - put in all the syntax bell-and-whistles in the hope/belief 
something will come along - leads to incompatibility.
It is then very hard to converge on an agreed standard later because 
significant communities have incompatible beyond-spec conventions and needs.

Interoperation of data matters.  Migration and change on the web of data 
is not like software upgrade.  Define a solid interoperable core.

	Andy

PS Jena basically does not care.  While the formal API has the RDF 
rules, internally it's pretty neutral.  It's really reader and writers 
that mediate these rules as data goes in and out.

Received on Wednesday, 6 February 2013 10:32:36 UTC