Bug in Xalan j_2_2_D10 preceding preceding-sibling

> I am witnessing a strange behavior with Xalan-j_2_2_D10. I am using
> xsl:key to construct a list of unique elements and later on using
> xsl:for-each to display these unique results. I am observing that both
> preceding and preceding-sibling are producing the same output, which is
> surprising since, preceding-sibling is supposed to return all the siblings
> that occur before the context node in the document order and belong to the
> same parent.
> 
> It would be clearer by reading the example below
> 
> source XML( This is a sample XML which has a similar structure to what I
> am working on)
> 
> <?xml version="1.0" encoding="ISO-8859-1"?>
> <!-- File Created on   03-Oct-2001 13:10:34-->
> <!DOCTYPE ALBUM_LIST>
> <ALBUM_LIST>
> 	<ALBUM ID="Q9001">
> 		<TRACK ID="001"/>
> 		<TRACK ID="002"/>
> 		<TRACK ID="003"/>
> 	</ALBUM>
> 	<ALBUM ID="Q9001">
> 		<TRACK ID="004"/>
> 		<TRACK ID="005"/>
> 		<TRACK ID="003"/>
> 	</ALBUM>
> 	<ALBUM ID="Q9001">
> 		<TRACK ID="006"/>
> 		<TRACK ID="002"/>
> 		<TRACK ID="003"/>
> 	</ALBUM>
> </ALBUM_LIST>
> 
> I want to extract the list of all unique tracks, which would be tracks
> with ID 001,002,003,004,005,006. Here is the XSL that I am using
> 
> <?xml version="1.0"?>
> <xsl:stylesheet xmlns:xsl="http://www.w3.org/1999/XSL/Transform"
> version="1.0" xmlns:lxslt="http://xml.apache.org/xslt">
> <xsl:output method="xml" indent="yes" />
> <xsl:strip-space elements="*" />
> <xsl:key name="ALBUM_TRACK"
> match="ALBUM/TRACK[not(@ID=preceding::node()/@ID)]" use="@ID"/>
> 
> <xsl:template match="/" >
>   <objects>
>     <xsl:apply-templates select="ALBUM_LIST/ALBUM"/>
>   </objects>
> </xsl:template>
> 
> <xsl:template match="ALBUM">
>     <xsl:for-each
> select="TRACK[generate-id()=generate-id(key('ALBUM_TRACK',@ID))]">
>       <object name="TRACK">
> 	<xsl:value-of select="@ID"/>
>       </object>
>     </xsl:for-each>
> </xsl:template>
> </xsl:stylesheet>
> 
> 
> OUTPUT
> 
> <?xml version="1.0" encoding="UTF-8"?>
> <objects>
> <object  name="TRACK">001</object>
> <object  name="TRACK">002</object>
> <object  name="TRACK">003</object>
> <object  name="TRACK">004</object>
> <object  name="TRACK">005</object>
> <object  name="TRACK">006</object>
> </objects>
> 
> This looks like a very simple XSL transformation. Now if I replace
> preceding by preceding-sibling in the xsl:key definition, still the OUTPUT
> is the SAME. Though there is a significant performance gain. I believe
> that this is wrong behavior and probably a bug in Xalan. How can preceding
> and preceding-sibling produce the same result. Using preceding-sibling,
> the xsl-key should NOT store the unique Tracks, as it cannot look beyond
> the parent of Track namely Album, but it seems that preceding-sibling
> looks into all the ALBUM elements and thus selects unique TRACK elements.
> 
> Also, if I replace preceding:: by following:: axis in the same key, there
> is a drastic difference in performance time. I would believe that just by
> replacing preceding:: by following::  without any other change, the XSL
> transformation time should remain the same, but it is not so in my case.
> 
> with preceding::, the total time to run transformation is 30 while with
> following:: the time is 8 minutes.
> 
> So, looks like there are 2 anomalies, I am using a much large source xml,
> about 5MB long.
> 
> Please let me know the reason of these two anomalies I am observing and
> whether this really is a bug or not
> 
> Nisheet

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