Not related to your question, but my pet peeve is code that proceesses data element by element, when it could be handled all at once.

There's no need to process your output line by line. If you print an array on it's own:

the array elements are printed without any separator. If this is a CGI script, separators are ignored by the browser, anyway, so that should be the fastest solution. But, for some odd reason, it is actually quite slow.

If it isn't HTML, and it isn't convenient to add newline characters when the array is generated, you'll need to introduce the newlines at the print. If you have an array in a double-quoted strring, the array elements are printed separated by the value of the variable $". So temporarily re-defined $" to be newline, and use the array in a string ... the string will need a newline at the end.

{
    local $" = "\n";
    print PROG "@array\n";
}
[download]

That's equivalent to using join to convert the array into a string, and equally fast.

print PROG join( "\n", @array), "\n";
[download]

That makes me think the underlying code is very much identical. Don't forget that Perl built-ins ( and natively coded module routines ) are fast, while re-implementing equivalent constructs in perl is slower.

I timed the options, one hundred iterations of each with an array of one million elements. join and local are 6 times as fast as manual looping and 3 times as fast as printing the array outside a string.

use Benchmark;
for ( 1..1000000 ) {
    push @a, "$i";
}
open PROG, ">/dev/nul";

timethese( 100], {
   manual => sub {for $line ( @a ){ print PROG "$line\n";}},
   join   => sub {print PROG join( "\n", @a ), "\n" },
   local  => sub {local $"="\n"; print PROG "@a\n";}
   html  => sub {print PROG @a;}
   });

##### output ->
$ perl t.pl
Benchmark: timing 100 iterations of join, local, manual...
  html:133 wallclock secs (131.79 usr+  0.15 sys = 131.94 CPU) @  0.76
+/s (n=100)
  join: 45 wallclock secs (44.04 usr +  0.01 sys = 44.05 CPU) @  2.27/
+s (n=100)
 local: 45 wallclock secs (44.51 usr +  0.00 sys = 44.51 CPU) @  2.25/
+s (n=100)
manual:275 wallclock secs (274.17 usr+  0.04 sys = 274.21 CPU) @  0.36
+/s (n=100)
[download]