Re: [xsl] follow up re: key execution speed

[tcn@xxxxxxxxxxxxx (Trevor Nash)]

Wild idea: include your zip lookup file in the stylesheet either by
external entity reference or xsl:include (in the latter case you have
to put a stylesheet element in the zip file).  In both cases you will
have to invent a namespace for it.

This *may* be enough to convince Xalan that its the same document
every time rather than a new one, hence it will only rebuild the index
when you prepare the stylesheet.  I believe Saxon can be made to
behave this way, but Xalan may be different.

If I were you I think I would write a simple Java class to read the
zip table and access it from an extension function.  Its not
difficult, and the memory footprint will be less than a tenth of the
XSLT method.

Regards,
Trevor Nash

>I wrote in last week and received much good advice on using keys to perform
>lookups during XSL execution.  I was advised, properly I am sure, that keys
>are damn fast and the best way to do nearly anything.  However my problem is
>a little unusual, and is resulting in unsatisfactory performance.
>
>The trick is this: while I have been advised that keys are speed for any
>lookup that will be performed more than once in a transform, my situation is
>a little more complicated. I only do this lookup once per transform, but I
>perform this transform many times (thousands!) on different source
>documents.  I was hoping to get the speed benefit of the key across these
>multiple transformations.
>
>My XSL templates make use of two data files, Zips.xml, which maps zip codes
>to dealer codes, and Dealers.xml, which contains information about each
>dealer. The idea is to use a key (ziptorc) over the (very large - 400K) Zips
>file (looks like <ZIP zip='10001' rc='2'/>) to perform the zip-to-dealer
>mapping, then to use another key over the dealer file (rctodealer) to
>perform a generic templating of the dealer informaiton. My templates are as
>follows:
>
><xsl:key name="ziptorc" match="ZIP" use="@zip"/>
><xsl:key name="rctodealer" match="DEALER" use="@rc"/>
><xsl:template name="TARGETBYZIP">
>	<!-- target by ZIP -->
>	<xsl:variable name="ZIP">
>		<xsl:value-of select="ZIP/@val"/>
>	</xsl:variable>
>	<xsl:variable name="RC">
>		<xsl:for-each select="document('Zips.xml')">
>			<xsl:value-of select="key('ziptorc', $ZIP)/@rc"/>
>		</xsl:for-each>
>	</xsl:variable>
>	<xsl:for-each select="document('Dealers.xml')">
>		<xsl:apply-templates select="key('rctodealer', $RC)"/>
>	</xsl:for-each>
></xsl:template>
>
><xsl:template match="DEALER">
>	<!-- generic template for displaying dealer information here -->
></xsl:template>
>
>I am using Xalan, creating and persisting a Templates object. I am then
>creating a Transformer object to perform each transform. Performance is very
>slow; seems to me that it is reloading the data files and recreating the
>keys every time it performs the transform.
>
>Is there some set of changes, to the XSL or to the Java, that will enable me
>to get the blazing performance I require?
>
>Thanks very much for your help.
>
> XSL-List info and archive:  http://www.mulberrytech.com/xsl/xsl-list

--
Traditional training & distance learning,
Consultancy by email

Melvaig Software Engineering Limited
voice:     +44 (0) 1445 771 271 
email:     tcn@xxxxxxxxxxxxx

 XSL-List info and archive:  http://www.mulberrytech.com/xsl/xsl-list