[xsl] Is it OK for xsl:output to affect the construction of an unserialized result tree?

Subject: [xsl] Is it OK for xsl:output to affect the construction of an unserialized result tree?
From: Elliotte Rusty Harold <elharo@xxxxxxxxxxxxxxx>
Date: Tue, 20 Jul 2004 11:13:11 -0400
This is a practical issue relating to proposed Xalan-J bug:


Consider this stylesheet:

<?xml version="1.0"?> <xsl:stylesheet version="1.0"
<!-- test a stylesheet that use method="text" -->
<xsl:output method="text"/>

<xsl:template match="/">
<xsl:processing-instruction name="test">PIs are not treated as literals in


Suppose we transform a DOM Document, XOM document, SAX InputSource or some other such representation of an XML document. Furthermore suppose that instead of serializing the output as a stream or a string we create another DOM document, XOM document, etc. Should the xsl:output element have any effect?

My claim is that it should not. Anything said in the xsl:output element applies only when the document is serialized to a an actual XML document represented as a sequence of characters or bytes, not before. This is what earlier versions of Xalan did. Faced with this case, they would produce a result tree containing six child nodes of different kinds,
element, comment, and processing instructions.

However, in the last few versions, Xalan produces a single text node containing all the element content. The elements, comments, and processing instructions should be reduced to plain text only when the document is serialized, not before. In my opinion, Xalan appears to be jumping the gun when the output method is text and incorrectly assuming that it can generate a single text node. However, at least on of the Xalan developers disagrees with me, and thinks this is acceptable and indeed desirable behavior.

What does everyone else think? I've always assumed xsl:output only applied in serialization, not result tree construction but Joe Kesselman thinks the spec language is at best ambiguous on this point. Interpretations from working group members would be particularly helpful.

One thing that worries me is that if this is made optional (because xsl:output is support clearly optional according to the spec) then this becomes yet another point where different implementations behave differently with no clear way for client code to figure out what's likely to happen.


Elliotte Rusty Harold
Effective XML (Addison-Wesley, 2003)
http://www.cafeconleche.org/books/effectivexml http://www.amazon.com/exec/obidos/ISBN%3D0321150406/ref%3Dnosim/cafeaulaitA

Current Thread