Quentin wrote:

> 1---(7)-(8)-(9)
>      |   |   |
> 2---(4)-(5)-(6)
>      |   |   |
> 3---(1)-(2)-(3)
>      |   |   |
>      |   |   |
>      4   5   6

> You make pin 1 high, then read 4, then 5, then 6 to see which one is
> high, then you make pin 1 low and 2 high and read 4, 5, 6 again, etc.

  That's not *quite* correct.

  Firstly, pins you are using as an input must be given a known state.
The PIC has built-in pull-ups on port B which you may care to use or you
can use external pull-up resistors for all keyboard lines.

  Note that for protection against shorts to ground, which are rather
more likely than shorts to supply, it is usual to use pull-ups rather
than pull-downs as in a short to ground (being in effect the same as a
switch closure to ground) the current will be safely limited by the
pull-up.

  Secondly, you "strobe" each row (or if you wish, column) in turn by
making it an output and pulling it *low*, but the rows you are *not*
strobing, you do *not* pull high, but rather open-circuit them by making
them inputs.

  If both rows and columns are part of port B and you elected to use the
internal pull-ups, then the non-selected rows will be pulled up by the
pull-ups as well as the columns but this does not matter (in fact, you
don't really want any inputs to float).  The point is that by this means
multiple switch closures in any given row will *not* cause high and low
outputs to conflict.

  When one row, say pin 1, is low, then any column may be pulled down by
a corresponding closed switch on that row, while all other columns are
pulled up by their pull-ups.  By pulling each row low in turn, and
reading (generally in one I/O operation, using a shift operation to
test each in turn) all columns, all the switches can be read.

  You may note a couple of cunning tricks which may or may not be
useful.  One is to hold *all* rows low whilst reading the columns,
waiting for *any* closure before strobing them individually.  This makes
it very simple to perform a status check on the keyboard as a whole, it
can even be done in SLEEP using the Port B interrupt-on-change.

  Also, a quite fast way of finding a single key closure (i.e., does not
work if more than one key is pressed), is to pull all rows low and read
the columns, then pull all columns low and read the rows.  As long as
you only see one column and one row pulled down as a result, this
uniquely indicates the closure.

  Note that the PORT register is kept clear to do this, (do not use
read-modify-write instructions on it) and only the TRIS register is
changed (use the TRIS instruction!) but the PORT is read each time.
--
  Cheers,
        Paul B.