Re: [xslt] Isn't it time to make libxslt multi-threaded?

[Liam R E Quin <liam holoweb net>]

On Thu, 2012-07-26 at 11:31 +0200, Stefan Behnel wrote:
> Дмитрий Грибов, 26.07.2012 10:17:
[...]
> > Saxon runs massively faster on Java? That's something really new for me.
> 
> Note that Liam wrote "using explicit typing on variables". As I read it, he
> claims that XSLT 2.0 can be executed more efficiently than XSLT 1.0 when
> the XSLT code is statically typed using XML Schema, likely because runtime
> optimisation of untyped XSLT code may not be sufficient (or implemented
> well enough, or available at all, or whatever). Sounds more like a tooling
> problem to me.

Sure, it's a tooling problem - libxslt is a f*cking crap optimizer. But
that's partly because Saxon has had a lot more active work over many
years, and partly because XSLT 2 has a number of changes intended to
help improve performance. Mike Kay has published papers about the work
he did.

JVMs have also advanced, so that even if Saxon's compiled stylesheets
are only, say, twice as fast as xsltproc's, you get a win. It does (as I
said) depend on your stylesheets.

> Personally, I find a processor for XSLT 1.0 plus the usual extensions (and
> easy extensibility from user code) much more appealing than one for XSLT
> 2.0 that includes all that XML Schema cruft. But maybe that's just me.

It's just you :-) I'd really miss the easier debugging and better error
checking from being able to say "this variable must contain an integer",
the extra performance, a portable, standard way to create multiple
output documents, user-defined functions written in XSLT, etc etc.

Note that variable typing (the "as" attribute) does not require "all
that XML Schema cruft".

Liam

-- 
Liam Quin - XML Activity Lead, W3C, http://www.w3.org/People/Quin/
Pictures from old books: http://fromoldbooks.org/
Ankh: irc.sorcery.net irc.gnome.org freenode/#xml