Menu

Automating Stylesheet Creation

September 7, 2005

Bob DuCharme

Since the early days of XSLT, many have asked whether it was possible to automate the creation of XSLT stylesheets. The general idea of filling out a form or dragging some icons around, then clicking a button and seeing a productive stylesheet generated from your input has always appealed to people. However, the problem of generating working XSLT syntax from the result of someone clicking on pull-down menus and radio buttons has not attracted many takers.

Breaking the problem down into two parts, though, makes it much easier. Step one, the generation of a file like the following from XForms, Tcl/TK, scripting plus an HTML forms interface, or Visual Basic, sounds much more tractable:

<conversion>

  <delete>

    <name>foobarNum</name>

    <name>weather</name>

  </delete>



  <strip>

    <name>email</name>

  </strip>



  <reorder>

    <parent>purchaseOrder</parent>

    <children>

      <name>contact</name>

      <name>shippingAddress</name>

      <name>items</name>

    </children>

  </reorder>



  <wrap>

    <name>shippingAddress</name>

    <wrapper>addresses</wrapper>

  </wrap>



  <attribute2element>

    <name>purchaseOrder/@date</name>

    <elName>orderDate</elName>

  </attribute2element>



  <custom>

    <name>item/@img</name>

  </custom>



</conversion>

If anyone has a successful crack at developing a graphical user interface that creates such a file (and, ideally, reads one, lets the user edit it, and writes out the saved one), let me know and I'll mention it in a future column. Meanwhile, I'll show you how to do step two of the XSLT-generation app: how one stylesheet can create another using the file above (which I've named conversionSpecs.xml) as the source for the generating stylesheet. The tasks that conversionSpecs.xml describes above are somewhat general-purpose for XML transformations, and certainly won't cover all your needs; the stylesheet-generation stylesheet below provides a model for other tasks that you might want to list in a file like the one above and then convert to working XSLT code.

Writing Stylesheet-Generating Stylesheets

The key trick when writing stylesheets that generate other stylesheets is the use of the xsl:namespace-alias element, which I described in further detail in the "Using XSLT to Output XSLT" section of an earlier column titled Namespaces and XSLT Stylesheets. The following shows the beginning of my stylesheet-generation stylesheet. (The complete stylesheet and sample files are available here.) The template rules for the generated stylesheet will have the namespace prefix "wh" in the generating stylesheet, where they're mapped to the dummy namespace "whatever." The xsl:namespace-alias element below tells the XSLT processor to map the "wh" elements to the same namespace as the "xsl" prefix in the generated stylesheet. Because this is the http://www.w3.org/1999/XSL/Transform namespace, this will make the "wh" elements in the generated stylesheet proper XSLT instructions.

<xsl:stylesheet

            xmlns:xsl="http://www.w3.org/1999/XSL/Transform"

            xmlns:wh="whatever"

            xmlns:exsl="http://exslt.org/common"

            extension-element-prefixes="exsl"

            version="1.0">



  <xsl:strip-space elements="*"/>

  <xsl:output indent="yes"/>



  <xsl:namespace-alias result-prefix="xsl"

                       stylesheet-prefix="wh"/>

Other parts of the stylesheet setup above include the declaration of a namespace used for a stylesheet extension element, which I'll cover below, and xsl:strip-space and xsl:output elements to make the generated stylesheet a little easier to read, because generated code is often tough on the eyes.

The generating stylesheet's first template rule creates a template rule in the generated stylesheet that will delete all elements named by name elements in the conversionSpec.xml file's delete element. With the version of conversionSpec.xml shown above, we want it to generate this:

<wh:template match="foobarNum|weather"/> 

The template rule that creates this uses the xsl:element element to create an element named "wh:template" and an xsl:attribute element to create a match attribute for wh:template. It creates the attribute's value by iterating through the delete element's name children and adding each name element's value and a pipe delimiter if another name is coming up.

<!-- Create template rule to delete elements. -->

<xsl:template match="delete">

  <xsl:element name="wh:template">

    <xsl:attribute name="match">

      <xsl:for-each select="name">

        <xsl:value-of select="."/>

        <xsl:if test="following-sibling::name">

          <xsl:text>|</xsl:text>

        </xsl:if>

      </xsl:for-each>

    </xsl:attribute>

    <!-- Created element is empty. -->

  </xsl:element>

</xsl:template>

Once the attribute is created, nothing else is necessary, because as the sample wh:template element above shows, we want to create an empty template rule that does nothing when it finds elements described by the match condition.

The basic pattern of this template rule works for other stylesheet tasks in which one template rule in the generated stylesheet handles multiple elements. The generating stylesheet's next template rule, which creates a template rule that will output the contents of any matched elements without their tags, resembles the one above, with the new part bolded.

 <!-- Create template rule to strip tags. -->

<xsl:template match="strip">

  <xsl:element name="wh:template">

    <xsl:attribute name="match">

      <xsl:for-each select="name">

        <xsl:value-of select="."/>

        <xsl:if test="following-sibling::name">

          <xsl:text>|</xsl:text>

        </xsl:if>

      </xsl:for-each>

    </xsl:attribute>

    <wh:apply-templates/>

  </xsl:element>

</xsl:template>

Once this template rule finishes the match condition listing the relevant elements, it adds a wh:apply-templates element to the result, creating an instruction for the generated stylesheet that looks like this (remember, conversionSpecs.xml only had one name child of its strip element):

<wh:template match="email">

    <wh:apply-templates/>

  </wh:template>

The two template rules we've seen so far from the generating stylesheet follow a common pattern that will be useful in any stylesheet that generates other stylesheets. The next three template rules follow another pattern that is handy when you need to generate a separate template rule for each element named in the conversion instructions as the object of a specific task.

The template rule below creates a template rule in the generated stylesheet for each reorder element in the conversion instructions. The generated template rule will name the element to reorder in its match condition, copy all the attributes verbatim, and then use a series of wh:apply-templates instructions to output the parent element's children in the order specified in the name children of the reorder element's children element:

<!-- Create a template rule for each element

     that needs to be reordered. -->

<xsl:template match="reorder">

  <wh:template match="{parent}">

    <wh:copy>

      <wh:apply-templates select="@*"/>

      <xsl:for-each select="children/name">

        <wh:apply-templates select="{.}"/>

      </xsl:for-each>

    </wh:copy>

  </wh:template>

</xsl:template>  

The second of the three template rules that creates a generated template for each element that needs its own match condition, finds the conversion instruction's wrap elements and uses its two child elements to create a generated template rule that will wrap the named element, when found, with an element named by the wrapper element:

<!-- Create a template rule for each element that

     needs to be wrapped in another element. -->

<xsl:template match="wrap">

  <wh:template match="{name}">

    <xsl:element name="{wrapper}">

      <wh:copy>

        <wh:apply-templates select="@*|node()"/>

      </wh:copy>

    </xsl:element>

  </wh:template>

</xsl:template> 

The last of these three template rules creates a template rule that will convert attributes to subelements when it finds an attribute2element element in the conversion instructions:

 <!-- Create a template rule for each attribute that

     gets converted to a subelement. -->

<xsl:template match="attribute2element">

  <wh:template match="{name}">

    <wh:element name="{elName}">

      <wh:value-of select="."/>

    </wh:element>

  </wh:template>

</xsl:template>

The generating stylesheet's next template rule tells the XSLT processor not to apply any template rules to name children of custom elements, because the generating stylesheet's final template rule will grab them when it needs them.

<xsl:template match="custom/name"/> 

The last template rule is the setup one that gets triggered upon finding the root of the conversion instructions document. It creates the document element of the generated style (the wh:stylesheet element), calls any other necessary template rules in the generating stylesheet, and then does two things. First, it adds a default template rule to the generated stylesheet that will copy any nodes that don't have other template rules specified for them. If your generated stylesheet will copy most of the nodes it finds without changing them, this is a good default template rule to add to your generated stylesheet. But you may want your generated stylesheet to use a different default template rule. If so, this is a good place for it.

<xsl:template match="/">

 <wh:stylesheet version="1.0">



  <xsl:apply-templates/>



   <!-- Default template rule in generated stylesheet. -->

   <wh:template match="@*|node()">

    <wh:copy>

     <wh:apply-templates select="@*|node()"/>

     </wh:copy>

   </wh:template>



<!-- If needed, generate an included

     stylesheet for custom code. -->

<xsl:if test="conversion/custom">



     <wh:include href="custElements.xsl"/>



    <exsl:document href="custElements.xsl" indent="yes">

     <wh:stylesheet version="1.0">

      <xsl:for-each select="conversion/custom/name">

       <wh:template match="{.}">

        <xsl:comment> Add custom code here </xsl:comment>

        </wh:template>

          </xsl:for-each>

        </wh:stylesheet>

     </exsl:document>

   </xsl:if>



  </wh:stylesheet>

</xsl:template> 

The most complex job of the generating stylesheet's final template rule is the generation of template rules for elements and other source nodes that require custom handling. You're lucky if your stylesheet generation system can automatically generate 90 percent of what you need, but you still need to account for the other 10 percent.

The conversionSpecs.xml file has a custom element with only one name child to indicate a source tree node that requires this special handling: img attributes of item elements. The generating stylesheet generates a separate stylesheet named custElements.xsl, where the custom code for such nodes can be written by hand to process any nodes named in conversion/custom elements. The generating stylesheet also adds a wh:include instruction to the main generated stylesheet so that an XSLT processor that executes it uses the template rules from the custom code as well. (Once you've added your customized code to custElements.xsl, keep a copy in a separate directory, because this generating stylesheet overwrites any existing custElements.xsl stylesheets each time it's run.)

The generating stylesheet creates custElements.xsl using the exsl:document extension element, one of the EXSLT extensions to XSLT 1.0 specification that I recently described in the column titled Extending XSLT with EXSLT. (Libxslt supports exsl:document, but for Saxon, use the XSLT 2.0 xsl:result-document instruction instead.) Inside this exsl:document element is a wh:stylesheet element that does the basic setup for the custElements.xsl stylesheet being created. That wh:stylesheet element will have a wh:template element for each custom/name element found in the document. To prevent these template rules from showing up as single-tag empty elements in the custElements.xsl stylesheet, an xsl:comment element creates a stub to be replaced by the actual XSLT logic for those elements.

Running Stylesheet-Generating Stylesheets

    

Also in Transforming XML

Appreciating Libxslt

Push, Pull, Next!

Seeking Equality

The Path of Control

Using Stylesheet Schemas

If all this talk of generating/generated stylesheets is confusing, an overview and diagram of the script that runs them all should make it clearer. (In the sample documents zip file, there is a gentest.bat driver file for Windows machines and a gentest.sh one for Linux machines. Both assume that the free libxslt command-line utility xsltproc is in your path.) The driver script has two lines, not counting the line that shows you the result:

  1. generating.xsl reads conversionSpecs.xml and creates the generated.xsl and custElements.xsl stylesheets based on the information found in conversionSpecs.xml.

  2. The generated stylesheet is tested: generated.xsl gets applied to sampleDoc.xml and stores the result in sampleOut.xml. It deletes the foobarNum and weather elements, strips the tags from any email elements, and executes the other tasks originally listed in conversionSpecs.xml.

diagram showing stylesheet generation workflow

Generating Stylesheets in Your Workflow

How does all this reduce the workload in your office? In the ideal scenario, when a new content type shows up, someone who doesn't necessarily know XSLT creates a new version of conversionSpecs.xml for that content type (perhaps using a GUI tool, as I described above) and then uses generating.xsl to create the stylesheet necessary to convert the new content type to the XML that your system needs. The difficult part is determining the typical transformation tasks that your shop needs and then writing the XSL code in generating.xsl to automate these tasks. The generating.xsl stylesheet shown in this article demonstrates a few automation patterns that should give you a good head start.