UTC with Smoothed Leap Seconds (2011) | Hacker News Reader

UTC with Smoothed Leap Seconds (2011)

cl.cam.ac.uk

11 pointsacqq3 comments

This is much more recent than https://news.ycombinator.com/item?id=9017761 ( TC, TAI, and Unix time (1997) by djb ) and more convenient for most of common use cases.

In short, UTC with leap seconds is no problem at all, as long as exactly 86400 seconds exist in every day (1) for the common calendar calculations and the leap second is "smoothed" by the computer time synchronization code. Google already used this approach (and blogged about it while not crediting Kuhn, as far as I understood).

The common operating systems should just use UTC-SLS for the calendar-related time. Who needs the "real honest atomic seconds" already uses TAI.

More details also here:

https://news.ycombinator.com/item?id=9018367

1) POSIX already specifies that every day has exactly 86400 seconds for "Seconds Since the Epoch" and the current code relies on that:

http://pubs.opengroup.org/onlinepubs/9699919799/basedefs/V1_...

And still no really good justification why general society should have to pay the price of the needs of some specialist systems and astronomers. I'm disappointed that civil time decided that UTC instead of TAI made sense. The least we can do is fix UTC and let specialists deal with it.

And yes, while repeating a second at the end of the day is a terrible approach, it's not a worse sin than leap seconds in the first place. I do hope that slowing down the clock is generally picked up and accepted.

No, the leap second is not the problem at all. UTC has sense since the calendar day is more important for everybody than some ideas of some confused programmers. Who needs "just pure atomic seconds" already has TAI.

All the computer clocks are inaccurate and synchronize sometimes, often more than one second, and nobody worries about that or maintains any tables when his computer adjusted (try yourself, you'd soon see why you shouldn't care about the leap second). Likewise nobody would worry about the leap second were there not too much confusion among the programmers, enough to have the NTP implementation as it was and the people who used djb's or the code from Olsen tz to "account" for the leap second all the time for the calculations that didn't need that.

For some examples of confusion see:

http://stackoverflow.com/questions/7552104/is-a-day-always-8...

The least voted comment is right: "in POSIX, a day is exactly 86400 seconds by definition" (when we ignore the daylight saving time changes of the wall clocks and computers of the whole hour, of course).

Linus agrees it's a good decision:

http://www.wired.com/2015/01/torvalds_leapsecond/

"the POSIX approach puts the pain where it belongs, and hides it from the vast majority of cases that really don’t care. I think it was the right decision."

"The people and computers who really care about “atomic” time tend to be astronomers. For the rest of us – both people and computers – we’re probably better off taking the POSIX approach to it, and just say “who cares”, with a few unlucky people worrying about bugs happening because of the perversities of timekeeping."

Reply on news.ycombinator.com