This got accidentally send to Sujata, here it is, again, to the
list...(possibly out of date now, since the discussion has gone on)
-------- Original Message --------
Subject: Re: [xsl] Optimizing XSLT iteration
Date: Mon, 08 Oct 2007 03:48:12 +0200
From: Abel Braaksma <abel.online@xxxxxxxxx>
To: Sujata Gohad <sgohad@xxxxxxx>
Sujata Gohad wrote:
What is your expectation wrt to the time Saxon takes to transform
given the XSLT attached?
I've looked a bit at your source and I am a bit worried about its
maintainability, but that's not your major point here (there are large
xsl:when's that can be removed, and there are nested xsl:for-each that
serve no special purpose and can be replaced by matching
templates/apply-template pairs).
But my main point is the performance: it took my computer, while still
doing the math thing on the background and a dozen other things, about 5
minute 43 seconds to run against your augmented (up to 68MB) source
code (after I removed 5 obvious errors). That is still about 5 times
faster than your measurements and your system is faster (on paper) than
mine.
In lieu of this, I assume to be able to optimize your stylesheet to run
in about 20 to 30 seconds with about 80MB, based on my earlier timings.
But I wonder why there is so much difference between my timing and yours.
I use Saxon B 8.9.0.4 for Java. I ran it with the following command:
saxon -s sample.xml hvm_to_svg.xsl > out.xml
Do you perhaps run it from within an IDE? This may add a lot of extra
timing to interpreting and showing the result document (I use Eclipse,
running it within Eclipse is much-much-much slower with large documents).
Cheers,
-- Abel Braaksma
PS: it is now about 3.45 AM up here, time to call it a day ;)