PR9631 shows a case where vbasetype can be a TYPE_STUB, and since we were
not accounting for it, we crashed in TYPE_FIELD_BITPOS:

static int
gnuv3_baseclass_offset (struct type *type, int index, const bfd_byte *valaddr,
			CORE_ADDR address)
{
...
  vbasetype = TYPE_VPTR_BASETYPE (type);
  vbasetype_vptr_fieldno = get_vptr_fieldno (vbasetype, NULL);

  if (vbasetype_vptr_fieldno >= 0
      && TYPE_FIELD_BITPOS (vbasetype, vbasetype_vptr_fieldno) != 0) <<< here
    error (_("Illegal vptr offset in class %s"),
	   TYPE_NAME (vbasetype) ? TYPE_NAME (vbasetype) : "<unknown>");

This patch handles it, and adds new testcase.  Does it look like the
right thing to do here?  get_vptr_fieldno also always goes through
check_typedef right at the top.

OK?

-- 
Pedro Alves