> At first, my old implementation does something like:
> 
> <xsl:variable name="mayContinue">
> 	<xsl:choose>
> 		<xsl:when 
> test="some_big_test_over_big_xml_trees_which_was_very_slow">1<
> /xsl:when>
> 		<xsl:otherwise>0</xsl:otherwise>
> 	</xsl:choose>
> 
> 	<xsl:if test="$mayContinue=1">
> 
> 		<xsl:if 
> test="other_big_test_to_check_if_something_will_be_processed">
> 			<!-- print header -->
> 			<!-- process data and generate output -->
> 			<!-- print bottom -->
> 		</xsl:if>
> 
> 	</xsl:if>

It can't do exactly that because you can't access a variable within its own
definition, but perhaps you just left out the </xsl:variable> end tag.
> 
> .....
> 	<xsl:if test="$mayContinue=1">
> 		<xsl:variable name="processOutput">
> 			<!-- process data and generate output -->
> 		</xsl:variable>
> 
> 		<xsl:if test="string-length($processOutput)!=0">
> 			<!-- print header -->
> 			<xsl:value-of select="$processOutput"/>
> 			<!-- print bottom -->
> 		</xsl:if>
> 
> 	</xsl:if>
> 
> But I am afraid that when $processOutput will be too big (for 
> example 100000 
> lines of text output), test for it's string-length may be slow.
> 

As I mentioned, it depends on the processor as to whether it optimises this.
Although Saxon will optimise the test string-length($X) = 0 to the
equivalent of string($X) = '', it will still do the conversion of the result
tree fragment to a string, which is probably the expensive part of the
operation. It might be cheaper to test if $x contains any text nodes, that
is: test="xx:node-set($X)//text()" or in 2.0 simply test="$X//text()".

But as mentioned before, it all depends on what your XSLT processor chooses
to optimise, which you can only tell by measurement.

Michael Kay
http://www.saxonica.com/