Bravo!  Steve Thackery wrote a tremendous synopsis.  I'm saving it
too.  A few clarifications just in case:

> I'll just summarise that again, as a sanity check: a mark represents a
> logic 1 and is a negative voltage, a space represents a logic 0 and is
> a positive voltage.  A start bit is a space (positive), a stop bit is
> a mark (negative).

  Just note that Mario mentioned that his marks (stop) were positive and
his "space"s negative.  The UART or MCU sees things this way.  It
connects to the line via an inverting transceiver which inverts incoming
and outgoing voltage levels.  On the RS-232 line then, marks are
negative and "space"s positive.  OTOH, the "status" lines are positive-
true on the RS-232 line and inverted as the UART/ MCU sees them.

  Now, according to the RS-232 standard, in a stable state the RS-232
(AKA EIA-232) line is at least 3V either positive or negative, but not
more than 15V (or is that 18V?).  That is what the 1488/ 1489
transmitter/ receiver chip, or any other equivalent, provides.  In
practice, designers of peripheral devices, e.g. mice, frequently cheat
and use a mark of 0V and a space of 3 to 5V.

  The 1489 line receiver threshold unless its adjustment feature is
used, is at 1.5V, so these voltages are fine for non-critical (short-
range) applications.  This is quite convenient, as you see 0V is
interpreted as a marking or idle condition allowing hot-switching of
circuits (minimizes glitches) and all sorts of other tricks.  On the
status lines, an open or grounded line is OFF.

  This leads to mention of the "break" condition.  Where unipolar
current loop switching was used (in those Telex machines), mark meant
current on and space meant current off.  The original meaning of the
words in fact relates to paper tape markers using an ink roller.  A
telex machine in continuos mark sits quietly (sort of!) idling.  If the
line "breaks" however, it makes a terrific racket as without its magnet
holding at all, the mechanism continuously acts on dummy characters.

  This is used as a fault indication.  On UARTs, it is detected if the
stop bit is absent and most implementations thereafter lock out until an
apparent stop bit eventuates.  This then is the "break" condition and
there are deliberate ways of sending it.  You do have to be careful of
interpretation of the character detected (NUL) as the break was being
received.

> It is not uncommon to transmit two stop bits.  Interestingly, in this
> case you can configure the receiver to expect either one or two and it
> will still work.

  In most implementations, the UART is set *as a whole* for one or two
(or 1¸) stop bits.  While the extra stop bits were needed for the
mechanical implementations, UARTs only look for the first, so this
setting actually affects only the transmitter; the receiver sees the
extra merely as an inter-character delay.  If you expect any problems at
all, two stop bits is a good setting as, should the receiver get out of
synch due to noise pulses during a burst of characters, the extra delays
will pull it back into synch within about 8 "bad" characters at the
most.

> You might be interested to know that the worldwide telex standard uses
> a five bit code, with a stop bit one-and-a-half bits long!  This gives
> 7 1/2 bits per character.

  I have a SAGEM electronic telex, hopefully well-covered, in my
repository.  AFAIK, the Telex service became obsolete about 8 years ago.
(The FAX machine killed it!)  Some variants still operate on HF radio
though, and Amateurs still play with it.  Anybody know of residual
land-line use?

  I can imagine the Telex version of the recent Morsecodian thread...

Wagner Lipnharski wrote:

> The receiver needs to save or do something with each received byte, it
> demands some time, and normally this is done right after sensing (and
> during) the stop bit time.  Old and slow machines sometimes needs an
> extra time.

  That's talking about *machines*, but the significance is that "bit-
banging" implementations on MCUs may also need that extra time,
particularly at high baudrates.  Interrupt-driven and hardware
implementations generally don't.

> It doesn't make sense to program the receiver with more than one stop
> bit, since with ONE STOP BIT it will works with data arriving with 1,
> 1¸ or 2 stop bits, always.

  It's not significant.  You aren't programming the number of stop bits
in the receiver, only the transmitter.
-- 
  Cheers,
        Paul B.