Michael Kay wrote:

http://snelson.org.uk/~jpcs/higher-order-functions.html

Any reason why the subtyping relation is invariant for the argument
types instead of being more generally contravariant?

On a side note, am I right to feel that the evolution of XSLT lies
nowadays mainly with XPath? If so, it seems to me defeating the purpose
of having a transformation language being able natively to process its
programs (XPath is opaque for templates).