Re: [xsl] extrem performance break down

Subject: Re: [xsl] extrem performance break down
From: Michael Kay <mike@xxxxxxxxxxxx>
Date: Wed, 12 Dec 2012 10:17:27 +0000
I would have expected the first pattern to perform pretty badly, because of the use of [following::text()], which means that every time you look at a paragraph you also have to look at all the following text nodes.

But in your new pattern you are looking at all the preceding nodes of all the ancestors, and you are doing this for every element, not only for selected elements like p and ul.

You could probably get some improvement by reordering the predicates (the optimum order is some combination of putting the cheapest-to-evaluate predicate first, and putting the most-selective predicate first).

However, I would suggest a completely different approach. Whenever you have match patterns that need to search the document like this, it's often best to invert things: start by finding the nodes that match this complex condition and put this node-set in a global variable $special, then replace your pattern with

match="*[. intersect $special]"

Another point to be aware of with pattern matching in Saxon is that it will always work best if Saxon can determine that the pattern will only match elements of one particular element name. So you might well be better off breaking this into separate patterns that start

p[....]
ul[....]
span[.....]

Michael Kay
Saxonica

On 12/12/2012 08:14, Szabo, Patrick (LNG-VIE) wrote:
Hi,

I'm using Saxon-B 9.0.0.6 that comes with Oxygen 9 for profiling.
My Stylesheet is quite long so i don't wan't to post the whole thing but my problems seems to reside within one template anyway.

I have the following template:

<xsl:template match="*[self::p or self::ul][following::text()[starts-with(.,'Zum aktuellen Rechtsstand')] and ancestor::div[@class='comm.body'] and text()][1]" priority="20">
         ***something happens***
</xsl:template>

When i change the xpath to

<xsl:template match="*[not(ancestor-or-self::span[@class='heading' and count(preceding::span[@class='heading']) = 0])][self::p or self::ul or self::span[not(@class='heading' and count(preceding::span[@class='heading']) = 0)]][following::text()[starts-with(.,'Zum aktuellen Rechtsstand')] and ancestor::div[@class='comm.body'] and .//text()][1]" priority="20">
         **still same things happens**
</xsl:template>

it adds 104 seconds to the execution time. Problem is i need to change it because otherwise it matches thing it's not supposed to.

Can anyone tell me what exactly I'm doing that is causing this performance break ?
As soo as i revert that one change execution time goes down again.

cheers


. . . . . . . . . . . . . . . . . . . . . . . . . . Ing. Patrick Szabo Developer LexisNexis A-1030 Wien, Marxergasse 25

mailto:Patrick.Szabo@xxxxxxxxxxxxx
Tel.: +43 1 53452 1573
Fax.: +43 1 534 52 146

. . . . . . . . . . . . . . . . . . . . . . . . . .

Current Thread