This test doesn't work exactly the way Leo has envisioned it (creating
a special .CCLASS_* value for word boundaries), but it demonstrates
how I think PIR programmers will typically locate word boundaries.
Sent as a patch rather than applied directly to give others an
opportunity to review and comment.
Pm