2 ms·
The way Google smears leap seconds seems very practical. I can't think of a time when I needed better than 1/86400 sec accuracy/precision. Being off by a second
by karmakaze 6y ago
The way Google smears leap seconds seems very practical. I can't think of a time when I needed better than 1/86400 sec accuracy/precision. Being off by a second would be very noticeable. If you want/care, use a call that doesn't smear as that's the rarer requirement.
- ncmncm 6y agoIt means that during many hours the second is smeared out over, they are hundreds of milliseconds off of (almost) everybody else in the world -- in many cases even when others are also smearing. Since the only value in tracking a Standard time reference is to stay synchronized with others also following it, being hundreds of milliseconds off for hours on end is not much different from hours of downtime.
- karmakaze 6y agoSo then what I want is for everyone to smear the same way and stay in sync.
- ncmncm 6y agoWhat you want, then, is a Standard. Smearing is decidedly non-Standard. The Standard we have says to insert a leap second, or to be offline during the (short) period containing the leap second, if you can tolerate the downtime.
- karmakaze 6y agoStandard is what's agreed upon, nothing more or less--standards change.
- ncmncm 6y agoAnd, the best possible change would be to do away with leap seconds. Making everybody do the most complicated and difficult thing imaginable -- or choose between several of those, and have no correct choice -- would not be an improvement. The number of applications that cannot tolerate two seconds' pre-scheduled downtime every other year, or one minute each century, is much lower than some people imagine. They will get that, but unscheduled, regardless. The number of applications that need to be exactly synchronized to the astronomical position of the sun in the sky is even smaller. The number of those that would benefit from a more complicated time Standard is exactly zero.
- karmakaze 6y agoI completely agree that leap seconds were a stupid idea to put into general purpose machines. From where we are though, consistent smearing seems the least disruptive, and long term doesn't suffer from drift. Having a mix of machines that respect/ignore leap seconds could accumulate to more than 1 sec differences.
- ncmncm 6y agoAs it is I know of at least three completely different smearing regimes, all incompatible with one another and extremely complicated to implement. One thing is certain: anything complicated and difficult to implement will, in a large fraction of systems, not be implemented, or be implemented incorrectly, leading to needless hundreds-of-milliseconds disagreements between systems about what time it is. Drift is absolutely harmless almost everywhere, and easy to correct for in the rare, isolated cases where it is not. The only cases I know of where it matters involve spacecraft, and those certainly are not using UTC internally.