That is jumping through some hoops to avoid void. SuprisinglyUnsuprisingly (I just looked at it properly) the first one is a lot slower than the map in void context solution, the last one is slowest of all :)

Updated

added the for loop solution to the benchmark, it is nicely faster.

#!/usr/bin/perl
use strict;
use warnings;
use Benchmark;

my @strings;
my @array;

push @strings, $_ x 2 for ("a" .. "z");
for (1..1000) {
    my $i=int rand $#strings;
    push @array, $strings[$i]
}

sub for_loop {
    my %sums;
    $sums{$_}++ foreach @array;
    # print "$_ = $sums{$_}\n" foreach sort keys %sums;
}

sub map_void {
    my %sums;
    map {$sums{$_}++} @array;
    # print "$_ = $sums{$_}\n" foreach sort keys %sums;
}

sub nonvoid_1 {
    my %sums;
    %sums = map {$_ => ++$sums{$_} } @array;
    # print "$_ = $sums{$_}\n" foreach sort keys %sums;
}

sub nonvoid_2 {
    my %sums;
    my $last="";
    %sums = map {
        my $n;
        do { $n = (($last=$_) ... ($last ne $_)) } while $n =~ /E0$/;
        ($last => $n);
    } sort @array;
    # print "$_ = $sums{$_}\n" foreach sort keys %sums;
}

timethese( -3, {map_void  => \&map_void,
                for_loop  => \&for_loop,
                nonvoid_1 => \&nonvoid_1,
                nonvoid_2 => \&nonvoid_2 }
         );

__END__
Benchmark: running for_loop, map_void, nonvoid_1, nonvoid_2 for at lea
+st 3 CPU seconds...
  for_loop:  3 wallclock secs ( 3.13 usr +  0.01 sys =  3.14 CPU) @ 15
+79.94/s (n=4961)
  map_void:  3 wallclock secs ( 3.19 usr +  0.00 sys =  3.19 CPU) @ 11
+65.83/s (n=3719)
 nonvoid_1:  3 wallclock secs ( 3.16 usr +  0.00 sys =  3.16 CPU) @ 33
+0.06/s (n=1043)
 nonvoid_2:  3 wallclock secs ( 3.28 usr +  0.00 sys =  3.28 CPU) @ 13
+9.33/s (n=457)
[download]

Cheers,
R.