> In C9X, int16_t *MUST* have exactly 16 bits of precision, but sizeof
> can be anything >0.  (Well, any *integer* greater than 0.)

What does one do if an architecture simply doesn't have 16-bit addressability?
Does the compiler vendor not define int16_t, or is the compiler obligated to
emulate non-addressable quantities?  (I'm thinking in particular of the DSP
chip which has byte and 32-bit-word addressability, and that's it.)

Is it time for some C9X committee members to wake up with a crashed disk head
in their bed and a note reading "All the world's NOT a VAX"?   ;-)