Re: XPath perfomance in attributes or nodes?

I would like to ask the XSLT implementors on this list whether their
processors build keys lazily (i.e. on the first use of the key() function),
or does <xsl:key> have a cost even when the key is not used later ?
When is the cost of building the index offset by the performance gain ?

For example, suppose the input is 10000 elements like so :

<doc>
<elt att="1">
</elt>
<elt att="2">
</elt>
<elt att="3">
</elt>
...
</doc>

and there is an external parameter that gives the stylesheet
an attribute value to look for. Which would be faster :

<xsl:key name="idx" match="elt" use="@att" />
<xsl:param name="param" />

<xsl:template ...>
    ...
    <xsl:apply-templates select="key('idx',$param)" />
    ...
</xsl:template>

Or :

<xsl:param name="param" />

<xsl:template ...>
    ...
    <xsl:apply-templates select="/doc/elt[@att=$param]" />
    ...
</xsl:template>

Assuming the <xsl:apply-templates> above is executed
only once, it seems to me that the key-less version
should perform a bit faster. But as I have never implemented
an XSLT processor I would like to have the opinion of
masters of the subject.

Cheers,
--Jonathan

----- Original Message ----- From: "Peter Davis" <pdavis152@xxxxxxxxx>
To: <xsl-list@xxxxxxxxxxxxxxxxxxxxxx>
Sent: Thursday, March 07, 2002 3:52 AM
Subject: Re: XPath perfomance in attributes or nodes?

<xsl:key name="project-uid" match="project" use="@uid"/>
<!-- or use="uid" -->
<xsl:template ...>
 <xsl:apply-templates select="key('project-uid',
    '{6A97F258-93A7-4C5E-B712-052130A5208C}')[1]"/>
 <!-- [1] is probably not necessary -->
</xsl:template>

will be much faster than either approach.

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