I left the new nonoverloaded API using timevals, because that's what
xntpd currently uses.  We can change the API to use timespecs if you
want, but it means downgrading them to timevals in userspace.

I'd prefer to do whatever xntpd/ntpd maintainers think is best.
How about I ask Harlan Stenn?


>BUT...
>
>To do the copy or assignment, it has to load a pointer to the softc's
>timeval into a register... so rather than loading two timeval *'s
>into registers, just do one.


But because gcc can't do good alias analysis, it cant tell you didn't
change sc.  So you have to recompute the pointer into the softc each time.
It's a micro-optimization, but it can go if you really want.