3

The Unix timestamp given by:

int(time.time())

gives the number of seconds elapsed since 01/01/1970, without leap-seconds.

Just out of curiosity, how to get the true number of seconds elapsed since this date, leap-seconds included? (i.e. the distance between these two events on a time axis)

Notes:

  • Example: the timestamp range 867715190.000 .. 867715202.000 represents a "real-life duration" of 13 seconds (measured with a timer) because there was a leap-second this day of 1997, whereas the Unix timestamp has only increased of +12.

  • Example 2: the real-life time elapsed between 1/1/1970 and 1/1/2020 (12 leap-years in this 50-year interval) is (365*50+12)*24*3600 + number_leap_seconds and not (365*50+12)*24*3600. But we see datetime.datetime.utcfromtimestamp((365*50+12)*24*3600) is 2020-01-01 00:00:00, so obviously, the leap-seconds haven't been taken in consideration.

  • Linked to What does python return on the leap second, and Unix time and leap seconds

Basj
  • 41,386
  • 99
  • 383
  • 673
  • 1
    Based on my reading of both links you showed, you’re indeed getting the true number of seconds. It’s only that the conversion from this to a date is messed up (because some of those seconds are designated leap seconds). – Arya McCarthy Jan 03 '21 at 16:12
  • See "Example 2" in the edited question @AryaMcCarthy, this shows the leap seconds are not taken in consideration. The Unix timestamp is a representation of time (non-unique in the case of leap seconds!), but not an actual measure of elapsed time. – Basj Jan 03 '21 at 16:48

2 Answers2

2

The number of SI seconds between any two UTC timestamps since 1972-01-01 requires access to the list of leap seconds which have been introduced into UTC. This list is available as part of the IANA tzdata distribution and it can also be obtained from other sources.

Caution is required because the number of SI seconds between what was known as 1970-01-01 and 1972-01-01 is 2x365x24x60x60 + 1.999918 SI seconds because at 1970 the official time was determined not by cesium atoms but by actually measuring the rotation of the earth, so the official seconds were mean solar seconds not SI seconds.

Steve Allen
  • 326
  • 1
  • 3
0

I ran into this same issue, and you can work around it by using "right" time:

import os; import time; import datetime;
tai_epoch=datetime.datetime(1970, 1, 1, 0, 0, 10)
timezone=time.tzname[0];os.environ['TZ']='right/UTC';time.tzset()
now=datetime.datetime.utcnow()
leap_seconds=int(now.timestamp())-int((now-tai_epoch).total_seconds())
right_time=int(now.timestamp())+leap_seconds
os.environ['TZ']=timezone;time.tzset()

You can compare this against the regular UTC time like so:

utc_time=int(time.time())
print(right_time-utc_time)

which, at the moment, will give you 37.

Compholio
  • 733
  • 1
  • 9
  • 17
  • Thank you for the answer. Could you please add an explanation to your code? Just after a quick glance I am wondering: **1.** What is the meaning of the magic 10 seconds? `tai_epoch=datetime.datetime(1970, 1, 1, 0, 0, 10)` **2.** Is `right/UTC` a time zone? What does it mean? `os.environ['TZ']='right/UTC'` – pabouk - Ukraine stay strong Jun 14 '22 at 08:09
  • 1
    **1.** The magic 10 seconds is that on January 1st, 1970 (Unix epoch) the TAI time and the UTC time [differed by 10 seconds](https://en.wikipedia.org/wiki/Unix_time#TAI-based_variant). Technically, TAI time started with a difference of 10 seconds in 1972 - but those two years have no impact here. **2.** "right/UTC" is a special time zone that does not use leap seconds, so if you work exclusively in "right/UTC" time when doing math then you will get the correct number of seconds between two times. – Compholio Jun 14 '22 at 23:09