Time::HiRes - High resolution alarm, sleep, gettimeofday, interval timers
use Time::HiRes qw( usleep ualarm gettimeofday tv_interval nanosleep clock_gettime clock_getres clock_nanosleep clock stat lstat utime);usleep ($microseconds);nanosleep ($nanoseconds);ualarm ($microseconds);ualarm ($microseconds, $interval_microseconds);$t0 = [gettimeofday];($seconds, $microseconds) = gettimeofday;$elapsed = tv_interval ( $t0, [$seconds, $microseconds]);$elapsed = tv_interval ( $t0, [gettimeofday]);$elapsed = tv_interval ( $t0 );use Time::HiRes qw ( time alarm sleep );$now_fractions = time;sleep ($floating_seconds);alarm ($floating_seconds);alarm ($floating_seconds, $floating_interval);use Time::HiRes qw( setitimer getitimer );setitimer ($which, $floating_seconds, $floating_interval );getitimer ($which);use Time::HiRes qw( clock_gettime clock_getres clock_nanosleep ITIMER_REAL ITIMER_VIRTUAL ITIMER_PROF ITIMER_REALPROF );$realtime = clock_gettime(CLOCK_REALTIME);$resolution = clock_getres(CLOCK_REALTIME);clock_nanosleep(CLOCK_REALTIME, 1.5e9);clock_nanosleep(CLOCK_REALTIME, time()*1e9 + 10e9, TIMER_ABSTIME);my $ticktock = clock();use Time::HiRes qw( stat lstat );my @stat = stat("file");my @stat = stat(FH);my @stat = lstat("file");use Time::HiRes qw( utime );utime $floating_seconds, $floating_seconds, file...;TheTime::HiRes module implements a Perl interface to theusleep,nanosleep,ualarm,gettimeofday, andsetitimer/getitimer system calls, in other words, high resolution time and timers. See the"EXAMPLES" section below and the test scripts for usage; see your system documentation for the description of the underlyingnanosleep orusleep,ualarm,gettimeofday, andsetitimer/getitimer calls.
If your system lacksgettimeofday() or an emulation of it you don't getgettimeofday() or the one-argument form oftv_interval(). If your system lacks all ofnanosleep(),usleep(),select(), andpoll, you don't getTime::HiRes::usleep(),Time::HiRes::nanosleep(), orTime::HiRes::sleep(). If your system lacks bothualarm() andsetitimer() you don't getTime::HiRes::ualarm() orTime::HiRes::alarm().
If you try to import an unimplemented function in theuse statement it will fail at compile time.
If your subsecond sleeping is implemented withnanosleep() instead ofusleep(), you can mix subsecond sleeping with signals sincenanosleep() does not use signals. This, however, is not portable, and you should first check for the truth value of&Time::HiRes::d_nanosleep to see whether you have nanosleep, and then carefully read yournanosleep() C API documentation for any peculiarities.
If you are usingnanosleep for something else than mixing sleeping with signals, give some thought to whether Perl is the tool you should be using for work requiring nanosecond accuracies.
Remember that unless you are working on ahard realtime system, any clocks and timers will be imprecise, especially so if you are working in a pre-emptive multiuser system. Understand the difference betweenwallclock time and process time (in UNIX-like systems the sum ofuser andsystem times). Any attempt to sleep for X seconds will most probably end up sleepingmore than that, but don't be surprised if you end up sleeping slightlyless.
The following functions can be imported from this module. No functions are exported by default.
In array context returns a two-element array with the seconds and microseconds since the epoch. In scalar context returns floating seconds likeTime::HiRes::time() (see below).
Sleeps for the number of microseconds (millionths of a second) specified. Returns the number of microseconds actually slept. Can sleep for more than one second, unlike theusleep system call. Can also sleep for zero seconds, which often works like athread yield. See alsoTime::HiRes::sleep(), andclock_nanosleep().
Do not expect usleep() to be exact down to one microsecond.
Sleeps for the number of nanoseconds (1e9ths of a second) specified. Returns the number of nanoseconds actually slept (accurate only to microseconds, the nearest thousand of them). Can sleep for more than one second. Can also sleep for zero seconds, which often works like athread yield. See alsoTime::HiRes::sleep(),Time::HiRes::usleep(), andclock_nanosleep().
Do not expect nanosleep() to be exact down to one nanosecond. Getting even accuracy of one thousand nanoseconds is good.
Issues aualarm call; the$interval_useconds is optional and will be zero if unspecified, resulting inalarm-like behaviour.
Returns the remaining time in the alarm in microseconds, orundef if an error occurred.
ualarm(0) will cancel an outstanding ualarm().
Note that the interaction between alarms and sleeps is unspecified.
tv_interval ( $ref_to_gettimeofday [, $ref_to_later_gettimeofday] )
Returns the floating seconds between the two times, which should have been returned bygettimeofday(). If the second argument is omitted, then the current time is used.
Returns a floating seconds since the epoch. This function can be imported, resulting in a nice drop-in replacement for thetime provided with core Perl; see the"EXAMPLES" below.
NOTE 1: This higher resolution timer can return values either less or more than the coretime(), depending on whether your platform rounds the higher resolution timer values up, down, or to the nearest second to get the coretime(), but naturally the difference should be never more than half a second. See also"clock_getres", if available in your system.
NOTE 2: Since Sunday, September 9th, 2001 at 01:46:40 AM GMT, when thetime() seconds since epoch rolled over to 1_000_000_000, the default floating point format of Perl and the seconds since epoch have conspired to produce an apparent bug: if you print the value ofTime::HiRes::time() you seem to be getting only five decimals, not six as promised (microseconds). Not to worry, the microseconds are there (assuming your platform supports such granularity in the first place). What is going on is that the default floating point format of Perl only outputs 15 digits. In this case that means ten digits before the decimal separator and five after. To see the microseconds you can use eitherprintf/sprintf with"%.6f", or thegettimeofday() function in list context, which will give you the seconds and microseconds as two separate values.
Sleeps for the specified amount of seconds. Returns the number of seconds actually slept (a floating point value). This function can be imported, resulting in a nice drop-in replacement for thesleep provided with perl, see the"EXAMPLES" below.
Note that the interaction between alarms and sleeps is unspecified.
TheSIGALRM signal is sent after the specified number of seconds. Implemented usingsetitimer() if available,ualarm() if not. The$interval_floating_seconds argument is optional and will be zero if unspecified, resulting inalarm()-like behaviour. This function can be imported, resulting in a nice drop-in replacement for thealarm provided with perl, see the"EXAMPLES" below.
Returns the remaining time in the alarm in seconds, orundef if an error occurred.
NOTE 1: With some combinations of operating systems and Perl releasesSIGALRM restartsselect(), instead of interrupting it. This means that analarm() followed by aselect() may together take the sum of the times specified for thealarm() and theselect(), not just the time of thealarm().
Note that the interaction between alarms and sleeps is unspecified.
Start up an interval timer: after a certain time, a signal ($which) arrives, and more signals may keep arriving at certain intervals. To disable an "itimer", use$floating_seconds of zero. If the$interval_floating_seconds is set to zero (or unspecified), the timer is disabledafter the next delivered signal.
Use of interval timers may interfere withalarm(),sleep(), andusleep(). In standard-speak the "interaction is unspecified", which means thatanything may happen: it may work, it may not.
In scalar context, the remaining time in the timer is returned.
In list context, both the remaining time and the interval are returned.
There are usually three or four interval timers (signals) available: the$which can beITIMER_REAL,ITIMER_VIRTUAL,ITIMER_PROF, orITIMER_REALPROF. Note that which ones are available depends: true UNIX platforms usually have the first three, but only Solaris seems to haveITIMER_REALPROF (which is used to profile multithreaded programs). Win32 unfortunately does not have interval timers.
ITIMER_REAL results inalarm()-like behaviour. Time is counted inreal time; that is, wallclock time.SIGALRM is delivered when the timer expires.
ITIMER_VIRTUAL counts time in (process)virtual time; that is, only when the process is running. In multiprocessor/user/CPU systems this may be more or less than real or wallclock time. (This time is also known as theuser time.)SIGVTALRM is delivered when the timer expires.
ITIMER_PROF counts time when either the process virtual time or when the operating system is running on behalf of the process (such as I/O). (This time is also known as thesystem time.) (The sum of user time and system time is known as theCPU time.)SIGPROF is delivered when the timer expires.SIGPROF can interrupt system calls.
The semantics of interval timers for multithreaded programs are system-specific, and some systems may support additional interval timers. For example, it is unspecified which thread gets the signals. See yoursetitimer(2) documentation.
Return the remaining time in the interval timer specified by$which.
In scalar context, the remaining time is returned.
In list context, both the remaining time and the interval are returned. The interval is always what you put in usingsetitimer().
Return as seconds the current value of the POSIX high resolution timer specified by$which. All implementations that support POSIX high resolution timers are supposed to support at least the$which value ofCLOCK_REALTIME, which is supposed to return results close to the results ofgettimeofday, or the number of seconds since 00:00:00:00 January 1, 1970 Greenwich Mean Time (GMT). Do not assume that CLOCK_REALTIME is zero, it might be one, or something else. Another potentially useful (but not available everywhere) value isCLOCK_MONOTONIC, which guarantees a monotonically increasing time value (unlike time() or gettimeofday(), which can be adjusted). See your system documentation for other possibly supported values.
Return as seconds the resolution of the POSIX high resolution timer specified by$which. All implementations that support POSIX high resolution timers are supposed to support at least the$which value ofCLOCK_REALTIME, see"clock_gettime".
NOTE: the resolution returned may be highly optimistic. Even if the resolution is high (a small number), all it means is that you'll be able to specify the arguments to clock_gettime() and clock_nanosleep() with that resolution. The system might not actually be able to measure events at that resolution, and the various overheads and the overall system load are certain to affect any timings.
Sleeps for the number of nanoseconds (1e9ths of a second) specified. Returns the number of nanoseconds actually slept. The $which is the "clock id", as with clock_gettime() and clock_getres(). The flags default to zero butTIMER_ABSTIME can specified (must be exported explicitly) which means that$nanoseconds is not a time interval (as is the default) but instead an absolute time. Can sleep for more than one second. Can also sleep for zero seconds, which often works like athread yield. See alsoTime::HiRes::sleep(),Time::HiRes::usleep(), andTime::HiRes::nanosleep().
Do not expect clock_nanosleep() to be exact down to one nanosecond. Getting even accuracy of one thousand nanoseconds is good.
Return as seconds theprocess time (user + system time) spent by the process since the first call to clock() (the definition isnot "since the start of the process", though if you are lucky these times may be quite close to each other, depending on the system). What this means is that you probably need to store the result of your first call to clock(), and subtract that value from the following results of clock().
The time returned also includes the process times of the terminated child processes for which wait() has been executed. This value is somewhat like the second value returned by the times() of core Perl, but not necessarily identical. Note that due to backward compatibility limitations the returned value may wrap around at about 2147 seconds or at about 36 minutes.
As"stat" in perlfunc or"lstat" in perlfunc but with the access/modify/change file timestamps in subsecond resolution, if the operating system and the filesystem both support such timestamps. To override the standard stat():
use Time::HiRes qw(stat);Test for the value of &Time::HiRes::d_hires_stat to find out whether the operating system supports subsecond file timestamps: a value larger than zero means yes. There are unfortunately no easy ways to find out whether the filesystem supports such timestamps. UNIX filesystems often do; NTFS does; FAT doesn't (FAT timestamp granularity istwo seconds).
A zero return value of &Time::HiRes::d_hires_stat means that Time::HiRes::stat is a no-op passthrough for CORE::stat() (and likewise for lstat), and therefore the timestamps will stay integers. The same thing will happen if the filesystem does not do subsecond timestamps, even if the &Time::HiRes::d_hires_stat is non-zero.
In any case do not expect nanosecond resolution, or even a microsecond resolution. Also note that the modify/access timestamps might have different resolutions, and that they need not be synchronized, e.g. if the operations are
writestat # t1readstat # t2the access time stamp from t2 need not be greater-than the modify time stamp from t1: it may be equal orless.
As"utime" in perlfunc but with the ability to set the access/modify file timestamps in subsecond resolution, if the operating system and the filesystem, and the mount options of the filesystem, all support such timestamps.
To override the standard utime():
use Time::HiRes qw(utime);Test for the value of &Time::HiRes::d_hires_utime to find out whether the operating system supports setting subsecond file timestamps.
As with CORE::utime(), passing undef as both the atime and mtime will call the syscall with a NULL argument.
The actual achievable subsecond resolution depends on the combination of the operating system and the filesystem.
Modifying the timestamps may not be possible at all: for example, thenoatime filesystem mount option may prohibit you from changing the access time timestamp.
Returns the number of files successfully changed.
use Time::HiRes qw(usleep ualarm gettimeofday tv_interval);$microseconds = 750_000;usleep($microseconds);# signal alarm in 2.5s & every .1s thereafterualarm(2_500_000, 100_000);# cancel that ualarmualarm(0);# get seconds and microseconds since the epoch($s, $usec) = gettimeofday();# measure elapsed time# (could also do by subtracting 2 gettimeofday return values)$t0 = [gettimeofday];# do bunch of stuff here$t1 = [gettimeofday];# do more stuff here$t0_t1 = tv_interval $t0, $t1;$elapsed = tv_interval ($t0, [gettimeofday]);$elapsed = tv_interval ($t0); # equivalent code## replacements for time, alarm and sleep that know about# floating seconds#use Time::HiRes;$now_fractions = Time::HiRes::time;Time::HiRes::sleep (2.5);Time::HiRes::alarm (10.6666666);use Time::HiRes qw ( time alarm sleep );$now_fractions = time;sleep (2.5);alarm (10.6666666);# Arm an interval timer to go off first at 10 seconds and# after that every 2.5 seconds, in process virtual timeuse Time::HiRes qw ( setitimer ITIMER_VIRTUAL time );$SIG{VTALRM} = sub { print time, "\n" };setitimer(ITIMER_VIRTUAL, 10, 2.5);use Time::HiRes qw( clock_gettime clock_getres CLOCK_REALTIME );# Read the POSIX high resolution timer.my $high = clock_gettime(CLOCK_REALTIME);# But how accurate we can be, really?my $reso = clock_getres(CLOCK_REALTIME);use Time::HiRes qw( clock_nanosleep TIMER_ABSTIME );clock_nanosleep(CLOCK_REALTIME, 1e6);clock_nanosleep(CLOCK_REALTIME, 2e9, TIMER_ABSTIME);use Time::HiRes qw( clock );my $clock0 = clock();... # Do something.my $clock1 = clock();my $clockd = $clock1 - $clock0;use Time::HiRes qw( stat );my ($atime, $mtime, $ctime) = (stat("istics"))[8, 9, 10];In addition to the perl API described above, a C API is available for extension writers. The following C functions are available in the modglobal hash:
name C prototype--------------- ----------------------Time::NVtime NV (*)()Time::U2time void (*)(pTHX_ UV ret[2])Both functions return equivalent information (likegettimeofday) but with different representations. The namesNVtime andU2time were selected mainly because they are operating system independent. (gettimeofday is Unix-centric, though some platforms like Win32 and VMS have emulations for it.)
Here is an example of usingNVtime from C:
NV (*myNVtime)(); /* Returns -1 on failure. */SV **svp = hv_fetchs(PL_modglobal, "Time::NVtime", 0);if (!svp) croak("Time::HiRes is required");if (!SvIOK(*svp)) croak("Time::NVtime isn't a function pointer");myNVtime = INT2PTR(NV(*)(), SvIV(*svp));printf("The current time is: %" NVff "\n", (*myNVtime)());In ualarm() you tried to use number of microseconds or interval (also in microseconds) more than 1_000_000 and setitimer() is not available in your system to emulate that case.
You tried to use a negative time argument.
Something went horribly wrong-- the number of microseconds that cannot become negative just became negative. Maybe your compiler is broken?
In some platforms it is not possible to get an alarm with subsecond resolution and later than one second.
Some calls simply aren't available, real or emulated, on every platform.
Notice that the coretime() maybe rounding rather than truncating. What this means is that the coretime() may be reporting the time as one second later thangettimeofday() andTime::HiRes::time().
Adjusting the system clock (either manually or by services like ntp) may cause problems, especially for long running programs that assume a monotonously increasing time (note that all platforms do not adjust time as gracefully as UNIX ntp does). For example in Win32 (and derived platforms like Cygwin and MinGW) the Time::HiRes::time() may temporarily drift off from the system clock (and the original time()) by up to 0.5 seconds. Time::HiRes will notice this eventually and recalibrate. Note that since Time::HiRes 1.77 the clock_gettime(CLOCK_MONOTONIC) might help in this (in case your system supports CLOCK_MONOTONIC).
Some systems have APIs but not implementations: for example QNX and Haiku have the interval timer APIs but not the functionality.
In pre-Sierra macOS (pre-10.12, OS X) clock_getres(), clock_gettime() and clock_nanosleep() are emulated using the Mach timers; as a side effect of being emulated the CLOCK_REALTIME and CLOCK_MONOTONIC are the same timer.
gnukfreebsd seems to have non-functional futimens() and utimensat() (at least as of 10.1): therefore the hires utime() does not work.
Perl modulesBSD::Resource,Time::TAI64.
Your system documentation forclock(3),clock_gettime(2),clock_getres(3),clock_nanosleep(3),clock_settime(2),getitimer(2),gettimeofday(2),setitimer(2),sleep(3),stat(2),ualarm(3).
D. Wegscheid <wegscd@whirlpool.com> R. Schertler <roderick@argon.org> J. Hietaniemi <jhi@iki.fi> G. Aas <gisle@aas.no>
Copyright (c) 1996-2002 Douglas E. Wegscheid. All rights reserved.
Copyright (c) 2002, 2003, 2004, 2005, 2006, 2007, 2008 Jarkko Hietaniemi. All rights reserved.
Copyright (C) 2011, 2012, 2013 Andrew Main (Zefram) <zefram@fysh.org>
This program is free software; you can redistribute it and/or modify it under the same terms as Perl itself.
Perldoc Browser is maintained by Dan Book (DBOOK). Please contact him via theGitHub issue tracker oremail regarding any issues with the site itself, search, or rendering of documentation.
The Perl documentation is maintained by the Perl 5 Porters in the development of Perl. Please contact them via thePerl issue tracker, themailing list, orIRC to report any issues with the contents or format of the documentation.