RE: Stylesheet optimisation

> Do you think there is no possible
> way to boost performance and scalability of so-called "serious"
> stylesheets further than today's crop of processors?

No, I'm sure there are lots of improvements possible. It would be very rash
to think otherwise - after 30 years of relational databases there are still
people working on improved join algorithms - though there's definitely a law
of diminishing returns.

I suspect it splits into two sub-problems: simple stylesheets and complex
stylesheets. For simple stylesheets the challenge is to avoid doing all the
things that aren't necessary when the stylesheet is sufficiently simple
(perhaps like building the tree in memory). For complex stylesheets it's
some combination of all the things that we know about complex query
optimisation, compiler optimisation, and pattern matching.

In the middle ground I suspect there are a lot of stylesheets where
performance won't improve all that much from current levels. 

One of my concerns is that there are lots of things in XSLT (such as
numbering and elimination of duplicates) which currently have n-squared
performance. In some of these cases I think it's quite hard to detect the
cases where a more efficient algorithm is possible: the language itself
isn't declarative enough (or to put it another way, the declarative part of
the language is insufficiently powerful). But I'm not an expert on
optimisation so perhaps someone can correct me.

Mike Kay


 XSL-List info and archive:  http://www.mulberrytech.com/xsl/xsl-list