« Linked Data & Web Search Engines | Main | Document metadata using DC-HTML and using RDFa »

April 13, 2010

A small GRDDL (XML, really) gotcha

I've written previously here about DCMI's use of an HTML meta data profile for document metadata, and the use of a GRDDL profile transformation to extract RDF triples from an XHTML document. DCMI has had made use of an HTML profile for many years, but providing a "GRDDL-enabled" version is a more recent development - and it is one which I admit I was quietly quite pleased to see put in place, as I felt it illustrated rather neatly how DCMI was trying to implement some of the "follow your nose" principles of Web Architecture.

A little while ago, I noticed that the Web-based tools which I usually use to test GRDDL processing (the W3C GRDDL service and the librdf parser demonstrator) were generating errors when I tried to process documents which reference the profile. I've posted a more detailed account of my investigations to the dc-architecture Jiscmail list, and I won't repeat them all here, but in short it comes down to the use of the entity references (  and ©) in the profile document, which itself is subject to a GRDDL transformation to extract the pointer to the profile transformation.

The problem arises because XHTML defines those entity references in the XHTML DTD, i.e. externally to the document itself, and a non-validating XML processor is not required to read that DTD when parsing the document, with the consequence that it fails to resolve the references - and there's no guarantee that a GRDDL processor will employ a validating parser. There's a more extended discussion of these issues in a post by Lachlan Hunt from 2005 which concludes:

Character entity references can be used in HTML and in XML; but for XML, other than the 5 predefined entities, need to be defined in a DTD (such as with XHTML and MathML). The 5 predefined entities in XML are: &, <, >, " and '. Of these, you should note that ' is not defined in HTML. The use of other entities in XML requires a validating parser, which makes them inherently unsafe for use on the web. It is recommended that you stick with the 5 predefined entity references and numeric character references, or use a Unicode encoding.

And the GRDDL specification itself cautions :

Document authors, particularly XHTML document authors, who wish their documents to be unambiguous when used with GRDDL should avoid dependencies on an external DTD subset; specifically:

  • Explicitly include the XHTML namespace declaration in an XHTML document, or an appropriate namespace in an XML document.
  • Avoid use of entity references, except those listed in section 4.6 of the XML specification.
  • And, more generally, follow the rules listed for the standalone document validity constraint.

A note will be added to the DC-HTML profile document to emphasise this point (and the offending references removed).

I guess I was surprised that no-one else had reported the error, particularly as it potentially affects the processing of all instance documents. The fact that they hadn't does rather lends weight to the suspicion that I voiced here a few weeks ago that it may well be that few implementers are actually making use of the DC-HTML GRDDL profile transformation.


TrackBack URL for this entry:

Listed below are links to weblogs that reference A small GRDDL (XML, really) gotcha:


The comments to this entry are closed.



eFoundations is powered by TypePad