- From: Dan Brickley <danbri@w3.org>
- Date: Mon, 18 Mar 2002 15:23:25 -0500 (EST)
- To: Pat Hayes <phayes@ai.uwf.edu>
- cc: <www-archive@w3.org>
Pat, I've been thinking about a rathole topic that I tried to avoid in RDFCore last time it cropped up. A sanity check or logician's perspective would be welcome. Issue is URI-refs that while syntatically well formed per RFC2396, don't denote anything. Much of the RDF formalisation assumes that terms denote, but we need to deploy RDF in a Web where things aren't so clean. Example: xyz://foo.bar/ http://example.com/ urn:bogus-ns:etc These look and smell like URIrefs, and we can probably call them such, but (on some readings of the web-arch tealeaves) don't name anything. Two thoughts: (i) our lives might be easier if we could get web architectural blessing for the simplifying assumption that all Web names (URI-refs) denote something (perhaps defaulting to a /dev/null object). But if we can't asume this... (ii) a statement composed of a term that doesn't refer (assume some bogus URI scheme which will be bogus in all interpretations) is never going to be true. a graph containing such a statement isn't a healthy looking truth candidate either. I fear this means that bad/wrong/typo'd data, even if reified, will infect the truth-prospects of any RDF document that includes a reification of it (since reification doesn't currently stringify names/uris). eg <rdf:Statement> <rdf:subject rdf:resource="xyz://foo.bar/"/> </rdf:Statement> if xyz://foo.bar/ is a 'bad URI', one can't even reify statements that use it without getting into trouble. Contrast with, for example, <rdf:Statement> <rdf2:subjectTerm rdf2:stringifiedURI="xyz://foo.bar/"/> </rdf:Statement> ...which has some hope of being true, even if xyz://foo.bar/ is non-denoting. A rushed scribble cos a saw you were online. Am I making any sense? Copying www-archive for future reference, Dan
Received on Monday, 18 March 2002 15:23:26 UTC