RE: XSLT2: Keys vs. xsl:choose to distinguish groups

Why not use apply-templates to despatch based on the first member of the
group:

<xsl:for-each-group....
  <xsl:apply-templates select="current-group()[1]" mode="process-group"/>

<xsl:template match="A" mode="process-group">
  <xsl:for-each select="current-group()">
    ...

Michael Kay
http://www.saxonica.com/
 

> -----Original Message-----
> From: Yves Forkl [mailto:Y.Forkl@xxxxxx] 
> Sent: 15 March 2007 12:31
> To: xsl-list@xxxxxxxxxxxxxxxxxxxxxx
> Subject:  XSLT2: Keys vs. xsl:choose to distinguish groups
> 
> Hi list members,
> 
> several weeks ago, your much appreciated help enabled me to 
> tackle my grouping problem. Now I am thinking about some 
> refactoring, so I'll briefly resume it before asking you 
> again for help. It's basically a classic heading-type 
> grouping problem, but with some additional difficulties. From 
> a flat structure like
> 
> <A/>
> <x/>
> <B/>
> <x/>
> <y/>
> <C/>
> <x/>
> 
> I am building this hierarchical structure:
> 
> <part_A>
>    <item_specific_to_A/>
>    <A/>
>    <x/>
> </part_A>
> <part_B>
>    <B/>
>    <x/>
>    <item_specific_to_B/>
>    <y/>
> </part_B>
> <part_C>
>    <C/>
>    <item_specific_to_C/>
>    <x/>
> </part_C>
> 
> This is to say, each group needs specific processing. 
> Currently I am using xsl:choose to recognize a group by its "leader":
> 
> <xsl:for-each-group select="*" group-starting-with="A|B|C">
>    <xsl:choose>
> 
>      <xsl:when test="current-group()[1] = A">
>        <part_A>
>          <item_specific_to_A/>
>          <xsl:apply-templates select="current-group()"/>
>        </part_A>
>      </xsl:when>
> 
>      <!-- xsl:when statements for the other two cases left out -->
> 
>    </xsl:choose>
> </xsl:for-each-group>
> 
> <!-- template rules for copying the individual elements left out -->
> 
> 
> My question is: Is it feasible (and reasonable) to avoid 
> using xsl:choose here but instead select a group's treatment 
> based on a key (because some groups might need the same kind 
> of processing)? I think of having just a single call to the 
> key function which would then dispatch the call to one out of 
> a list of user-defined functions or named templates, which 
> would finally return the resulting sequence.
> 
> Am I dreaming here of a control flow "too dynamic" to be 
> legal in XSLT2? 
> If not, what could the key definition and function call look like?
> 
>    Yves