*Partial* reply for now.  Work beckons!

Donald Riedinger wrote:

> Are we all agreed that you need to sample a mechanical switch for at
> least 20 ms to see if it has actually settled to a different state
> after bouncing?

  No.  In *this* application, debouncing is neither necessary nor in any
way desirable.  If any switch closure is detected at any time, it *has*
been pressed; it was not an accident.

  My thesis was that if it is impossible to be sure whether two switches
were pressed at once and/ or which was first, then it is probably
"fairer" to simply accept a delay *of the order of* 50 µs in the
determination (this being well within the mechanical variances of the
switch design, let alone the human interface) and use a tie-breaking
algorithm that is *scrupulously* equitable.  My suggestions as made,
follow accordingly, and I think already cover most of your comments.
-- 
  Cheers,
        Paul B.