Skip to content

Multiple result documents

A 1.0 stylesheet produces exactly one output. xsl:result-document (2.0+) lifts that limit: a single transform can write many files — one HTML page per chapter, one document per invoice, a sitemap plus the pages it lists. The principal output stream still goes where it always did; each xsl:result-document opens a secondary one.

The basics

Wrap the content for a secondary file in xsl:result-document, giving it an href (resolved against the base output URI) and, optionally, its own output settings:

<xsl:template match="/catalog">
  <!-- principal output: an index -->
  <html><body>
    <h1>Catalogue</h1>
    <ul>
      <xsl:for-each select="cd">
        <li><a href="{position()}.html"><xsl:value-of select="title"/></a></li>
      </xsl:for-each>
    </ul>
  </body></html>

  <!-- one secondary file per cd -->
  <xsl:for-each select="cd">
    <xsl:result-document href="{position()}.html" method="html" indent="yes">
      <html><body>
        <h1><xsl:value-of select="title"/></h1>
        <p><xsl:value-of select="artist"/> — <xsl:value-of select="price"/></p>
      </body></html>
    </xsl:result-document>
  </xsl:for-each>
</xsl:template>

Run once, this emits the index as the main result plus 1.html, 2.html, 3.html alongside it. Each xsl:result-document can carry its own method, encoding, indent — they are independent documents, not fragments of one.

Per-document output settings

Because each result is independent, you can mix formats in one run — an HTML page and a companion JSON data file, say:

<xsl:result-document href="data.json" method="json">
  <xsl:sequence select="array { /catalog/cd ! map { 'title': string(title) } }"/>
</xsl:result-document>

See JSON output for method="json", and Producing XML output for the full set of xsl:output controls a result document accepts.

You cannot read what you are writing

Within one transform you may not open a result document at a URI and then doc() it back in — the result does not exist until the transform finishes. Splitting a job into a pipeline (write in pass 1, read in pass 2) is the way around it; on a streaming run, each result document is still written eagerly as it is encountered.

Validity and grouping

xsl:result-document is not allowed where it would sit inside a temporary tree being constructed — it must execute as an instruction, not as part of building a node to return. In practice: call it from a template or xsl:for-each body, not from inside an xsl:variable that builds an element.

Where to go next