Re: XPath support for using multiple ixml implementations

Hello,

I haven't actually done this in XProc, but I've done something pretty
similar. One of my test pipelines calls iXML over a sequence of inputs.
Half of these are expected to work, the other half to throw errors. The
processor dispatches them to p:invisible-xml wrapped in p:try, and does the
right thing with the results, whether outputs, or errors. The pipeline
output registers whether the good inputs worked and the bad inputs produced
errors. Discrepancies are reported.

This is nice to have, not only to sketch the boundaries of bad-vs-good in
the syntax, but also because it is easy to throw in a different toolchain
(iXML processor, XSLTs) and see that the results are still good - so, I get
regression testing around my defined inputs and 'source of truth' is where
I want it.

Since a goal for me is an unambiguous grammar, avoiding the problem of what
to do with ambiguities is something I don't have to think about.

But if it were something I'd be willing to accept - and especially if
coding for others to use (include future-self others), I'd also build a
test suite of instances that demonstrated the 'intent' apart from the tools