RE: [xsl] hardware xml / xslt

[Joseph Kesselman <keshlam at us dot ibm dot com>]

>> Who is the fastest, Mike? Caucho's Resin? libxml/xsl? It is 
>> not Saxon. Who do you think they should use?
> 
>I heard recently of one customer site where putting in xt
>instead of Xalan gave a 12-fold improvement

We *have* made some fairly major changes over the last year -- factoring 
out common XPath subexpressions, for example -- so the phrase "instead of 
Xalan" begs the question "Instead of which version of Xalan". On some 
stylesheets the gain has been significant, depending on exactly what 
you're doing. 

There's also the question of whether you're talking about traditional 
Xalan (interpretive mode) or compiled Xalan (XSLTC); essentially, Xalan 
now contains two quite different solutions, with different 
characteristics.



Traditionally, Xalan (like Saxon), has been putting a lot of its effort 
into nailing down the remaining loose corners of XSLT 1.0 compatability 
and bug-blatting, and has started looking at 2.0. XSLT is very much a 
90/10 proposition; compatability with the last 10% of the spec seems to 
create 90% of the performance issues... so a processor which has the goal 
of full compliance all the time may, in fact, suffer a performance hit. 
(And I'm still not convinced XSLT 2.0 will ever perform as well as XSLT 
1.0; we'll see.)

But Xalan now has a somewhat larger team, which should free some resources 
to return to performance work. I can't promise Xalan will suddenly become 
the top performer; all I can say is that we all want, and expect, it to 
continue to improve.

______________________________________
Joe Kesselman  / IBM Research

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