- From: Dan Connolly <connolly@w3.org>
- Date: 18 Jun 2002 16:05:26 -0500
- To: www-xml-schema-comments@w3.org
I got the impression that certain design decisions in XML Schema are justified by a requirement to be able to associate types with parts of documents. But in a discussion in xmlschema-dev in Jun 2001, I wasn't able to defend this position... [[[ in W3C XML Schema validation, the result includes not just a "yes, this is valid"/"no, not valid" but also "and this part of the input matched this part of the schema" i.e. "it has this type, is associated with this annotation" etc. ... the other languages won't give you type/annotation info as a result of checking. ]]] -- http://www.w3.org/2002/02/mid/3B38FC65.2FBC4C8B@w3.org http://lists.w3.org/Archives/Public/xmlschema-dev/2001Jun/0187.html But K.Kawaguchi disagreed: http://lists.w3.org/Archives/Public/xmlschema-dev/2001Jun/0189.html I then went to study the relax NG spec to see if I could understand it for myself. I never quite learned the answer... The thread continued; K.Kawaguchi concluded with: [[ The only difference, with regard to the type assignment, between RELAX NG and W3C XML Schema is that W3C XML Schema ensures that you can always tell the type of elements/attributes when you see its startElement SAX event, whereas RELAX NG doesn't. But the way W3C XML Schema ensures this is ad hoc. As a result, many grammars (including the one written by Geoff) are prohibited even though they satisfy this property. ]] http://lists.w3.org/Archives/Public/xmlschema-dev/2001Jun/0194.html I have to say that, after trying to capture XML Schema part 1 in larch, and looking[1] at the relaxNG formalization[2], I find myself agreeing with the "ad hoc" characterization of XML Schema part 1. [1] http://rdfig.xmlhack.com/2001/06/27/2001-06-27.html#993632552.869612 [2] http://www.oasis-open.org/committees/relax-ng/proofsystem.html So please, in any future XML Schema requirements document, explain exactly what requirements motivate various validation constrants; e.g. why <all> cannot have repetitions or be nested within a <sequence>. FYI, my interest in the matter is renewed by recent TAG discussions... 2.1.1 Augmented infosets, PSVI in Minutes of 17 June 2002 TAG teleconference http://www.w3.org/2002/06/17-tag-summary#psvi which followed from a large thread... # Potential new issue: PSVI considered harmful Tim Bray (Wed, Jun 12 2002) http://lists.w3.org/Archives/Public/www-tag/2002Jun/thread.html#85 That thread suggests a new requirement: that type augmentation should be orthogonal to validation. I haven't studied that suggestion well enough to endorse it yet, but FYI, you may get a request for it sometime soon. -- Dan Connolly, W3C http://www.w3.org/People/Connolly/
Received on Tuesday, 18 June 2002 17:05:11 UTC