I find the very idea of XSLT simply pointless. It is fundamentally structured around templates and doesn't accept anything beyond XML (even in XSLT 3.0, where you can have an initial template but can't get rid of it), making it cumbersome for general programming. It is also very verbose so it is not suitable for scripting as well. For XSLT to be useful you need a complex enough transformation (so you actually want templates) and nothing else (so you can avoid general programming); that use case would be pretty rare.
Sounds like declarative pattern matching which sounds just great to me [0].
I won't deny that XML sucks to write. But I'm not yet ready to throw the baby out with the bathwater.
Maybe we keep using XML as a machine-readable serialization format (and XSLT as serialization of pattern-matching data transformations!) and people use native data structures in their own programming languages. Or we all switch to S-expressions.
Pattern matching is great, but probably not if everything is pattern matching. Also I'm okay with XML as long as it doesn't power an urge to make everything XML and everything connected via the semantic web (lol). XSLT is a remnant of that urge and should have been a non-XML-based language. (The same criticism applies to, e.g. RDF vs. Notation3.)
