RE: [xsl] Comparing grouping techniques in terms of performance

["Michael Kay" <mhk@xxxxxxxxx>]

> Of course. But this email is not intended as a 100% solid 
> comparison test.
> But it is solid enough to prove my point, I think.

But I'm not sure what your point is. That one stylesheet doing A, B, C, and
D performs better than one doing P, Q, R, and S? I would like to understand
why it performs so much better, and that means isolating which of the
variables account for the difference. (In my experience, it might be
something quite unexpected).

> 
> But I would like to extend this test on another processor - I 
> could use
> yours for example... did you do any tests on this kind of 
> subject, Michael
> (comparing stylesheets in terms of performance)? 

Jeni Tennison did some tests on grouping in XSLT 1.0 a while back, one of
the things it revealed was that generate-id() in Saxon was at that time very
inefficient; but it has since been greatly improved.

> Is XSLT 2.0 doing better
> with regards on these kind of tests?
> 

There are some things that perform much better in 2.0 because you don't have
to code them in such a roundabout way. On the whole though I haven't done
much serious work on Saxon performance recently, I have been concentrating
on functionality. All I do is to check that performance doesn't slip back
from one release to the next, and occasionally fix specific performance bugs
if identified by users. There are many opportunities for optimization that
are stored up for future work.

Michael Kay