HTML LiteralResult performance and encoding

["Voytenko, Dimitry" <DV...@SECTORBASE.COM>]

Hi,

I tried to compare performance of LiteralResult output. So I created
stylesheet which I used to output 1, 2, 3, 4, 5, 6 thousand elements to file
using html output method and ISO-8859-1 encoding. I compared average time of
20 executions of one stylesheet with Xalan 1.0.1 and Xalan 2.2.D13
processors using following steps:
- loaded XSLT stylesheet from file
- executed it one time
- executed it 20 more times and measured time

Stylesheet looked like this:

<xsl:stylesheet version="1.0"
xmlns:xsl="http://www.w3.org/1999/XSL/Transform">
<xsl:output method="html" indent="no" encoding="ISO-8859-1"/>

<xsl:template match="/">
    <xsl:call-template name="html-body"/> <!-- * 1000 -->
</xsl:template>

<xsl:template name="html-body">
	<BODY1>BODY1</BODY1> <!-- 1 thousand elements to output -->
	<BODY2>BODY2</BODY2> <!-- 2 thousand elements to output -->
	<BODY3>BODY3</BODY3> <!-- 3 thousand elements to output -->
	<BODY4>BODY4</BODY4> <!-- 4 thousand elements to output -->
	<BODY5>BODY5</BODY5> <!-- 5 thousand elements to output -->
	<BODY6>BODY6</BODY6> <!-- 6 thousand elements to output -->
</xsl:template>

</xsl:stylesheet>

I found interesting the regression on Xalan 2.2.D13 performance with
increasing the number of elements.
The results are (average time running transformation 20 times in ms):
			1000	2000	3000	4000	5000	6000
Xalan 1.0.1		70	82	96	110	114	126
Xalan 2.2.D13	60	118	166	214	270	318
Diff.			14%	-44%	-73%	-95%	-137%	-152%

These results exclude time of loading XML into DTM or other internal
structures and so should depict pure transformation time. I used sax events
to load XML.

For Xalan 1.0.1 increasing number of output elements on 1000 gives average
13% of increased time. In case of 2.2.D13 it is in average 42% and in
particular 97% for increasing from 1000 to 2000.

After I changed encoding to UTF-8 in <xsl:output> results were almost
similar for both processors (for 6000 elements it was 122 ms). I don't
understand why outputting in encoding other than UTF-8 drastically decreases
performance in one case, but not in the other.
I checked conformance of both processors: Xalan 1.0.1 doesn't output
attribute values correctly in some cases, but I don't see how it could have
affected this case when I didn't use attributes at all.
Can somebody check SerializerToXYZ? May be it contains some excessive code.

Thanks,
Dmitry