I don't know if this is a limitation of the spec or not, but it makes it
difficult to use tc->subnames for a union.  For example, one of the unions
in the everything.idl test included with ORBit2 looks like this:

        union FixedLengthUnion switch (char) {
        case 'a': long x;
        case 'b': char y;
        case 'c': 
        case 'd': boolean z;
        case 'e': FixedLengthStruct v;
        };

For this union, we have the following information in the typecode:
  tc.sub_parts = 5
  tc.subnames = [ "x", "y", "z", "v" ]
  tc.subtypes = [ long, char, boolean, boolean, FixedLengthStruct ]
  tc.sublabels = [ 'a', 'b', 'c', 'd', 'e' ]

So the subnames array doesn't have any indication of the "case 'c':"
branch.  There doesn't seem to be any indication that the 'c' branch even
falls through.  I would have expected the subnames array to look like this:
  tc.subnames = [ "x", "y", "z", "z", "v" ]

Does this sound sensible?  What does the spec say about it?