Lists Home |
Date Index |
> -----Original Message-----
> From: Jonathan Robie [mailto:email@example.com]
> Sent: Friday, December 28, 2001 12:31 PM
> To: Paul T; Champion, Mike; firstname.lastname@example.org
> Subject: Re: [xml-dev] W3C's five new XQuery/Xpath2 working drafts -
> Still missing Updates
> But I don't quite grasp your argument. Certainly we need to provide
> validation for business rules, but shouldn't these business rules be
> defined on top of well-typed data?
I most certainly do NOT want a query to ignore purchase orders that have an
extra <p> tag in the <description> element, whatever the putative PO schema
says on the subject!
> They certainly are in relational
> databases. Do you think that RDBMS systems should simply stop using
> datatypes? If not, why do you think that XQuery should?
I'd appreciate some elaboration on the role of "types" in RDBMS systems,
preferably from someone who actually uses them in their day job. My very
limited experience in this area tends to support Dare and Paul's remarks
about constraints: most "types" I've ever used are simply strings,
dates, or integers; what matters are the primary - foreign key constraints
(to ensure that joins are meaningful), or business rules that are much more
practical to validate with procedural code.
This discussion got me curious, so I checked out Date's INTRODUCTION TO
DATABASE SYSTEMS to see what he says about the importance of types. He does
spend about 15 pages (in the 1995 edition) discussing domains and types, but
makes a couple of points I found intriguing: First, he implies that
user-defined types are difficult to work with and rarely used except by
experts. Also he (predictably) concludes "SQL supports almost none of the
ideas presented here", referring the reader to a 1-page summary in a
previous chapter on SQL's lamentable lack of support for Codd's ideas about
domains (which Date equates with types).
OK, Date is a bit of a crank about SQL, so let's see what others say... a
quick Googling of the subject also left me with the impression that doing
anything with SQL types other than distinguishing strings from dates from
integers from floats from binary is an invitation for interoperability
nightmares. So -- again, this is more or less a troll, in hopes that someone
will educate me -- I get the impression that "types" as discussed in this
thread are on the bleeding edge of RDBMS technology.
I KNOW that types are the blood dripping off the bleeding edge of XML
technology :~) I listened to too many of James Clark and Makoto Murata's
analyses of the W3C XML Schema spec in Florida recently to have any
illusions that it is the last word on the subject of XML schemas or
datatypes. Now that ISO is taking on the job of standardizing Relax NG and
Schematron as well as W3C XSD, the "it's the only standard we have, so we
must support it" argument goes away.
The W3C is most successful when the members butt heads, horse trade,
abstract away insignificant differences, and come up with mutually agreeable
ways of doing things that we all basically know how to do. It is least
successful when it tries to do computer science by committee. XPath 2.0 is
very much in the first category, and XQuery as XPath + constructors +
updates would not push the boundaries of solid knowledge very far either.
I just don't like the idea of holding up the standardization of
well-understood techniques in order to explore terra incognita.