8 Benchmark - benchmark running times of Perl code
12 use Benchmark qw(:all) ;
14 timethis ($count, "code");
16 # Use Perl code in strings...
18 'Name1' => '...code1...',
19 'Name2' => '...code2...',
22 # ... or use subroutine references.
24 'Name1' => sub { ...code1... },
25 'Name2' => sub { ...code2... },
28 # cmpthese can be used both ways as well
30 'Name1' => '...code1...',
31 'Name2' => '...code2...',
35 'Name1' => sub { ...code1... },
36 'Name2' => sub { ...code2... },
40 $results = timethese($count,
42 'Name1' => sub { ...code1... },
43 'Name2' => sub { ...code2... },
47 cmpthese( $results ) ;
49 $t = timeit($count, '...other code...')
50 print "$count loops of other code took:",timestr($t),"\n";
52 $t = countit($time, '...other code...')
54 print "$count loops of other code took:",timestr($t),"\n";
56 # enable hires wallclock timing if possible
57 use Benchmark ':hireswallclock';
61 The Benchmark module encapsulates a number of routines to help you
62 figure out how long it takes to execute some code.
64 timethis - run a chunk of code several times
66 timethese - run several chunks of code several times
68 cmpthese - print results of timethese as a comparison chart
70 timeit - run a chunk of code and see how long it goes
72 countit - see how many times a chunk of code runs in a given time
81 Returns the current time. Example:
85 # ... your code here ...
87 $td = timediff($t1, $t0);
88 print "the code took:",timestr($td),"\n";
92 Enables or disable debugging by setting the C<$Benchmark::Debug> flag:
95 $t = timeit(10, ' 5 ** $Global ');
100 Returns the number of iterations.
104 =head2 Standard Exports
106 The following routines will be exported into your namespace
107 if you use the Benchmark module:
111 =item timeit(COUNT, CODE)
113 Arguments: COUNT is the number of times to run the loop, and CODE is
114 the code to run. CODE may be either a code reference or a string to
115 be eval'd; either way it will be run in the caller's package.
117 Returns: a Benchmark object.
119 =item timethis ( COUNT, CODE, [ TITLE, [ STYLE ]] )
121 Time COUNT iterations of CODE. CODE may be a string to eval or a
122 code reference; either way the CODE will run in the caller's package.
123 Results will be printed to STDOUT as TITLE followed by the times.
124 TITLE defaults to "timethis COUNT" if none is provided. STYLE
125 determines the format of the output, as described for timestr() below.
127 The COUNT can be zero or negative: this means the I<minimum number of
128 CPU seconds> to run. A zero signifies the default of 3 seconds. For
129 example to run at least for 10 seconds:
133 or to run two pieces of code tests for at least 3 seconds:
135 timethese(0, { test1 => '...', test2 => '...'})
137 CPU seconds is, in UNIX terms, the user time plus the system time of
138 the process itself, as opposed to the real (wallclock) time and the
139 time spent by the child processes. Less than 0.1 seconds is not
140 accepted (-0.01 as the count, for example, will cause a fatal runtime
143 Note that the CPU seconds is the B<minimum> time: CPU scheduling and
144 other operating system factors may complicate the attempt so that a
145 little bit more time is spent. The benchmark output will, however,
146 also tell the number of C<$code> runs/second, which should be a more
147 interesting number than the actually spent seconds.
149 Returns a Benchmark object.
151 =item timethese ( COUNT, CODEHASHREF, [ STYLE ] )
153 The CODEHASHREF is a reference to a hash containing names as keys
154 and either a string to eval or a code reference for each value.
155 For each (KEY, VALUE) pair in the CODEHASHREF, this routine will
158 timethis(COUNT, VALUE, KEY, STYLE)
160 The routines are called in string comparison order of KEY.
162 The COUNT can be zero or negative, see timethis().
164 Returns a hash of Benchmark objects, keyed by name.
166 =item timediff ( T1, T2 )
168 Returns the difference between two Benchmark times as a Benchmark
169 object suitable for passing to timestr().
171 =item timestr ( TIMEDIFF, [ STYLE, [ FORMAT ] ] )
173 Returns a string that formats the times in the TIMEDIFF object in
174 the requested STYLE. TIMEDIFF is expected to be a Benchmark object
175 similar to that returned by timediff().
177 STYLE can be any of 'all', 'none', 'noc', 'nop' or 'auto'. 'all' shows
178 each of the 5 times available ('wallclock' time, user time, system time,
179 user time of children, and system time of children). 'noc' shows all
180 except the two children times. 'nop' shows only wallclock and the
181 two children times. 'auto' (the default) will act as 'all' unless
182 the children times are both zero, in which case it acts as 'noc'.
183 'none' prevents output.
185 FORMAT is the L<printf(3)>-style format specifier (without the
186 leading '%') to use to print the times. It defaults to '5.2f'.
190 =head2 Optional Exports
192 The following routines will be exported into your namespace
193 if you specifically ask that they be imported:
197 =item clearcache ( COUNT )
199 Clear the cached time for COUNT rounds of the null loop.
201 =item clearallcache ( )
203 Clear all cached times.
205 =item cmpthese ( COUNT, CODEHASHREF, [ STYLE ] )
207 =item cmpthese ( RESULTSHASHREF, [ STYLE ] )
209 Optionally calls timethese(), then outputs comparison chart. This:
211 cmpthese( -1, { a => "++\$i", b => "\$i *= 2" } ) ;
213 outputs a chart like:
219 This chart is sorted from slowest to fastest, and shows the percent speed
220 difference between each pair of tests.
222 c<cmpthese> can also be passed the data structure that timethese() returns:
224 $results = timethese( -1, { a => "++\$i", b => "\$i *= 2" } ) ;
225 cmpthese( $results );
227 in case you want to see both sets of results.
229 Returns a reference to an ARRAY of rows, each row is an ARRAY of cells from the
230 above chart, including labels. This:
232 my $rows = cmpthese( -1, { a => '++$i', b => '$i *= 2' }, "none" );
234 returns a data structure like:
237 [ '', 'Rate', 'b', 'a' ],
238 [ 'b', '2885232/s', '--', '-59%' ],
239 [ 'a', '7099126/s', '146%', '--' ],
242 B<NOTE>: This result value differs from previous versions, which returned
243 the C<timethese()> result structure. If you want that, just use the two
244 statement C<timethese>...C<cmpthese> idiom shown above.
246 Incidently, note the variance in the result values between the two examples;
247 this is typical of benchmarking. If this were a real benchmark, you would
248 probably want to run a lot more iterations.
250 =item countit(TIME, CODE)
252 Arguments: TIME is the minimum length of time to run CODE for, and CODE is
253 the code to run. CODE may be either a code reference or a string to
254 be eval'd; either way it will be run in the caller's package.
256 TIME is I<not> negative. countit() will run the loop many times to
257 calculate the speed of CODE before running it for TIME. The actual
258 time run for will usually be greater than TIME due to system clock
259 resolution, so it's best to look at the number of iterations divided
260 by the times that you are concerned with, not just the iterations.
262 Returns: a Benchmark object.
264 =item disablecache ( )
266 Disable caching of timings for the null loop. This will force Benchmark
267 to recalculate these timings for each new piece of code timed.
269 =item enablecache ( )
271 Enable caching of timings for the null loop. The time taken for COUNT
272 rounds of the null loop will be calculated only once for each
273 different COUNT used.
275 =item timesum ( T1, T2 )
277 Returns the sum of two Benchmark times as a Benchmark object suitable
278 for passing to timestr().
282 =head2 :hireswallclock
284 If the Time::HiRes module has been installed, you can specify the
285 special tag C<:hireswallclock> for Benchmark (if Time::HiRes is not
286 available, the tag will be silently ignored). This tag will cause the
287 wallclock time to be measured in microseconds, instead of integer
288 seconds. Note though that the speed computations are still conducted
289 in CPU time, not wallclock time.
293 The data is stored as a list of values from the time and times
296 ($real, $user, $system, $children_user, $children_system, $iters)
298 in seconds for the whole loop (not divided by the number of rounds).
300 The timing is done using time(3) and times(3).
302 Code is executed in the caller's package.
304 The time of the null loop (a loop with the same
305 number of rounds but empty loop body) is subtracted
306 from the time of the real loop.
308 The null loop times can be cached, the key being the
309 number of rounds. The caching can be controlled using
318 Caching is off by default, as it can (usually slightly) decrease
319 accuracy and does not usually noticably affect runtimes.
325 use Benchmark qw( cmpthese ) ;
332 outputs something like this:
334 Benchmark: running a, b, each for at least 5 CPU seconds...
342 use Benchmark qw( timethese cmpthese ) ;
344 $r = timethese( -5, {
350 outputs something like this:
352 Benchmark: running a, b, each for at least 5 CPU seconds...
353 a: 10 wallclock secs ( 5.14 usr + 0.13 sys = 5.27 CPU) @ 3835055.60/s (n=20210743)
354 b: 5 wallclock secs ( 5.41 usr + 0.00 sys = 5.41 CPU) @ 1574944.92/s (n=8520452)
362 Benchmark inherits from no other class, except of course
367 Comparing eval'd strings with code references will give you
368 inaccurate results: a code reference will show a slightly slower
369 execution time than the equivalent eval'd string.
371 The real time timing is done using time(2) and
372 the granularity is therefore only one second.
374 Short tests may produce negative figures because perl
375 can appear to take longer to execute the empty loop
376 than a short test; try:
380 The system time of the null loop might be slightly
381 more than the system time of the loop with the actual
382 code and therefore the difference might end up being E<lt> 0.
386 L<Devel::DProf> - a Perl code profiler
390 Jarkko Hietaniemi <F<jhi@iki.fi>>, Tim Bunce <F<Tim.Bunce@ig.co.uk>>
392 =head1 MODIFICATION HISTORY
394 September 8th, 1994; by Tim Bunce.
396 March 28th, 1997; by Hugo van der Sanden: added support for code
397 references and the already documented 'debug' method; revamped
400 April 04-07th, 1997: by Jarkko Hietaniemi, added the run-for-some-time
403 September, 1999; by Barrie Slaymaker: math fixes and accuracy and
404 efficiency tweaks. Added cmpthese(). A result is now returned from
405 timethese(). Exposed countit() (was runfor()).
407 December, 2001; by Nicholas Clark: make timestr() recognise the style 'none'
408 and return an empty string. If cmpthese is calling timethese, make it pass the
409 style in. (so that 'none' will suppress output). Make sub new dump its
410 debugging output to STDERR, to be consistent with everything else.
411 All bugs found while writing a regression test.
413 September, 2002; by Jarkko Hietaniemi: add ':hireswallclock' special tag.
417 # evaluate something in a clean lexical environment
418 sub _doeval { no strict; eval shift }
421 # put any lexicals at file scope AFTER here
427 our(@ISA, @EXPORT, @EXPORT_OK, %EXPORT_TAGS, $VERSION);
430 @EXPORT=qw(timeit timethis timethese timediff timestr);
431 @EXPORT_OK=qw(timesum cmpthese countit
432 clearcache clearallcache disablecache enablecache);
433 %EXPORT_TAGS=( all => [ @EXPORT, @EXPORT_OK ] ) ;
437 # --- ':hireswallclock' special handling
441 sub mytime () { time }
446 if (eval 'require Time::HiRes') {
447 import Time::HiRes qw(time);
448 $hirestime = \&Time::HiRes::time;
454 if (grep { $_ eq ":hireswallclock" } @_) {
455 @_ = grep { $_ ne ":hireswallclock" } @_;
456 *mytime = $hirestime if defined $hirestime;
458 Benchmark->export_to_level(1, $class, @_);
461 our($Debug, $Min_Count, $Min_CPU, $Default_Format, $Default_Style,
462 %_Usage, %Cache, $Do_Cache);
468 $Default_Format = '5.2f';
469 $Default_Style = 'auto';
470 # The cache can cause a slight loss of sys time accuracy. If a
471 # user does many tests (>10) with *very* large counts (>10000)
472 # or works on a very slow machine the cache may be useful.
477 sub debug { $Debug = ($_[1] != 0); }
480 my $calling_sub = (caller(1))[3];
481 $calling_sub =~ s/^Benchmark:://;
482 return $_Usage{$calling_sub} || '';
485 # The cache needs two branches: 's' for strings and 'c' for code. The
486 # empty loop is different in these two cases.
489 delete $Cache{"$_[0]c"}; delete $Cache{"$_[0]s"};
492 sub clearallcache () {
500 sub disablecache () {
505 # --- Functions to process the 'time' data type
507 sub new { my @t = (mytime, times, @_ == 2 ? $_[1] : 0);
508 print STDERR "new=@t\n" if $Debug;
511 sub cpu_p { my($r,$pu,$ps,$cu,$cs) = @{$_[0]}; $pu+$ps ; }
512 sub cpu_c { my($r,$pu,$ps,$cu,$cs) = @{$_[0]}; $cu+$cs ; }
513 sub cpu_a { my($r,$pu,$ps,$cu,$cs) = @{$_[0]}; $pu+$ps+$cu+$cs ; }
514 sub real { my($r,$pu,$ps,$cu,$cs) = @{$_[0]}; $r ; }
515 sub iters { $_[0]->[5] ; }
518 $_Usage{timediff} = <<'USAGE';
519 usage: $result_diff = timediff($result1, $result2);
525 die usage unless ref $a and ref $b;
528 for (my $i=0; $i < @$a; ++$i) {
529 push(@r, $a->[$i] - $b->[$i]);
534 $_Usage{timesum} = <<'USAGE';
535 usage: $sum = timesum($result1, $result2);
541 die usage unless ref $a and ref $b;
544 for (my $i=0; $i < @$a; ++$i) {
545 push(@r, $a->[$i] + $b->[$i]);
551 $_Usage{timestr} = <<'USAGE';
552 usage: $formatted_result = timestr($result1);
556 my($tr, $style, $f) = @_;
558 die usage unless ref $tr;
561 warn "bad time value (@t)" unless @t==6;
562 my($r, $pu, $ps, $cu, $cs, $n) = @t;
563 my($pt, $ct, $tt) = ($tr->cpu_p, $tr->cpu_c, $tr->cpu_a);
564 $f = $Default_Format unless defined $f;
565 # format a time in the required style, other formats may be added here
566 $style ||= $Default_Style;
567 return '' if $style eq 'none';
568 $style = ($ct>0) ? 'all' : 'noc' if $style eq 'auto';
569 my $s = "@t $style"; # default for unknown style
570 my $w = $hirestime ? "%2g" : "%2d";
571 $s=sprintf("$w wallclock secs (%$f usr %$f sys + %$f cusr %$f csys = %$f CPU)",
572 $r,$pu,$ps,$cu,$cs,$tt) if $style eq 'all';
573 $s=sprintf("$w wallclock secs (%$f usr + %$f sys = %$f CPU)",
574 $r,$pu,$ps,$pt) if $style eq 'noc';
575 $s=sprintf("$w wallclock secs (%$f cusr + %$f csys = %$f CPU)",
576 $r,$cu,$cs,$ct) if $style eq 'nop';
577 $s .= sprintf(" @ %$f/s (n=$n)", $n / ( $pu + $ps )) if $n && $pu+$ps;
583 print STDERR "$msg",timestr($t),"\n" if $Debug;
586 # --- Functions implementing low-level support for timing loops
588 $_Usage{runloop} = <<'USAGE';
589 usage: runloop($number, [$string | $coderef])
595 $n+=0; # force numeric now, so garbage won't creep into the eval
596 croak "negative loopcount $n" if $n<0;
597 confess usage unless defined $c;
598 my($t0, $t1, $td); # before, after, difference
600 # find package of caller so we can execute code there
601 my($curpack) = caller(0);
603 while (($pack) = caller(++$i)) {
604 last if $pack ne $curpack;
607 my ($subcode, $subref);
608 if (ref $c eq 'CODE') {
609 $subcode = "sub { for (1 .. $n) { local \$_; package $pack; &\$c; } }";
610 $subref = eval $subcode;
613 $subcode = "sub { for (1 .. $n) { local \$_; package $pack; $c;} }";
614 $subref = _doeval($subcode);
616 croak "runloop unable to compile '$c': $@\ncode: $subcode\n" if $@;
617 print STDERR "runloop $n '$subcode'\n" if $Debug;
619 # Wait for the user timer to tick. This makes the error range more like
620 # -0.01, +0. If we don't wait, then it's more like -0.01, +0.01. This
621 # may not seem important, but it significantly reduces the chances of
622 # getting a too low initial $n in the initial, 'find the minimum' loop
623 # in &countit. This, in turn, can reduce the number of calls to
624 # &runloop a lot, and thus reduce additive errors.
625 my $tbase = Benchmark->new(0)->[1];
626 while ( ( $t0 = Benchmark->new(0) )->[1] == $tbase ) {} ;
628 $t1 = Benchmark->new($n);
629 $td = &timediff($t1, $t0);
630 timedebug("runloop:",$td);
634 $_Usage{timeit} = <<'USAGE';
635 usage: $result = timeit($count, 'code' ); or
636 $result = timeit($count, sub { code } );
643 die usage unless defined $code and
644 (!ref $code or ref $code eq 'CODE');
646 printf STDERR "timeit $n $code\n" if $Debug;
647 my $cache_key = $n . ( ref( $code ) ? 'c' : 's' );
648 if ($Do_Cache && exists $Cache{$cache_key} ) {
649 $wn = $Cache{$cache_key};
651 $wn = &runloop($n, ref( $code ) ? sub { } : '' );
652 # Can't let our baseline have any iterations, or they get subtracted
655 $Cache{$cache_key} = $wn;
658 $wc = &runloop($n, $code);
660 $wd = timediff($wc, $wn);
661 timedebug("timeit: ",$wc);
662 timedebug(" - ",$wn);
663 timedebug(" = ",$wd);
673 $_Usage{countit} = <<'USAGE';
674 usage: $result = countit($time, 'code' ); or
675 $result = countit($time, sub { code } );
679 my ( $tmax, $code ) = @_;
683 if ( not defined $tmax or $tmax == 0 ) {
684 $tmax = $default_for;
685 } elsif ( $tmax < 0 ) {
689 die "countit($tmax, ...): timelimit cannot be less than $min_for.\n"
694 # First find the minimum $n that gives a significant timing.
695 for ($n = 1; ; $n *= 2 ) {
696 my $td = timeit($n, $code);
697 $tc = $td->[1] + $td->[2];
703 # Get $n high enough that we can guess the final $n with some accuracy.
704 my $tpra = 0.1 * $tmax; # Target/time practice.
705 while ( $tc < $tpra ) {
706 # The 5% fudge is to keep us from iterating again all
707 # that often (this speeds overall responsiveness when $tmax is big
708 # and we guess a little low). This does not noticably affect
709 # accuracy since we're not couting these times.
710 $n = int( $tpra * 1.05 * $n / $tc ); # Linear approximation.
711 my $td = timeit($n, $code);
712 my $new_tc = $td->[1] + $td->[2];
713 # Make sure we are making progress.
714 $tc = $new_tc > 1.2 * $tc ? $new_tc : 1.2 * $tc;
717 # Now, do the 'for real' timing(s), repeating until we exceed
727 # The 5% fudge is because $n is often a few % low even for routines
728 # with stable times and avoiding extra timeit()s is nice for
730 $n = int( $n * ( 1.05 * $tmax / $tc ) );
733 my $td = timeit($n, $code);
740 $ttot = $utot + $stot;
741 last if $ttot >= $tmax;
743 $ttot = 0.01 if $ttot < 0.01;
744 my $r = $tmax / $ttot - 1; # Linear approximation.
745 $n = int( $r * $ntot );
746 $n = $nmin if $n < $nmin;
749 return bless [ $rtot, $utot, $stot, $cutot, $cstot, $ntot ];
752 # --- Functions implementing high-level time-then-print utilities
756 return $n == 0 ? $default_for : $n < 0 ? -$n : undef;
759 $_Usage{timethis} = <<'USAGE';
760 usage: $result = timethis($time, 'code' ); or
761 $result = timethis($time, sub { code } );
765 my($n, $code, $title, $style) = @_;
768 die usage unless defined $code and
769 (!ref $code or ref $code eq 'CODE');
772 croak "non-integer loopcount $n, stopped" if int($n)<$n;
773 $t = timeit($n, $code);
774 $title = "timethis $n" unless defined $title;
776 my $fort = n_to_for( $n );
777 $t = countit( $fort, $code );
778 $title = "timethis for $fort" unless defined $title;
782 $style = "" unless defined $style;
783 printf("%10s: ", $title) unless $style eq 'none';
784 print timestr($t, $style, $Default_Format),"\n" unless $style eq 'none';
786 $n = $forn if defined $forn;
788 # A conservative warning to spot very silly tests.
789 # Don't assume that your benchmark is ok simply because
790 # you don't get this warning!
791 print " (warning: too few iterations for a reliable count)\n"
793 || ($t->real < 1 && $n < 1000)
794 || $t->cpu_a < $Min_CPU;
799 $_Usage{timethese} = <<'USAGE';
800 usage: timethese($count, { Name1 => 'code1', ... }); or
801 timethese($count, { Name1 => sub { code1 }, ... });
805 my($n, $alt, $style) = @_;
806 die usage unless ref $alt eq 'HASH';
808 my @names = sort keys %$alt;
809 $style = "" unless defined $style;
810 print "Benchmark: " unless $style eq 'none';
812 croak "non-integer loopcount $n, stopped" if int($n)<$n;
813 print "timing $n iterations of" unless $style eq 'none';
815 print "running" unless $style eq 'none';
817 print " ", join(', ',@names) unless $style eq 'none';
819 my $for = n_to_for( $n );
820 print ", each" if $n > 1 && $style ne 'none';
821 print " for at least $for CPU seconds" unless $style eq 'none';
823 print "...\n" unless $style eq 'none';
825 # we could save the results in an array and produce a summary here
826 # sum, min, max, avg etc etc
828 foreach my $name (@names) {
829 $results{$name} = timethis ($n, $alt -> {$name}, $name, $style);
836 $_Usage{cmpthese} = <<'USAGE';
837 usage: cmpthese($count, { Name1 => 'code1', ... }); or
838 cmpthese($count, { Name1 => sub { code1 }, ... }); or
839 cmpthese($result, $style);
843 my ($results, $style);
846 ($results, $style) = @_;
849 my($count, $code) = @_[0,1];
850 $style = $_[2] if defined $_[2];
852 die usage unless ref $code eq 'HASH';
854 $results = timethese($count, $code, ($style || "none"));
857 $style = "" unless defined $style;
859 # Flatten in to an array of arrays with the name as the first field
860 my @vals = map{ [ $_, @{$results->{$_}} ] } keys %$results;
863 # The epsilon fudge here is to prevent div by 0. Since clock
864 # resolutions are much larger, it's below the noise floor.
865 my $rate = $_->[6] / ( $_->[2] + $_->[3] + 0.000000000000001 );
870 @vals = sort { $a->[7] <=> $b->[7] } @vals;
872 # If more than half of the rates are greater than one...
873 my $display_as_rate = $vals[$#vals>>1]->[7] > 1;
880 $display_as_rate ? 'Rate' : 's/iter',
881 map { $_->[0] } @vals
884 push @rows, \@top_row;
885 @col_widths = map { length( $_ ) } @top_row;
887 # Build the data rows
888 # We leave the last column in even though it never has any data. Perhaps
889 # it should go away. Also, perhaps a style for a single column of
890 # percentages might be nice.
891 for my $row_val ( @vals ) {
894 # Column 0 = test name
895 push @row, $row_val->[0];
896 $col_widths[0] = length( $row_val->[0] )
897 if length( $row_val->[0] ) > $col_widths[0];
899 # Column 1 = performance
900 my $row_rate = $row_val->[7];
902 # We assume that we'll never get a 0 rate.
903 my $rate = $display_as_rate ? $row_rate : 1 / $row_rate;
905 # Only give a few decimal places before switching to sci. notation,
906 # since the results aren't usually that accurate anyway.
921 my $formatted_rate = sprintf( $format, $rate );
922 push @row, $formatted_rate;
923 $col_widths[1] = length( $formatted_rate )
924 if length( $formatted_rate ) > $col_widths[1];
926 # Columns 2..N = performance ratios
928 for ( my $col_num = 0 ; $col_num < @vals ; ++$col_num ) {
929 my $col_val = $vals[$col_num];
934 elsif ( $col_val->[0] eq $row_val->[0] ) {
939 my $col_rate = $col_val->[7];
940 $out = sprintf( "%.0f%%", 100*$row_rate/$col_rate - 100 );
943 $col_widths[$col_num+2] = length( $out )
944 if length( $out ) > $col_widths[$col_num+2];
946 # A little wierdness to set the first column width properly
947 $col_widths[$col_num+2] = length( $col_val->[0] )
948 if length( $col_val->[0] ) > $col_widths[$col_num+2];
953 return \@rows if $style eq "none";
955 # Equalize column widths in the chart as much as possible without
956 # exceeding 80 characters. This does not use or affect cols 0 or 1.
957 my @sorted_width_refs =
958 sort { $$a <=> $$b } map { \$_ } @col_widths[2..$#col_widths];
959 my $max_width = ${$sorted_width_refs[-1]};
961 my $total = @col_widths - 1 ;
962 for ( @col_widths ) { $total += $_ }
965 while ( $total < 80 ) {
966 my $min_width = ${$sorted_width_refs[0]};
968 if $min_width == $max_width;
969 for ( @sorted_width_refs ) {
980 my $format = join( ' ', map { "%${_}s" } @col_widths ) . "\n";
981 substr( $format, 1, 0 ) = '-';