Slashdot Mirror
Leap Second May Be On the Chopping Block (ieee.org)
szotz writes: The days of 61-second minutes may be coming to an end. The World Radiocommunication Conference is meeting for nearly the entire month of November, and one of the hot-button issues is what to do about the leap second. The addition to UTC is supposed to keep atomic time aligned with Earth's rotation, but past leap seconds have caused server crashes, and some are worried that future problems could be even worse. Going into the conference, it doesn't look like there's much of a consensus on what to do. One official is expecting weeks of debate.
Midnight is always 00:00. There is no such hour as 24:00, but alas the ID10Ts at apple think there is and put one into Yosemite.
This is a large part of the problem. More often than not these issues stem from people trying to roll their own time handling code / int'l address code / i18n / etc rather than using one of the standard (and well-tested) libraries available in their language.
Time is hard to get right, addresses are hard to get right, i18n is hard to get right. Don't roll your own. There's a thousand edge cases you haven't accounted for.
How close do we need to be to the astronomical time?
My argument would be that we don't. Who cares if the seasons shift to different months over a few hundred years? It's not going to be important within the lifetime of anyone reading this. If it snows in June instead of December I just don't see that as an actual problem. Pick an epoch and count from there. No need to keep the seasons matched to the calendar for eternity. In fact if we leave Earth, doing that will quickly become highly impractical.
The problem with leap seconds vs. leap days is that leap seconds do not follow a regular pattern. It's fairly simple to write code that follows the leap year rules: February 29 occurs when (year mod 4 == 0 and year mod 100 != 0) or year mod 400 == 0
The timing of leap seconds, on the other hand, are chosen on a case-by-case basis by some standards body (IERS) and announced, usually with only about six month's notice. Thus hardcoded rules for leap seconds are not a good idea, and you have to have some means of distributing leap second announcements to all systems where clock accuracy is critical. Either that or you accept a one-second margin of error, don't count the leap second and correct the clock later.