The patch works for me, and I agree the test_xmlrpc is an xmlrpc issue.

Perhaps unrelated to this issue, but I think it makes this whole unicode
getargs situation fragile: I could not understand why the 'z' case (on
the switch where this patch applies, starting on line 879 on the patched
file) does a PyString_Check(arg) and a PyUnicode_Check(arg), while the
corresponding section on case 's' (line 813) only does PyUnicode_Check(arg).

Is this really an inconsistency? Maybe this should go into an issue to
cleanup this switch, according to the comment at its beginning:

	/* XXX WAAAAH!  's', 'y', 'z', 'u', 'Z', 'e', 'w', 't' codes all
	   need to be cleaned up! */