DOCBOOK-APPS: Re: speedy XSLT processor for win

[Norman Walsh <ndw at nwalsh dot com>]

/ Daniel Veillard <veillard@redhat.com> was heard to say:
| On Thu, Aug 22, 2002 at 07:38:20AM -0400, Norman Walsh wrote:
|> / Bob Stayton <bobs@caldera.com> was heard to say:
|> | I'll bet you are chunking out a lot of files.  If so, then
|> | you are probably I/O bound.  I get similar results on
|> 
|> I/O bound, or just working really hard to calculate all of the
|> navigational links.
|> 
|> Chunking really big documents requires some potentially expensive
|> operations over the document tree.
|
|   Basically the next and previous are recomputed *everytime* for
| each chunk, independantly that next->prev is the current start node.
| At least minimal caching here would help, okay I know one cannot
| override variables, but still what a waste of power... 

I suppose the answer is a two-pass process, which could be arranged
with extension functions in XSLT 1.0 and will be available in XSLT 2.0
natively.

/me mutters something about even more options and hair loss.

|   Can't you just compute the boundaries once, use a key() to store them
| and come back to linear cost for this computation ?

Take a look at the xpath expressions used to compute previous and next.
If you can short circuit that into a key(), dinner's on me.

                                        Be seeing you,
                                          norm

-- 
Norman Walsh <ndw@nwalsh.com>      | Life is a great bundle of little
http://www.oasis-open.org/docbook/ | things.--Oliver Wendell Holmes
Chair, DocBook Technical Committee |