Re: junit test... for xslt2?

A couple of remarks here.

   If we speak about the result of a "whole transform", I tend to
see it as defined by an input and an output, so the result can be
tested against an expected output.  As a few values can not
depend on the input (like generate-id()) we need a mechanism to
compare two documents (where we can say: "do not compare text
nodes with whitespaces", or "do not compare the value of the
attributes 'id'").

This is where I'm hoping 'structure checking' is possible
using xpath only? The poor mans xml diff. I wonder how
well Schematron fits this model.

But then this is more "functional testing" than "unit testing".

For unit testing, I'd say one component has only a few
responsibilities on the result it produces.  It does not always
define its whole output.  Some parts are the responsibility of
other components (like called functions or other template rules).
And in unit testing, I think assertion expressions (what I think
you mean by "you just need to execute XPaths against it") is
really more convenient than comparing two entire trees.

This backs up your idea of keeping templates 'simple' or single
purpose? Surely the difference is simply one of 'level'?
Test one template vs test a tree of templates?

   Here is a simplified exerpt of a test suite for the EXPath HTTP
Client:

     <t:call function="http:send-request">
        <!-- some param here... -->
     </t:call>
     <t:expect test="count($t:result) eq 2"/>
     <t:expect test="$t:result[1] instance of element(http:response)"/>
     <t:expect test="$t:result[1]/xs:integer(@status) eq 200"/>
     <t:expect test="$t:result[2]/*">
        <pass>...</pass>
     </t:expect>

--
Dave Pawson
XSLT XSL-FO FAQ.
http://www.dpawson.co.uk