RE: SAXON 5.0 is available

[Kay Michael <Michael.Kay@xxxxxxx>]

> What is 'significantly faster' ? Is it at least 2x faster ? Indeed,
> with Saxon 4.5, I made some test and the compiled style sheet appeared
> to compute exactly twice as fast as the interpreted version, so seems
> to be worth enough. The style sheet I took as example was one you
> provide with SAXON 'book.xsl'.

Yes, I was getting about 2x faster at Saxon 4.5 for many test runs, though
the larger the source document, the smaller the factor. Since then I have
improved the interpreter to do a great deal more of the work at stylesheet
initialisation time, so it is becoming harder to achieve this ratio. If the
speed-up is less than 2x then I think there's not much point, it's easier
just to sit in the garden and watch the hardware get cheaper. But I haven't
given up yet. 

I think for a significant boost in performance over the levels we're getting
from today's interpreters, we have to look to other techniques, e.g. doing
some kind of analysis of the stylesheet and using it to decide what access
paths to build into the source document tree. Plus a lot of conventional
optimisation, I expect, such as handling tail recursion and detecting common
subexpressions. I don't expect to get into those areas myself - I think it
needs someone with a research budget.

Mike Kay


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