> Granted, dynamic validation (=at runtime) is always possible,
> but it is
> certainly not a big help. Errors are discovered after having gone to
> production, doesn't sound good to me.
Actually, it's an enormous help compared with not doing it at all.
I agree that static checking, where possible, is even better. Saxon doesn't
do any static checking against a schema yet.
> I am sure you will not try to add exact static typing based
> on Relax NG for XSLT in the next version of Saxon?
No, but I wouldn't rule it out eventually.
> There are papers that analyze
> this for subsets of XSLT
Can you give me a reference? I'm not aware of such papers.
> but doing it exactly, for the full
> language,
> is - as far as I remember - undecidable.
Even with XML Schema, it's unprovable that a given stylesheet is guaranteed
to create valid output (to do strict static typing, in the style of the
XQuery static typing option, we would have to throw out
xsl:apply-templates). But pragmatically, I think it's possible to give
static warnings for a good proportion of user errors. And I suspect that
most of the heuristics to do this are equally applicable to DTDs or Relax
NG.
Michael Kay
