Two transformations on same input data

Belch · December 2nd, 2011, 06:24 AM

Hi all,

I have some pretty flat input xml, for example:

Code:

<root>
   <person1name>Bob<person1name>
   <person2name>Fred<person2name>
   <person3name>Harry<person3name>
</root>

I have an xsl file which contains a template called "tidyup", this improves the structure, e.g.:

Code:

<root>
   <people>
      <person id="1">
         <name>Bob</name>
      </person>
      <person id="2">
         <name>Fred</name>
      </person>
      ..and so on

I then have a "main" xsl file which is called by a server, this is the stylesheet used to transform the xml from one format into another.
My aim is for it to first call the tidyup stylesheet to improve the input xml structure, and then transform the tidied xml into a completely new format.

So far my "main" stylesheet looks like:

Code:

...
<xsl:include href="tidyup.xsl" />

<xsl:template match="*">
   <xsl:copy>
      <xsl:copy-of select="@*" />
      <xsl:apply-templates />
   </xsl:copy>
</xsl:template>

<xsl:template match="root">
   <xsl:call-template name="tidyup" />
</xsl:template>
	
<xsl:template match="/">
   <newxml>
      <xsl:value-of select="//name[.='Fred']" />
   </newxml>
</xsl:template>

As you can see from above, I am trying to output any "name" nodes with the value Fred.
However, the second template (matching to "/") seems to be trying to run against the original xml, so I would need to look for "//person2name" instead.

Please could someone offer any advice as to how I can get the second template (matching to "/") to run against the tidied xml (i.e. the result of the first template, matching to "root" and calling "tidyup").

Any help appreciated, thanks.

mhkay · December 2nd, 2011, 06:38 AM

Splitting a complex transformation into a sequence of simpler transformations is an excellent design technique (usually called 'pipelines') so you're on the right track.

There are two ways of implementing a pipeline of transformations: single-stylesheeet and multi-stylesheet. With a single-stylesheet approach (your current approach), you need to use modes, one mode for each phase of processing. Template rules for the first mode should say mode="phase1" and use apply-templates mode="phase1", etc for subsequent phases. Capture the result of each processing phase in a variable (can be local or global, for example

Code:

<xsl:variable name="phase1output">
  <xsl:apply-templates select="/" mode="phase1">
</xsl:variable>
<xsl:template match="/">
  <xsl:apply-templates select="$phase1output" mode="phase2"/>
</xsl:template>

With a two-stylesheet approach you need some external logic to cause the second stylesheet to be applied to the output of the first (or in Saxon you could use saxon:next-in-chain in xsl:output). The external logic might be your own Java code, or Ant, or XProc, or a shell-script, or a framework such as Coccoon or Orbeon.

The multi-stylesheet approach needs more coordination but gives you much more flexibility in reusing code and constructing pipelines dynamically.

Belch · December 20th, 2011, 11:00 AM

Thanks for the info Michael, I haven't had chance to try this out yet so when I get around to it I may have some more questions, but this should point me in the right direction.

Belch · April 30th, 2012, 06:38 AM

Hi again, I have finally revisited this and am having an issue with the initial "phase1" transform.
My stylesheet is now along the lines of:

Code:

<xsl:include href="tidyup.xslt" />
	
<xsl:variable name="tidyup_output">
	<xsl:apply-templates select="/" mode="untidy" />
</xsl:variable>
	
<xsl:template match="*">
	<xsl:apply-templates select="$tidyup_output" mode="tidied" />
</xsl:template>

The output XML was blank, so I tried changing the select attribute of the tidyup_output variable apply-templates command to "/|*", which then just output the XML tag values, and not the tags as well.

I have change the content of the "xsl:template match.." template to just output the value of the variable "tidyup_output" to see exactly what it is doing, and get the same results as above, so it appears the issue lies with the initial tidyup transformation, which is handled in an included stylesheet.

The included "tidyup.xslt" stylesheet has a template command:

Code:

<xsl:template match="*|/" mode="untidy">

Previously on the tidyup stylesheet, I just named the template "tidyup", and called this in the main stylesheet, and the XML tags were copied along with the data inside them.

This must be a difference between apply-templates and call-template but I can't see how to copy the XML tags as well as data - any help would be appreciated on this.

Thanks,

mhkay · April 30th, 2012, 06:59 AM

Sorry, you haven't supplied enough code to allow a reader to find your bug.

When you say it outputs "the tag values", I assume you mean that it outputs the text nodes only - that is, the content occurring between the tags, which would more correctly be described as the non-tag values. The normal reason for this is that none of your template rules have matched any element nodes, and the most common reason for this is that you haven't taken namespaces into account.

A good diagnostic technique at this point is to display the value of the variable $tidyup_output (using xsl:message and xsl:copy-of, or using your preferred XSLT debugger) so that you can determine which phase of processing contains the error.

Belch · May 3rd, 2012, 05:25 AM

Hi Michael,

I went back through your earlier post and over my code again and have managed to get it working now. I think my template match attributes were incorrect but I have sorted it now.

Thanks for your help with this.