- From: Chimezie Ogbuji <chimezie@gmail.com>
- Date: Wed, 22 Aug 2007 17:26:55 -0400
- To: www-tag@w3.org
Pat, thanks for calling out this overlap with the ongoing discussion. One point of clarification is *all* the discussion in that particular thread has to do with RDF URIs specifically. The difference may be a little more than subtle, considering ISSUE-58 and the URNs Namesapces and Registries finding address URIs in general and not URIs in RDF specifically . The concerns (as I see it) have more to do with automated machine interaction and not human interation. It dips into both the URNs Namespaces and Registries finding as well as with ISSUE-58. The key point which links the two is the suggestion that (even when used within RDF graphs) HTTP URIs should be preferred generally. Below are my comments to Norm's response: 1. http: != dereference Yes, this is very clearly stated in the URNs Namespaces and Registries finding. However, we cannot have our cake and eat it too. Specifically, there seems to be a bit of a conflict with: 1) The prominent suggestion that HTTP URIs should be used pervasively, 2) The equally prominent suggestion that it is a good idea to provide representations for these URIs, 3) ISSUE-58 and 4) The 'qualifier' above that the use of the HTTP scheme does not mandate dereference. As Noah suggests, the qualifier can be interpreted by the author as a suggestion that providing representations is not necessary. In addition, it can be interpreted by the consumer as a suggestion to not bother attempting to (arbitrarily) dereference these URIs (many don't). If a little bit of ambiguity was the only price being paid, it wouldn't be much of a concern, however, ISSUE-58 clearly identifies the cost as much more than ambiguity alone. I believe the appropriate recommendation, guideline, etc.. would be one which includes a clearly articulated set of scenarios which demonstrate when 'arbitrary' HTTP dereference (though not mandated) is useful for automatons/agents and when it might not be so useful (XML namespaces, for example). 2. The dereference problem is scheme independent The second part of this particular point assumes there will *inevitabely* be a need to dereference these (insert your favorite other scheme here) URIs. This is not always true, especially when the URIs in question are RDF URIs. RDF URIs and their use have a model-theoretic mechanism for making claims about the world. In most cases, these claims (very mathematical in nature) are meant to be much more authoritative than what representation you might get from dereferencing the URIs themselves especially when the claims are subject to much more fine-grained constraints through the use of a formal (OWL) ontology. The only caveat might be where the representations retrieved describe the very OWL ontologies which capture these constraints. In this case, and with ISSUE-58, as a backdrop it would seem prudent to review current practice in this regard or (perhaps) suggest some best practices. However, again this is specific to RDF and ISSUE-58 applies to all usage of URIs. RDF URIs have a different usage pattern than the typical Web scenario and the literature should consider this divergence. In addition, the dereference problem is not entirely scheme independent. Actually, it *only* applies to those URI schemes which are (formally) associated with a transport protocol (ironically, the same scheme(s) which are the subject of suggestion for their pervasive use). -- Chimezie
Received on Wednesday, 22 August 2007 21:27:03 UTC