optimization of XSLT code - xslt

Question

While searching for some profiling tools for XSLT, I came across this post. Since a lot of people there suggested to just post the code and offered to give feedback on that, I was wondering if anyone could give me some feedback on mine. I tried this (http://www.saxonica.com/documentation/#!using-xsl/performanceanalysis), but the output html is not very detailed.
I'm new to XSLT and usually work with python/perl, where regex support is much better (however, I won't rule out the possibility that it's just my very basic understanding of XSLT). For the purpose of this project however, I had to work with XSLT. It could be that I'm forcing it to do things in a very unnatural way. Any comments -on performance in particular, but anything else is also welcome, as I'd like to learn- are welcome!
<?xml version="1.0" encoding="UTF-8"?>
<xsl:stylesheet version="2.0" xmlns:xsl="http://www.w3.org/1999/XSL/Transform" xmlns:fo="http://www.w3.org/1999/XSL/Format" xmlns:xs="http://www.w3.org/2001/XMLSchema" xmlns:fn="http://www.w3.org/2005/xpath-functions">
<xsl:template name="my_terms">
<xsl:variable name="excludes" select="not (codeblock or draft-comment or filepath or shortdesc or uicontrol or varname)"/>
<!-- leftover example of how to work with excludes var -->
<!--<xsl:if test=".//*[$excludes]/text()[contains(.,'access management console')]"><li class="prodterm"><b>PB QA:access management console should be "AppCenter"</b></li></xsl:if>-->
<!-- Loop through all sentences and check for deprecated stuff -->
<xsl:for-each select=".//*[$excludes]/text()">
<xsl:variable name="sentenceList" select="tokenize(., '[\.!\?:;]\s+')"/>
<xsl:variable name="segment" select="."/>
<!-- main sentence loop -->
<xsl:for-each select="$sentenceList">
<xsl:variable name="sentence" select="."/>
<!-- very rudimentary sentence length check -->
<xsl:if test="count(tokenize(., '\W+')) > 30"> <li class="prodterm"><b>Sentence too long:</b> <xsl:value-of select="."/></li></xsl:if>
<!-- efforts to flag the shady case of the gerund -->
<xsl:if test="matches(., '\w+ \w+ing (the|a)')">
<!-- some extra checks to weed out the false positives -->
<xsl:if test="not(matches(., '\b(on|about|for|before|while|when|after|by|a|the|an|some|all|every) \w+ing (the|a)', '!i')) and not(matches(., 'during'))">
<li class="prodterm"><b>Possible unclear usage of gerund. If so, consider rewriting:</b> <xsl:value-of select="."/></li>
</xsl:if>
</xsl:if>
<!-- comma's after certain starting phrases -->
<xsl:if test="matches(., '^\s*Therefore[^,]')"><li class="prodterm"><b>Use a comma after starting a sentence with 'Therefore':</b> <xsl:value-of select="."/></li></xsl:if>
<xsl:if test="matches(., '^\s*(If you|Before|When)[^,]+$')"><li class="prodterm"><b>Use a comma after starting a sentence with 'Before', 'If you' or 'When':</b> <xsl:value-of select="."/></li></xsl:if>
<!-- experimenting with phrasal verbs (if there are a lot of verbs in phrasalVerbs.xml, it will be better to have this as the main loop (and do it outside the sentence loop)) -->
<xsl:for-each select="document('phrasalVerbs.xml')/verbs/verb[matches($sentence, concat('.* ', ./#text, ' .*'))]">
<xsl:variable name="verbPart" select="."/>
<xsl:for-each select="$verbPart/particles/particle/#text[matches($sentence, .) and not(matches($sentence, concat($verbPart/#text, ' ', .)))]">
<xsl:variable name="particle" select="."/>
<li class="prodterm"><b>Separated phrasal verb found in:</b> <xsl:value-of select="$sentence"/></li>
</xsl:for-each>
</xsl:for-each>
<!-- checking if conditionals (should be followed by then) -->
<xsl:if test="matches($sentence, '^\s*If\b', '!i') and not(matches($sentence, '\bthen\b', '!i'))"><li class="prodterm"><b>Conditional If found, but no then:</b> <xsl:value-of select="."/></li></xsl:if>
<!-- very dodgy way of detecting passive voice -->
<!--<xsl:if test="matches($sentence, '\b(are|can be|must be) \w+ed\b', '!i')"><li class="prodterm"><b>PB QA:Possible passive voice. If so, consider using active voice for:</b> <xsl:value-of select="."/></li></xsl:if>-->
<xsl:for-each select='document("generalDeprecatedTermsAndPhrases.xml")/terms/dt'>
<xsl:variable name="pattern" select="./#pattern"/>
<xsl:variable name="message" select="./#message"/>
<xsl:variable name="regexFlag" select="./#regexFlag"/>
<!-- <xsl:if test="matches($sentence, $pattern, $regexFlag)"> -->
<xsl:if test="matches($sentence, concat('(^|\W)', $pattern, '($|\W)'), $regexFlag)"> <!-- This is the work around for not being able to use \b when variable is passed on inside matches() -->
<li class="prodterm"><b><xsl:value-of select="$message"/> in: </b> <xsl:value-of select="$sentence"/> </li>
</xsl:if>
</xsl:for-each>
</xsl:for-each>
</xsl:for-each>
</xsl:template>
</xsl:stylesheet>
To get an idea, the stripped down version of my "generalDeprecatedTermsAndPhrases.xml" looks like this:
<dt pattern='to be able to' message="Use 'to' instead of 'to be able to'" regexFlag="i"></dt>
</terms>

Answer 1

The reason that Saxon's profile is not very detailed is that your code is so monolithic: it's all in one great big template rule.
However, being monolithic isn't by itself the cause of any performance problems.
First observation is a functionality problem: your variable
<xsl:variable name="excludes" select="not (codeblock or draft-comment or filepath or shortdesc or uicontrol or varname)"/>
doesn't do what you think. It's evaluated with the root document node as the context item, and its value is a boolean which is true if the outermost element has a name which is not one of those listed. So I think your xsl:for-each that uses [$excludes] as a predicate is applying to all elements, whereas I suspect you intended it to apply to selected elements. I don't know how much that affects the performance.
The main influence on performance will be the cost of evaluating the regular expressions. The best way to find out which ones are causing the problem is to measure the impact of removing them one-by-one. When you've isolated the problem, there may be a way of rewriting the regular expression to make it perform better (e.g. by making it avoid backtracking).