On Thu, Dec 11, 2003 at 08:59:08AM +0100, Jirka Kosek wrote:
> Steinar Bang wrote:
>
> >OK. So <xref> and <link> don't work in single pass profiling?
> >
> >Pity. Any chance of fixing it at some point in time? Does it need a
> >new XSL standard?
>
> It works in Saxon, but not in xsltproc. Week ago I proposed
> clarification of EXSLT node-set() function. It seems that EXSLT folks
> are willing to refine definition of this function. If this proposal is
> accepted I think that Daniel (author of xsltproc) will change xsltproc
> to behave in accordance with EXSLT. Then single pass profiling will work
> also in xsltproc.
>
> See:
>
> http://lists.fourthought.com/pipermail/exslt/2003-December/000982.html
Well, if that mean that all keys must be recomputed on all node-set()
results, beware this could cost *a lot* at runtime. What I suspect is that
the keying property that you obtain from Saxon is due to the fact that nodes
are reused, if you really build a new document, then it's a completely
different process, a node which was a key in the first document may not
be a key in the resulting node set and vice-versa, because the key will have
to be recomputed from scratch. Looking at just how much (too much) processing
is done by the docbook stylesheets, the fact that I implement that fully
could have a very serious impact on performances, that could be true for
others processors too.
At some point I may simply just drop the ball, all these node-set and
functions extensions are radical extensions from the XSLT-1.0 spec, they
*completely* change some of the assumptions I made when I engineered
the libxslt implementation based on XSLT-1.0 spec, especially w.r.t. to the
lifetime of objects, scope of access of objects etc...
Daniel
--
Daniel Veillard | Red Hat Network https://rhn.redhat.com/
| libxml GNOME XML XSLT toolkit http://xmlsoft.org/
http://veillard.com/ | Rpmfind RPM search engine http://rpmfind.net/
