[ntp:questions] Re: Time set backward

Garrett Wollman wollman at csail.mit.edu
Sat Jan 7 22:29:33 UTC 2006


In article <dpohfm$k5p$1 at dewey.udel.edu>,
David L. Mills <mills at udel.edu> wrote:

>If POSIX has no way to discipline the clock frequency, the clock
>frequency is not disciplined and the platform not reliably
>synchronized.

POSIX makes no claim about whether the clock has any connection
whatsoever with reality.  POSIX must provide for a wide variety of
environments, including non-networked machines with free-running
clocks that may be set from a repair technician's wristwatch once
every three years.  As I said, it's only a fluke that the
clock_settime() interface is even in POSIX -- prior to the Real Time
Extensions work there simply was no way to set the clock at all, just
as there still is no way to create a new user, reboot the system, or
perform any number of other administrative operations.  POSIX is just
an API standard; it does not try to include every possible function an
operating system needs to have.

I tried to interest people in an interface which would at least allow
applications to *tell* whether the clock was synchronized with
anything, but unfortunately timekeeping in POSIX is a deep quagmire,
and any time anyone brings up any changes to the timekeeping
interfaces, they are immediately overwhelmed with demands to fix the
Y2038 Problem, the Leap Second Problem, the struct timespec problem,
the Time Zone Thread-Safety Problem, and innumerable other pet
defects.  I am certainly not going to spend all my spare time dealing
with that, and I don't know anyone else with the stomach for it.

-GAWollman

-- 
Garrett A. Wollman    | As the Constitution endures, persons in every
wollman at csail.mit.edu | generation can invoke its principles in their own
Opinions not those    | search for greater freedom.
of MIT or CSAIL.      | - A. Kennedy, Lawrence v. Texas, 539 U.S. 558 (2003)




More information about the questions mailing list