man Time::Stopwatch () - Use tied scalars as timers
NAME
Time::Stopwatch - Use tied scalars as timers
SYNOPSIS
use Time::Stopwatch; tie my $timer, 'Time::Stopwatch';
do_something(); print "Did something in $timer seconds.\n";
my @times = map { $timer = 0; do_something_else(); $timer; } 1 .. 5;
DESCRIPTION
The Time::Stopwatch module provides a convenient interface to timing functions through tied scalars. From the point of view of the user, scalars tied to the module simply increase their value by one every second.
Using the module should mostly be obvious from the synopsis. You can provide an initial value for the timers either by assigning to them or by passing the value as a third argument to tie().
If you have the module Time::HiRes installed, the timers created by Time::Stopwatch will automatically count fractional seconds. Do not assume that the values of the timers are always integers. You may test the constant CWTime::Stopwatch::HIRES to find out whether high resolution timing is enabled.
A note on timing short intervals
Time::Stopwatch is primarily designed for timing moderately long intervals (i.e. several seconds), where the overhead imposed by the tie() interface does not matter. With Time::HiRes installed, it can nonetheless be used for even microsecond timing, provided that appropriate care is taken.
- •
- Explicitly initialize the timer by assignment. The first measurement taken before resetting the timer will be a few microseconds longer due to the overhead of the tie() call.
- •
- Always subtract the overhead of the timing code. This is true in general even if you're not using Time::Stopwatch. (High-level benchmarking tools like Benchmark.pm do this automatically.) See the code example below.
- •
- Take as many measurements as you can to minimize random errors. The Statistics::Descriptive module may be useful for analyzing the data. This advice is also true for all benchmarking.
- •
- Remember that a benchmark measures the time take to run the benchmark. Any generalizations to real applications may or may not be valid. If you want real world data, profile the real code in real use.
The following sample code should give a relatively reasonable measurement of a the time taken by a short operation:
use Time::HiRes; # high resolution timing required use Time::Stopwatch;
use Statistics::Descriptive; my $stat = Statistics::Descriptive::Sparse->new();
tie my $time, 'Time::Stopwatch'; # code timer tie my $wait, 'Time::Stopwatch'; # loop timer
while ($wait < 60) { # run for one minute my $diff = 0; $time = 0; do_whatever(); $diff += $time; $time = 0; $diff -= $time; $stat->add_data($diff); }
print("count: ", $stat->count(), " iterations\n", "mean: ", $stat->mean(), " seconds\n", "s.d.: ", $stat->standard_deviation(), " seconds\n");
Note that the above code includes the time of the subroutine call in the measurement.
BUGS
Since tied scalars do not (yet?) support atomic modification, use of operators like CW$t++ or CW$t *= 2 on timers will cause them to lose the time it takes to fetch, modify and store the value. I might be able to get around this by overloading the return value of CWFETCH, but I doubt if it's worth the trouble. Just don't do that.
There is no way to force low-resolution timing if Time::HiRes has been installed. I'm not sure why anyone would want to, since int() will do just fine if you want whole seconds, but still..
CHANGE LOG
- 1.00 (15 Mar 2001)
- Explicitly localized CW$SIG{__DIE__} when testing for Time::HiRes availability. Added A note on timing short intervals to the POD documentation. Bumped version to 1, no longer beta.
- 0.03 (27 Feb 2001)
- Modified tests to give more information, reduced subsecond accuracy test to 1/10 seconds to allow for inaccurate select() implementations. Tweaked synopsis and README.
SEE ALSO
Time::HiRes, tie in perlfunc
For a higher-level approach to timing, try (among others) the modules Time::SoFar, Devel::Timer, or Benchmark. Also see the profiling modules Devel::DProf, Devel::SmallProf and Devel::OpProf.
AUTHORS
Copyright 2000-2001, Ilmari Karonen. All rights reserved.
This library is free software; you can redistribute it and/or modify it under the same terms as Perl itself.
Address bug reports and comments to: perl@itz.pp.sci.fi