[ntp:hackers] ITU and Leap second elimination

Warner Losh imp at bsdimp.com
Tue Oct 1 15:14:41 UTC 2013


On Sep 30, 2013, at 3:14 AM, Mike S wrote:

> UTC, with leap (or anti-leap?) seconds, is monotonic (and does not exhibit discontinuity).
> 
> Implying it is not monotonic is simply admitting that one does not understand the proper representation of UTC.

Strictly speaking, you are correct. However, the problem comes from the non-uniform radix of UTC. That's the real underlying problem. For systems that can't handle it, the most common failure is a non-monotonic time because a non-uniform radix cannot be easily represented in encodings that assume a uniform radix. See, for example, POSIX time_t, where leap seconds explicitly do not exist. Or to NTP time stamps, for that matter, where only through out of band data can you tell you are in a leap second.

So while it is not pedantically speaking correct, there is a non-uniformity around leap seconds, and that non-uniformity is the root of many problems. "Non-monotonic" captures the essence of the problem and is a term more people are familiar with, but is not the pedantic description of the pathology.

Warner




More information about the hackers mailing list