[LEAPSECS] Toasting Unix timestamp 1234567890

Magnus Danielson magnus at rubidium.dyndns.org
Sat Feb 14 08:12:56 EST 2009

Zefram skrev:

> Magnus Danielson wrote:

>> Thus, the TAI-UTC difference was 4.213170 + (40587-39126) x 0.002592s =

>> 8.000082 s.


> Yes. This lets you calculate the number of *TAI* seconds since the

> Unix epoch. There were 63072001.999918 TAI seconds (exactly) in UTC's

> version of 1970 and 1971 together.

Indeed. This is useful for those believing in SI/TAI seconds for as the
second distance of time_t, their reference epoch is inconveniently
offset in this fashion, making them have an offset of 25,999918 s from
POSIX (right now - propper reference scale should be TAI).

Those believing in rubber seconds up to 1972 and then SI/TAI seconds
from that will get the more convenient 10 s offset from 1 Jan 1972,
making them having an offset of 24 s from POSIX.

Using the current POSIX definition UTC is mapped into time_t, skipping
leap seconds but otherwise maintaining UTC distances. It will thus
honour the rubber seconds (after the fact). However, celebrating
1234567890 seconds of time_t makes no sense at the time that time_t
reads 1234567890 since it is not the number of seconds from the
reference epoch, it is a form of "mock seconds" to make the scales fit.
24 leap seconds have shifted the offset of the scale and 1,999918 rubber
seconds too.

We could then add that original reference epoch was given in GMT, which
should be interpreted as UT1. I have not found any circular B online
covering 1 Jan 1970, but using that information we could deliver the
UT1-UTC and UT1-TAI offsets for the reference epoch for the pre-POSIX

>> So depending on which interpretation you choose... I see some 3-4

>> different times occuring. The spread amongst them is about 26 s or so.


> I think it's clear that Unix time has the well-established naive mapping

> to some form of UT. You can pick UT1 or UTC, giving answers that differ

> by a fraction of a second. Anything that secularly counts other than

> 86400 per UT day isn't Unix time: this includes counting either UTC or

> TAI seconds.

It is naive yes...


More information about the LEAPSECS mailing list