[LEAPSECS] Do lawyers care (know) about leap seconds?

Steffen Nurpmeso sdaoden at yandex.com
Wed Oct 1 07:29:00 EDT 2014

 |Warner Losh <imp at bsdimp.com> wrote:
 |> No. The basic point is that people are ignoring the standard because it
 |> is hard to implement.
 |Given the perpetual arguments on this list, I am not surprised by the
 |reaction of the people participating in the UK consultation: techies
 |should buckle down and implement it properly.


 |But my experience in the IETF is that it is normal for engineers to work
 |around or ignore awkward requirements when the cost of complying is too
 |high. See the thousands of IETF documents that never lead to a deployed


 |And successful standards usually follow a successful implementation,
 |rather than the other way round, mainly because there's no substitute for
 |practical experience when it comes to ironing out the interop and
 |deployment difficulties.

 |So I wonder how to effectively communicate the surprisingly large effects
 |that seemingly small technical details can have on the success or failure
 |of a standard. Especially to non-technical people who are rightly
 |impressed by the fondleslab in their pocket and wonder, if phones can be
 |so smart, why is time so dumb? And to technical people who have less
 |experience of the mind numbing futility of standards development.

It would be much easier to ask POSIX for a CLOCK_TAI
clock_gettime(3) if a TAI clock would be easily accessible on
end-user systems.  Unfortunately NTP doesn't include an 8-byte TAI
field in their packet but only send out UTC.  Ironically NTP
primary servers are defined as (RFC 5905, 2.):

  A primary server is synchronized to a reference clock directly
  traceable to UTC (e.g., GPS, Galileo, etc.).

That is, to my understanding, that atomic clock signals are used
by NTP but then this TAI is thrown away instead of being delivered
as a regular part of NTP.

I think it would be an immense improvement if TAI would be
delivered as part of NTP and made available to normal user-space
programs via a new CLOCK_TAI.  Even better would be an additional
CLOCK_LEAPDRIFT that simply returns the current (at the time of
the system call) relative distance.

This approach would satisfy all parties: humans can continue to
enjoy the cultural achievement of a clock that exactly describes
their home planet, and engineers can use TAI for satisfying
airplane schedule calculations for businessmen.


More information about the LEAPSECS mailing list