Thoughts on XSLT grouping

A

<!--
  From http://www.w3.org/TR/xslt-30/
  A standard grouping exercise to wrap a group of elements
-->
<xsl:template match="body">
  <chapter>
    <xsl:for-each-group select="*" group-starting-with="h2">
      <section title="{self::h2}">
        <xsl:for-each select="current-group()[self::p]">
          <para><xsl:value-of select="."/></para>
        </xsl:for-each> 
      </section>
    </xsl:for-each-group>
  </chapter>
</xsl:template>

B

<!--
  This can be rewritten using a subtree to create a node at the group level
  to match on.
-->
<xsl:template match="body">
  <chapter>
    <xsl:for-each-group select="*" group-starting-with="h2">
      <xsl:variable name="group" as="element(group)">
        <group>
          <xsl:sequence select="current-group()"/>
        </group>
      </xsl:variable>
      <xsl:apply-templates select="$group"/>
    </xsl:for-each-group>
  </chapter>
</xsl:template>

<!--
  The problem with this method is that it breaks the context of the nodes in
  the group. You can't do something in this template like ancestor::body/foo
-->
<xsl:template match="group">
  <section title="{h2}">
    <xsl:for-each select="p">
      <para><xsl:value-of select="."/></para>
    </xsl:for-each> 
  </section>
</xsl:template>

C

<!--
  What I'd like is to have some kind of psuedo-node e.g. group(), that 
  would be created by the for-each-group instruction that could be
  part of a match pattern. Within the pseudo-node the current-group() elements
  would be accessible via the child axis and the current-grouping-key() 
  function would be defined
-->
<xsl:template match="body">
  <chapter>
    <xsl:for-each-group select="*" group-starting-with="h2">
      <xsl:apply-templates select="current-group()"/>
    </xsl:for-each-group>
  </chapter>
</xsl:template>

<xsl:template match="group()">
  <section title="{h2}">
    <xsl:apply-templates select="p"/>
  </section>
</xsl:template>

<xsl:template match="p">
  <para><xsl:value-of select="."/></para>
</xsl:template>

If you want something to match on at group level, you could use two passes that process the whole document. The first pass groups, the next pass matches on the groups. We often do multi-pass-whole-document transformations because we can save snapshots of the whole document after each pass for debugging purposes. Of course memory consumption and also transformation time may be an issue.

I don’t fathom which modifications would be necessary on the XDM in order to support this. I believe most people wouldn’t think that the benefits outweigh the additional complexity.