Bruce D'Arcus wrote:
> On May 20, 2007, at 5:02 PM, Dan Brickley wrote:
>
>>> I was certainly presuming with the idea that both literal and blank
>>> nodes would not be allowed. That might indeed be "nice but not
>>> practical"; not sure.
>>
>> The latter is not permitted by the RDF specs, since the bnode
>> variation is implied by the URI one. You're simply removing a piece of
>> information from the RDF graph, and RDF allows information to be
>> removed without problem.
>
> Not following Dan. What is "not allowed"?
>
> And what's the upshot? That defining a property as an IFP would not
> preclude bnodes?
I mean that in RDF, parties defining schemas don't get to say that
values of their properties must always use URIs. Yes, so defining some
property as IFP doesn't require occurances of that property to always be
provided with URIs.
This is because
<Document>
<dc:identifier rdf:resource="uuid:123412341234" rdfs:label="xyz" />
<dc:title>A document</dc:title>
</Document>
implies
<Document>
<dc:identifier rdfs:label="xyz" />
<dc:title>A document</dc:title>
</Document>
...since RDF is designed for partial information exchange, and always
allows bits of information to be dropped.
Of course one might define document formats that did express such extra
constraints, but they don't come for free with RDF. And RDF apps would
cheerfully ignore them when the encoded triples get mixed into general
purpose RDF environments.
Does that make any more sense?
Dan
|