JRuby-Nokogiri fails to parse XSLT stylesheet

[denisdefreyne]

Nokogiri on JRuby fails to recognise <xsl:value-of select="$foo"/> and fails with an exception.

Steps to reproduce

Execute the following code on JRuby:

input = <<-EOS
<?xml version="1.0" encoding="utf-8"?>
<xsl:stylesheet version="1.0" xmlns:xsl="http://www.w3.org/1999/XSL/Transform">
  <xsl:output method="xml" version="1.0" encoding="utf-8" indent="yes"/>
  <xsl:template match="/">
    <h1><xsl:value-of select="$foo"/></h1>
  </xsl:template>
</xsl:stylesheet>
EOS

require 'nokogiri'
::Nokogiri::XSLT(input)

Expected output

% ruby test.rb
%

Tested on MRI 1.9.3.

Actual output

% ruby test.rb
RuntimeError: could not parse xslt stylesheet
  parse_stylesheet_doc at nokogiri/XsltStylesheet.java:166
                 parse at /Users/ddfreyne/.gem/jruby/1.9.3/gems/nokogiri-1.6.1-java/lib/nokogiri/xslt.rb:30
                  XSLT at /Users/ddfreyne/.gem/jruby/1.9.3/gems/nokogiri-1.6.1-java/lib/nokogiri/xslt.rb:13
                (root) at test2.rb:19

Tested on JRuby 1.7.11.

Notes

Removing the <xsl:value-of select="$foo"/> makes the error disappear.

[headius]

The actual error message here, hidden by the Java impl, is "Variable or parameter 'foo' is undefined." Is something eagerly trying to parse that value out and look it up?

[headius]

Full trace: https://gist.github.com/headius/180d6ffe99d0e6527483

It appears that the XSLT engine is attempting to resolve $foo. Perhaps there's a config to turn that off?

[headius]

Ok, so all resources I looked at indicated that this stylesheet needs an xsl:variable or xsl:param tag to declare $foo as a variable that will be provided later. Perhaps libxml is being so forgiving it allows this to be a deferred variable with it's not immediately present?

This patch to your stylesheet makes it parse fine:

--- blah.rb 2014-04-21 08:34:25.000000000 -0500
+++ blah2.rb    2014-04-21 08:34:21.000000000 -0500
@@ -2,6 +2,7 @@
 <?xml version="1.0" encoding="utf-8"?>
 <xsl:stylesheet version="1.0" xmlns:xsl="http://www.w3.org/1999/XSL/Transform">
   <xsl:output method="xml" version="1.0" encoding="utf-8" indent="yes"/>
+  <xsl:variable name="foo"/>
   <xsl:template match="/">
     <h1><xsl:value-of select="$foo"/></h1>
   </xsl:template>

So the question then is whether libxml is too permissive (which is usually the case) or xalan is too strict. I could not find anything obvious in the W3C spec about when XSL variables must resolve, but I fed your stylesheet into a number of online xslt transformers and they all failed to parse it.

I'm inclined to call this another case where libxml is wrong by allowing the stylesheet to compile in the first place.

[denisdefreyne]

Introducing the xsl:variable does the trick.

In the nanoc test case, which was a bit larger than the one provided in this issue, I invoke the XSLT with an external param (similar to xslt.transform(doc, ['key', 'value']) given in the Nokogiri::XSLT::Stylesheet documentation). Swapping the xsl:variable with xsl:param works in this case.

It’s good to know that libxml/libxslt is occasionally too permissive.

Thanks for investigating!