- From: <Noah_Mendelsohn@lotus.com>
- Date: Tue, 26 Jun 2001 20:31:23 -0400
- To: Dan Connolly <connolly@w3.org>
- Cc: elgey@dstc.qut.edu.au, kohsukekawaguchi@yahoo.com, xmlschema-dev@w3.org
Dan Connolly writes: >> I think you can get at the PSVI via XSV Yes, with the -r switch I believe. It produces an extremely verbose dump, which is frankly mostly useful (and mostly intended) for debugging and testing. It does demonstrate that maintaining and producing the PSVI is computationally tractable, and I expect that at least much of the PSVI will be showing up in widely deployed processors via APIs soon. Certainly, anyone wishing to migrate from DTDs to schema will be interested in features such as attribute value defaults; as you point out, neither Relax nor Trex attempts to support such features of DTDs. I suspect, but don't know for sure, that future versions of XSL and XPath may support templates such as "match on all integers", from which it would be possible when producing HTML from a stylesheet to "format all negative integers in red". That would depend on the PSVI (in this case type information) being available. As to the reason for limitations on <all>. My recollection is that certain members of the workgroup claimed that experiences with SGML had shown that support for generalized <all> would be complex and expensive. Perhaps this was due to the general wish to use DFA technology, perhaps for more subtle reasons. The initial proposal was therefore to stick with the XML 1.0 decision and provide no such support for the so-called "AND connector". Some of us pointed out that a limited form of <all>, allowing only elements, is a particularly good match to database columns, programming language structures, and other systems in which fields must be unique and named, but not ordered. The limited version of <all> was introduced primarily as an 80/20 trade-off to meet an interesting fraction of this need. Implementation of the limited <all> in schema is very simple. Later, the argument was made that allowing minOccurs=0 (I.e. Optional elements in an <all> group) added only modest complexity and was a significant value. It also has what I believe to be the appealing property that it is symmetric with the rules for attributes: at most once each of the named elements, with some being required if you want. So, that is my unofficial recollection of how the design came to be as it is. ------------------------------------------------------------------------ Noah Mendelsohn Voice: 1-617-693-4036 Lotus Development Corp. Fax: 1-617-693-8676 One Rogers Street Cambridge, MA 02142 ------------------------------------------------------------------------
Received on Tuesday, 26 June 2001 20:36:41 UTC