>Since API changes have been mentioned in this thread, let me just note
>that POSIX.1b timers can be set with either absolute or relative
>triggers.  

That's cool.   But it's not really the $64k question.
Does POSIX.1b  also  specify distinct timers which decrement
in  ``real time''  and  ``current time of day'', respectively?

The issue at hand is which rate of time to measure the sleep with
(when settimeofday() jumps the clock and makes the two distinct),
rather whether the encoding of the sleep time is relative or absolute.

I do think that relative triggers makes most sense for `real time',
and absolute triggers makes most sense for `current time of day',
but I don't know what POSIX.1b says about that.

If the POSIX.1b does not address the separate-time-flow problem,
then to address that issue, we still need to to invent an new API.