Disallow BNode identifier as Graph Name

[gkellogg]

We discussed if it is appropriate for a Graph name to be a BNode identifier on yesterday's teleconference, and on #rdf-wg IRC. There are a couple of concerns: RDF Concepts defines a Named Graph as being an {IRI, Graph} pair, clearly this does not include a BNode Identifier. However, JSON-LD strays in some other areas, where it makes sense, for example using a BNode Identifier as a predicate. The later is for historical purposes, as RDF/XML can't express this, but the choice to limit Graph names to IRIs is not historical.

One of the issues is that the scope of BNodes is problematic, and it can't really be said that a BNode identifier is a name. Furthermore, as a BNode Identifier only has meaning in the scope of a document, you can't reference the same Named Graph from more than one document.

PROPOSAL: Limit the value of a Graph Name to be an IRI.

[lanthaler]

RESOLVED: This group re-affirms that the common practice when naming a graph is to use either an IRI or a blank node identifier. The JSON-LD specification remains unchanged. When expressing graphs in Linked Data, the graph name SHOULD be an IRI.

[lanthaler]

RESOLVED: JSON-LD will continue to support blank node identifiers for properties and graph names. When converting data to RDF 1.1, the specification will not introduce any special checks to handle these specific cases. It is up to the implementations to figure out how to convert this data to something conformant to RDF 1.1.

[gkellogg]

To clarify, the spec will say that the blank nodes as predicates or graph names lead to undefined behavior when transforming to RDF. Conforming implementations can drop statements with bnode predicate triples, or graphs with bnode names.

[lanthaler]

We have been discussing bnode ids as graph names for quite some time. The last discussion we had was in last week's telecon. The position of the CG didn't change and in the RDF WG consensus was found (some time ago) around allowing JSON-LD to deviate in very specific ways from RDF - which includes bnode ids as graph names.

Unless I hear objections, I will therefore close this issue in 24 hours.