I re-ran these benchmarks with a minor optimization on the 'scalar(@x)' portion of the loop. It basically proved that the compiler isn't performing an optimization on the limit term in the loop. I'm a little surprised that since the array isn't being modified, and no functions are being called, the compiler didn't optimize that out to a fixed value. Of course, I'm speaking as an old 'C' coder...

Benchmark: timing 10000 iterations of for1000, for10000, for10000a, fo
+r1000a, foreach1000, foreach10000...
   for1000: 73 wallclock secs (60.06 usr +  0.16 sys = 60.22 CPU)
  for10000: 795 wallclock secs (632.63 usr +  1.39 sys = 634.02 CPU)
 for10000a: 508 wallclock secs (398.29 usr +  0.89 sys = 399.18 CPU)
  for1000a: 45 wallclock secs (37.13 usr +  0.06 sys = 37.19 CPU)
foreach1000: 30 wallclock secs (24.32 usr +  0.06 sys = 24.38 CPU)
foreach10000: 360 wallclock secs (276.53 usr +  0.69 sys = 277.22 CPU)

use Benchmark;

sub foreach1000()
{
    my @x = (1..1000);
    foreach(@x)
    {
        my $z = $_;
    }
}

sub for1000()
{
    my @x = (1..1000);
    for(my $y = 0; $y < scalar(@x); $y++)
    {
        my $z = $x[$y];
    }
}

sub for1000a()
{
    my @x = (1..1000);
    my $len = scalar(@x);
    for(my $y = 0; $y < $len; $y++)
    {
        my $z = $x[$y];
    }
}

sub foreach10000()
{
    my @x = (1..10000);
    foreach(@x)
    {
        my $z = $_;
    }
}

sub for10000()
{
    my @x = (1..10000);
    for(my $y = 0; $y < scalar(@x); $y++)
    {
        my $z = $x[$y];
    }
}

sub for10000a()
{
    my @x = (1..10000);
    my $len = scalar(@x);
    for(my $y = 0; $y < $len; $y++)
    {
        my $z = $x[$y];
    }
}

timethese( 10000, { for1000 => 'for1000()', 
            for1000a => 'for1000a()',
            foreach1000 => 'foreach1000()', 
            for10000 => 'for10000()', 
            for10000a => 'for10000a()',
            foreach10000 => 'foreach10000()' });
[download]

--Chris

e-mail jcwren