NAME
Time::UTC::Now - determine current time in UTC correctly
SYNOPSIS
use Time::UTC::Now qw(
now_utc_rat now_utc_sna now_utc_flt now_utc_dec);
($day, $secs, $bound) = now_utc_rat;
($day, $secs, $bound) = now_utc_rat(1);
($day, $secs, $bound) = now_utc_sna;
($day, $secs, $bound) = now_utc_sna(1);
($day, $secs, $bound) = now_utc_flt;
($day, $secs, $bound) = now_utc_flt(1);
($day, $secs, $bound) = now_utc_dec;
($day, $secs, $bound) = now_utc_dec(1);
use Time::UTC::Now qw(utc_day_to_mjdn utc_day_to_cjdn);
$mjdn = utc_day_to_mjdn($day);
$cjdn = utc_day_to_cjdn($day);
DESCRIPTION
This module is one answer to the question "what time is it?". It determines the current time on the UTC scale, handling leap seconds correctly, and puts a bound on how inaccurate it could be. It is the rigorously correct approach to determining civil time. It is designed to interoperate with Time::UTC, which knows all about the UTC time scale.
UTC (Coordinated Universal Time) is a time scale derived from International Atomic Time (TAI). UTC divides time up into days, and each day into seconds. The seconds are atomically-realised SI seconds, of uniform length. Most UTC days are exactly 86400 seconds long, but occasionally there is a day of length 86401 s or (theoretically) 86399 s. These leap seconds are used to keep the UTC day approximately synchronised with the non-uniform rotation of the Earth. (Prior to 1972 a different mechanism was used for UTC, but that's not an issue here.)
Because UTC days have differing lengths, instants on the UTC scale are identified here by the combination of a day number and a number of seconds since midnight within the day. In this module the day number is the integral number of days since 1958-01-01, which is the epoch of the TAI scale which underlies UTC. This is the convention used by the Time::UTC
module. That module has some functions to format these numbers for display. For a more general solution, use the utc_day_to_mjdn
function to translate to a standard Modified Julian Day Number or the utc_day_to_cjdn
function to translate to a standard Chronological Julian Day Number, which can be used as input to a calendar module.
FUNCTIONS
Time determination
Each of these functions determines the current UTC time and returns it. They vary in the form in which the time is returned. In each case, the function returns a list of three values. The first two values identify a current UTC instant, in the form of a day number (number of days since the TAI epoch) and a number of seconds since midnight within the day. The third value is an inaccuracy bound, as a number of seconds, or undef
if no accurate answer could be determined.
If an inaccuracy bound is returned then the function is claiming to have answered correctly, to within the specified margin. That is, some instant during the execution of the function is within the specified margin of the instant identified. (This semantic differs from older current-time interfaces that are content to return an instant that has already passed.) The inaccuracy bound describes the actual time represented in the return values, not some internal value that was rounded to generate the return values.
The inaccuracy bound is measured in UTC seconds; that is, in SI seconds on the Terran geoid as realised by atomic clocks. This differs from SI seconds at the computer's location, but the difference is only apparent if the computer hardware is significantly time dilated with respect to the geoid.
If undef
is returned instead of an inaccuracy bound then the function could not find a trustable answer. Either the clock available was not properly synchronised or its accuracy could not be established. Whatever time could be found is returned, but the function makes no claim that it is accurate. It should be treated with suspicion. In practice, clocks of this nature are especially likely to misbehave around leap seconds.
Each function will die
if it can't find a plausible time at all. If the DEMAND_ACCURACY parameter is supplied and true then it will also die if it could not find an accurate answer, instead of returning with undef
for the inaccuracy bound.
- now_utc_rat([DEMAND_ACCURACY])
-
All three return values are in the form of
Math::BigRat
objects.This retains full resolution, is future-proof, and is easy to manipulate, but beware that
Math::BigRat
is currently rather slow. If performance is a problem then consider using one of the functions below that return the results in other formats. - now_utc_sna([DEMAND_ACCURACY])
-
The day number is returned as a Perl integer. The time since midnight and the inaccuracy bound (if present) are each returned in the form of a three-element array, giving a high-resolution fixed-point number of seconds. The first element is the integral number of whole seconds, the second is an integral number of nanoseconds in the range [0, 1000000000), and the third is an integral number of attoseconds in the same range.
This form of return value is fairly efficient. It is convenient for decimal output, but awkward to do arithmetic with. Its resolution is adequate for the foreseeable future, but could in principle be obsoleted some day.
It is presumed that native integer formats will grow fast enough to always represent the day number fully; if not, 31 bits will overflow late in the sixth megayear of the Common Era. (Average day length by then is projected to be around 86520 s, posing more serious problems for UTC.)
- now_utc_flt([DEMAND_ACCURACY])
-
All the results are returned as native Perl numbers. The day number is returned as a Perl integer, with the same caveat as for
now_utc_sna
. The other two items are floating point numbers.This form of return value is very efficient and easy to manipulate. However, its resolution is limited, rendering it obsolete in the near future unless floating point number formats get bigger.
- now_utc_dec([DEMAND_ACCURACY])
-
Each of the results is returned in the form of a string expressing a number as a decimal fraction. These strings are of the type processed by Math::Decimal, and are always returned in Math::Decimal's canonical form.
This form of return value is fairly efficient and easy to manipulate. It is convenient both for decimal output and (via implicit coercion to floating point) for low-precision arithmetic. Math::Decimal can be used for high-precision arithmetic. Its resolution is unlimited.
Day count conversion
- utc_day_to_mjdn(DAY)
-
This function takes a number of days since the TAI epoch and returns the corresponding Modified Julian Day Number (a number of days since 1858-11-17 UT). MJDN is a standard numbering for days in Universal Time. There is no bound on the permissible day numbers.
- utc_day_to_cjdn(DAY)
-
This function takes a number of days since the TAI epoch and returns the corresponding Chronological Julian Day Number (a number of days since -4713-11-24). CJDN is a standard day numbering that is useful as an interchange format between implementations of different calendars. There is no bound on the permissible day numbers.
TECHNIQUES
There are several interfaces available to determine the time on a computer, and most of them suck. This module will attempt to use the best interface available when it runs. It knows about the following, in descending order of preference:
- ntp_adjtime(), ntp_gettime(), adjtimex()
-
These interfaces were devised for Unix systems using the Mills timekeeping model, which is intended for clocks that are synchronised via NTP (the Network Time Protocol). The timekeeping model is detailed in http://www.eecis.udel.edu/~mills/database/memos/memo96b.ps.
These interfaces gives some leap second indications, and an inaccuracy bound on the time returned. Both are faulty in their raw form, but they are corrected by this module. (Those interested in the gory details are invited to read the source.) Resolution 1 us, or on some systems 1 ns.
- clock_gettime(CLOCK_UTC)
-
This is a proposed interface for relatively simple (simpler than the NTP interface) acquisition of precise real time. It is defined by http://www.cl.cam.ac.uk/~mgk25/posix-clocks.html.
This interface gives an explicit leap second indication, and so permits this module to correctly tick through leap seconds. It also provides a limited kind of inaccuracy bound: it will only provide a reading at all if it is thought to be accurate to within a second. No tighter inaccuracy bound can be given. Resolution 1 ns.
- clock_gettime(CLOCK_REALTIME)
-
This is a POSIX interface.
Misbehaves around leap seconds, and does not give an inaccuracy bound. Resolution 1 ns.
- GetSystemTimeAsFileTime()
-
This is part of the Win32 API of Microsoft Windows.
Misbehaves around leap seconds, and does not give an inaccuracy bound. Resolution of the interface is 100 ns.
- gettimeofday()
-
This is a long-standing Unix interface, so named because it was the interface to the "time-of-day clock".
Misbehaves around leap seconds, and does not give an inaccuracy bound. Resolution 1 us.
- Time::Unix::time()
-
This is derived from the original Unix
time()
function, which was also adopted by the C library standard and by Perl. Various systems have different epochs and resolutions for thetime()
function, so it is not usable by this module on its own. TheTime::Unix
module corrects for the varying epochs across OSes.Misbehaves around leap seconds, and does not give an inaccuracy bound. Resolution 1 s.
The author would welcome patches to this module to make use of high-precision interfaces, along the lines of ntp_adjtime()
, on non-Unix operating systems.
OS-SPECIFIC NOTES
The author would appreciate any new information for this section, especially reports of experiences with this module under OSes not listed in this section.
Cygwin
Expected to use clock_gettime(CLOCK_REALTIME) now that support for it has been added to this module, which gives resolution 1 ns but no uncertainty bound and is discontinuous at leap seconds. Formerly used gettimeofday(), which gives resolution 1 us but no uncertainty bound and is discontinuous at leap seconds. There is no interface that supplies an uncertainty bound or correct leap second handling.
FreeBSD
Uses ntp_gettime(), which gives resolution 1 us or 1 ns (depending on system configuration) and uncertainty bound.
Kernel bug #42089 causes ntp_gettime() on affected systems to denominate time in nanoseconds regardless of the configured resolution. Outside the first millisecond of each second, this provides an out-of-range microseconds count, which this module detects and rejects, falling back to clock_gettime(CLOCK_REALTIME) or gettimeofday() (thus getting no uncertainty bound and discontinuities at leap seconds). Inside the first millisecond of each second, ntp_gettime()'s response looks correct and is accepted, yielding incorrect time (ranging over the whole second) with an uncertainty bound that is insufficient to cover this inaccuracy. It is infeasible for this module to detect that the time is affected by this bug. The bug can be avoided on any particular host by configuring it for nanosecond resolution.
Linux
Uses ntp_adjtime(), which gives resolution 1 us or 1 ns (depending on system configuration) and uncertainty bound.
Solaris
Uses ntp_gettime(), which gives resolution 1 us and uncertainty bound.
Windows
Uses the native GetSystemTimeAsFileTime(). Observed clock resolution is 10 ms, but lower-order digits are supplied (filled with noise) down to the API resolution of 100 ns. There is no uncertainty bound, and there are discontinuities at leap seconds. There is no interface that supplies an uncertainty bound or correct leap second handling.
SEE ALSO
AUTHOR
Andrew Main (Zefram) <zefram@fysh.org>
COPYRIGHT
Copyright (C) 2006, 2007, 2009, 2010, 2012, 2017 Andrew Main (Zefram) <zefram@fysh.org>
LICENSE
This module is free software; you can redistribute it and/or modify it under the same terms as Perl itself.