ElementTree() causes the node given to *move* to a new document, and
tostring() was (unnecessarily) uses it. This should fix Henrik's
xpath bug, and at least paper over Paul's XSLT segfault bug, for
which I've added a test. I say paper over as the fundamental problem
of mixing nodes from XSLT result trees with anything else remains.
I should think about what happens with ElementTree() in general when
it's fed an element. I wonder what (c)ElementTree's behavior is in
this respect, though I fear it actually will happily share the node
between two documents. I will investigate.
--HG--
branch : trunk