> My recollection isx that  POSIX rules forbids Bernstein's idea.

That is my recollection too, but the same POSIX section mandated a
leap year where there was none, so one has to doubt the level of
thought that went into that section.

> The kernel time would be about 30 seconds out form wall time, and will
> continue to drift further out of step, as (given the slowdown in
> rotation fo the Earth) TAI and UTC slowly diverge.  There is a
> nonzero user community where this is absolutely unacceptable.

My gps and the whole gps system for that matter is 13 seconds (I
think) out from wall-time.  The last observed UTC-GPS offset is simply
stored and added to any displayed values.  The whole thing is quite
user-transparent.

The fact that the kernel time gets wrenched around is a huge pain in
the butt for my user-level mapping code.  I need to structure
everything to keep interval-time and wall-time just on the off-chance
that an interval spanned one of the two leap-second points and the
leap second did occur.

To me it seems like ntp could just as well appended a line to a
leap-seconds file as to step the kernel's time.

-wolfgang
-- 
Wolfgang Rupprecht    <wolfgang@wsrcc.com>     http://www.wsrcc.com/wolfgang/
Coming soon: GPS mapping tools for Open Systems. http://www.gnomad-mapping.com/