RE: [xsl] Using Saxon 8.5 and collection() to process a directory of XML files

["Welch Andrew (ELS)" <A.Welch@xxxxxxxxxxxx>]

> > I now have Saxon 8.5b processing a directory of xml files and 
> > generating the required html files, using the following:
> > 
> > <xsl:for-each select="collection(concat($xmlDir, '?select=*.xml'))">
> > 	<xsl:result-document href="{$outputDir}/{$fileName}.html">
> > 	  ...
> > 	</xsl:result-document>
> > 	<xsl:for-each select="saxon:discard-document(.)" /> 
> </xsl:for-each>
> > 
> > -Am I using saxon:discard-document() correctly... Seems odd?
> 
> There's a danger Saxon will optimize it out if you use it in 
> a context where the result of the function (which is the same 
> as the input) isn't used. So a safer usage is:
> 
> > <xsl:for-each select="collection(concat($xmlDir, '?select=*.xml'))">
>     <xsl:for-each select="saxon:discard-document(.)">
> > 	<xsl:result-document href="{$outputDir}/{$fileName}.html">
> > 	  ...
> > 	</xsl:result-document>
> >   </xsl:for-each>
> > </xsl:for-each>
> 
> or you could tag it on to the outer for-each:
> 
>  <xsl:for-each select="collection(concat($xmlDir,
> '?select=*.xml'))/saxon:discard-document(.)">

I don't think the any keys generated for each document are being discarded
with it.  For example:

<xsl:for-each select="collection(concat($xmlDir, '?select=*.xml'))">
	<xsl:if test="saxon:discard-document(.)">Discarded.</xsl:if>
</xsl:for-each>

This seems to work fine, memory usage hovers around the same amount.  If a
call to a key is added:

<xsl:for-each select="collection(concat($xmlDir, '?select=*.xml'))">
	<xsl:value-of select="key('somekey','someval')"/>
	<xsl:if test="saxon:discard-document(.)">Discarded.</xsl:if>
</xsl:for-each>

Memory usage grows until an OutOfMemoryError is thrown.

Is there a workaround for this?


> > -Is it possible to continue the processing should one of the source 
> > files contain an error (say a parse error)?
> > 
> 
> Yes, there's a query parameter 
> ?select=*.xml;on-error=ignore|warning|fail

This works well, thanks.

andrew.