- From: Jeremy Carroll <jjc@hplb.hpl.hp.com>
- Date: Fri, 28 Feb 2003 23:52:06 +0000
- To: pat hayes <phayes@ai.uwf.edu>
- CC: Jeremy Carroll <jjc@hplb.hpl.hp.com>, w3c-rdfcore-wg@w3.org
> I took this comment as a rhetorical question meaning, "why bother even > getting into canonicalization if you have implementation variance?" and > hence suggesting a fourth option, which you did not consider: > > D. Ignore XML canonicalization and treat XML literals as strings, ie the > L2V mapping is the identity. > > Then the entire rdf:XMLliteral datatype machinery is just an elaborate > way of encoding the old 'XML bit', which I thought was the original > intent in any case. Introducing XML canonicalization seems to have been > one those neat ideas that got slipped in without too much discussion and > has turned out to be a tar-pit. I am particularly concerned that this > ugly mess is now centrally included in the very core of RDF. I would > hope that many 'cheap and cheerful' RDF engines wouldn't even want to > know about XML, still less about XML canonicalization. This really does not meet the requirements ... XML parsers really really have variability, when building RDF/XML parsers we have to work out how to deal with that. So the simple webont examples where they want a single well-defined denotation of some literal constructed with an rdf:parseType="Literal" cannot be addressed simply by saying "use the original string". In some real contexts there isn't a string to use (e.g. parsing a DOM tree). We could have put all the work in the parser, and then the semantics could just use the string - that may be your preference, but it's too late now. In practice I would expect a webont impl to work that way. However, I also believe in practice that there will be cheaper parsers for low footprint environments which don't do this. Jeremy
Received on Friday, 28 February 2003 18:52:25 UTC