> For saxon:serialize, I first need to understand how to invoke 
> Florent's serializer.
> 
> Incidentally, Michael, I am puzzled by one aspect of the 
> design of saxon:serialize - namely the passing of an 
> xsl:output element.
> 
> If I was writing a built-in extension function, I would take 
> advantage of access to the processor internals, and pass the 
> value of the xsl:output name attribute. 

When it's used in XSLT, the second argument can indeed be the name of an
xsl:output declaration in the stylesheet. The other option, of providing a
run-time xsl:output element, is there for two reasons:

* It works in XQuery.

* It allows a completely dynamic specification of the serialization
parameters.

Of course, the specification has evolved in a slightly ad-hoc way looking at
only one user requirement at a time - you generally get a better spec if you
apply a number of minds to the task and try and analyze all the requirements
in advance, hence the suggestion of doing it under EXSLT auspices.

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