back
24 comments
"But Britain, along with Canada and China, would like to keep the current keeping system, arguing that, in the 40 years that leap seconds have been gracefully inserted in our midst — most recently in 2008 — there have been no problems to speak of, and the worriers have greatly exaggerated the potential for havoc. Remember Y2K?"

No problems? He must not be aware of the 2008 adjtimex() bug in the linux kernel: https://lkml.org/lkml/2009/1/2/389

This bug brought down nearly 100k systems for my employer at midnight UTC, worldwide, in unison. It was not a fun evening.

Hmm, the proposal to do away with the leap second seems yet another instance of humanity's practice of taking decisions that are expedient in the short term but will create problems in the long term (sun rises at noon in a few thousand years?!)

Every computer in the past several years I've installed (Windows, Mac, Linux) has had NTP enabled. Almost every cell phone ships with "set time from network" enabled. Wristwatches sales are way down => few people set their watches manually.

Keep the leap second! :)

The sun rising at noon in the far future long after we're all dead is not a problem for us, and it wouldn't be a problem for the people living in that future either. On the other hand, software bugs due to unnecessary complexity are an actual problem, now and in the future.
Umm, I think your comment exhibits exactly the attitude I was arguing against. I'm not smart enough to argue conclusively why one should care about what happens after one's death, but this greek proverb moved me a lot.

    A society grows great when old men plant trees
    whose shade they know they shall never sit in.
Example: http://www.gardinclothing.com/blog/?p=31
I think not being able to guarantee that there always 7200 seconds between 11:00pm and 1:00am the next day could be a worse problem than having the sun rise at noon in a few thousand years.
why is this different from leap years?
Leap years are defined by a known algorithm: If the year is evenly divisible by 4, then it's a leap year, unless it's divisible by 100, unless it's divisible by 400. This defines all future leap years. Because of this, we /do/ know how much time will be between two particular dates. Leap years can be calculated in advance as they represent a well-known, fixed skew between the gregorian calendar and the actual calendar.

Contrast this with leap seconds, which represent a variable skew and cannot be predicted by algorithm. Leap seconds are determined by measurement of the solar cycle and are merely announced at 6 month intervals. Because of this we cannot use solar time with seconds-granularity more than 6 months into the future, and a table of past leap-seconds is needed to accurately calculate time in the past. Worth noting: POSIX time functions do none of this for you.

There is another system of time, TAI, which is used for scientific measurements which rely on seconds-granularity at arbitrary intervals. TAI is skewed from solar time and will eventually not match the 24 hour solar cycle.

It seems to me that leap minutes would perform no worse and cause a lot less pain from rarity.
Code that is exercised once in a century (see: y2k and similar timestamp problems) is buggier than code that is exercised once every six months (leap seconds) or in real time (NTP, which given the poor precision of internal clocks, everyone should be running anyway).
I'll pick mild bugs once a century over milder bugs every couple years.
Only that you'll have to correct this more and more in time because earth's second isn't constant (actually this is exactly why they started to use the leap second). So why do it? Just to get a /Universal/ time in sync with Earth?
Nope. Let's get rid of this horrible date-time system we have. I can't wait for gataca or the singularity to take away these crazy time and productivity wasters.
Astronomical observations at small resolutions of time already have to compensate for so many variables that the current leap second system does not help them any. Seriously, look at the formulae and data collected by these guys http://www.iers.org/ and then say "astronomers need leap seconds to point their telescopes".

More likely, somewhere some budget for astronomers will be reduced by the move to a terrestrial source of time.

The author sounds knowledgeable, but the reference to Y2K as an overreaction makes me think he hasn't a clue.

Kenneth, boychik... how much effort do you think it took to make sure Y2K didn't become a memorable problem?

We'll have to reset everything at Y10K anyway.
I'm wondering why people are trying to get the leap second removed from UTC. If people have situations where leap seconds are problematic, why not just switch to TAI which already doesn't have leap seconds?
It's because too many things are defined in terms of UTC; if you start saying, "Okay, let's start using TAI from now on.", now you have the complexity of trying to remember whether you're supposed to use TAI or UTC for this particular protocol and/or data format, so in practice people will stick with UTC there.

On the other hand, removing leap seconds from UTC is trivial - they're inserted manually anyway, so you just stop doing it.

'required' (if only for the sentence starting with "As a time-nut, a small and crazy fraternity") reading: http://queue.acm.org/detail.cfm?id=1967009
Epoch time vs. time of day http://www.ucolick.org/~sla/leapsecs/epochtime.html
Why don't we recalibrate the atomic clocks, that are systematically running late?

I assume that the variations in the Earth's rotations are not predictable, and that it is this impossible to do it exactly, but they could be conservatively slowed down in order to increase the delay between leap seconds.

The earth is slowing down. No matter how you calibrate the clocks, you'll constantly have to change it. At some point the second was accurate to the earth.

And in any case the second is very precisely defined based on physical objects (specifically how long it takes an atom of cesium to oscillate a specific number of times).

Changing it would ripple through all sorts of things, for example the length of objects would change (or more accurately the number we use to measure the length of an object would be different). The number used to represent the length of an object is based on how far light travels in a certain amount of time. Change that time, you change how far the light traveled, which would then give a new definition for how long a meter is.

Then define two types of seconds.

* The reference second, used for the metric system definitions and science in general.

* The real-life second, based on the rotation of the earth.

If you want to keep a Cesium reference for every day time, you can define the second as either 9,192,631,770 or 9,192,631,771 oscillations in a given proportion. X short seconds followed by Y long seconds, with the X/Y ratio decreasing daily. When X becomes 0, increment both numbers and start over.

Are you trolling? Joking?

A: How exactly do you propose changing the speed of millions of clocks worldwide to match your new second?

B: Given that the entire difference is one second every year or so, I guess they could just adjust by one second every once in a while. i.e. you are back where you started.

Why can't we just add a leap minute every century, or a leap ten minutes every millennium?