[time-nuts] leapseconds, converting between GPS time (week, second) and UTC

[Martin Burnicki]

Fiorenzo Cattaneo wrote:
> Yes of course you are right. NTP timebase (based on UTC with an epoch
> of 1900-01-01 00:00:00, or its representation in Unix time - seconds
> since 1970-01-01 00:00:00) is timezone independent.
> 
> I just wanted to make the point that in the IT world (I've worked for
> Amazon, Microsoft, Twitter and now I work for another major IT
> supplier) we keep spending billions of dollars to maintain really
> crappy and cruddy libraries which deal with timezones, and still get
> plenty of bugs every time.
> Not to mention NTP issues when leapseconds are addded

It's not an NTP issue. NTP only forwards the leap second announcement to
the kernel, if the kernel supports this. It's up to the kernel how a
leap second is handled. Most *ix kernels just step the time back by
default, which is confusing for applications that haven't been designed
to cope with such steps.

Except for latest Windows server versions, Windows doesn't care about
leap seconds at all, and the system time simply off by 1 second after a
leap second event. This offset persists until some time synchronization
software corrects it.

> (the only
> company which handled the last leap second correctly was google, as
> they slowly slewed at the rate of 1s/hour).

This is just a hack. During the slewing the time is off by up to +/- 0.5
s, or even up to 1 s depending on the kind of slewing. This can mess up
applications that require a very accurate absolute time.

Martin