It's pretty easy just to assume that "it" is so. "It" may of course be any of many things, but in this case "it" is that "push and pop are obviously faster than unshift and shift". Clearly push and pop work at the growable end of the array so obviously there is an efficiency advantage. I was curious to see how much faster the push/pop pair were than the unshift/shift pair. The answer may come as something of a surprise - it did to me!

use strict;
use warnings;
use Benchmark qw(cmpthese);

my @template = (1) x 200;

cmpthese
  (
  -10,
    {
    'push pop', sub {pushPop ();},
    'push shift', sub {pushShift ();},
    'unshift pop', sub {unshiftPop ();},
    'unshift shift', sub {unshiftShift ();},
    }
  );

sub pushPop
{
my @from = @template;
my @to;
for (0..100)
  {
  push @to, pop @from for @from;
  push @from, pop @to for @to;
  }
}

sub pushShift
{
my @from = @template;
my @to;
for (0..100)
  {
  push @to, shift @from for @from;
  push @from, shift @to for @to;
  }
}

sub unshiftPop
{
my @from = @template;
my @to;
for (0..100)
  {
  unshift @to, pop @from for @from;
  unshift @from, pop @to for @to;
  }
}

sub unshiftShift
{
my @from = @template;
my @to;
for (0..100)
  {
  unshift @to, shift @from for @from;
  unshift @from, shift @to for @to;
  }
}
[download]

               Rate unshift shift      push pop   unshift pop    push 
+shift
unshift shift 129/s            --           -0%           -4%         
+  -6%
push pop      129/s            0%            --           -3%         
+  -6%
unshift pop   134/s            4%            3%            --         
+  -2%
push shift    137/s            6%            6%            2%         
+   --
[download]

There are two big surprises here, at least for me:

• The stack pairs (push/pop and unshift/shift) run at the same speed
• The queue pairs (push/shift and unshift/pop) run at close to the same speed and are both slightly faster than the stack pairs

Every now and then a problem comes up that could be solved using either of the stack pairs. In the past I would always choose push/pop. Now I'm happier thinking about the best expression of the problem and not being distracted by assumed efficiency issues.

Even more interesting though is the queue result which was quite unexpected. Note though that the fastest pair was push/shift. After a little contemplation this is not quite so surprising. After all shift is often used to pull arguments out of the list passed to a sub and push is often used to tack elements on to the end of a list. These are used much more than pop and unshift so it makes sense that there is some optimisation for them. Obvious after the benchmark anyway. :)