SUBSTR OR REGEX: WHICH WILL YOU FAVOUR?

[Anonymous Monk]

Recently, I wrote a script for an individual, which is to take three consecutive letter or number from a line of number or string, until the whole line is "taken care of"{ I hope you get that.}. Of course, I rolled out my while loop, with a regex to "pick out" these three consecutive letters like:

...
        $str .= $1 . $/ while ( $ARGV[0] =~ m/(.{3})/g );
...

Then, I was told NO, please can you use substr function? Hummm! Ok I will. Though that doesn't come to me naturally. I felt like, why become verbose when you can be simple and "efficient"? So, I did.

...
use constant LEN => 3;
...
        my $offset = 0;
        my $length = LEN;

        while ( $_ = substr $ARGV[0], $offset, $length ) {
            last if length != LEN;
            $str .= $_ . $/;
            $offset += $length;
        }
...

Of course, it worked, and the input is coming directly from the CLI - Command Line Interface. However, I decided on my own to Benchmark these. Though, I know that substr should be faster than regex usage. So, I called up the benchmark module and did this:

#!usr/bin/perl -w
use strict;
use constant LEN => 3;
use Benchmark qw(:all);

my $str;
my $count = -2;

my $re = timethese(
    $count,
    {
    substring => sub {

        my $offset = 0;
        my $length = LEN;

        while ( $_ = substr $ARGV[0], $offset, $length ) {
            last if length != LEN;
            $str .= $_ . $/;
            $offset += $length;
        }
    },

    regex => sub {

        $str .= $1 . $/ while ( $ARGV[0] =~ m/(.{3})/g );
    }
  }
);

cmpthese($re);

And my result confirmed by pre-informed mind. With this:

Benchmark: running regex, substring for at least 2 CPU seconds...
     regex:  1 wallclock secs ( 2.09 usr +  0.00 sys =  2.09 CPU) @ 14
+030.14/s (n=29323)
 substring:  3 wallclock secs ( 2.14 usr +  0.00 sys =  2.14 CPU) @ 23
+225.70/s (n=49703)
             Rate     regex substring
regex     14030/s        --      -40%
substring 23226/s       66%        --

Oh! Yea! So, I thought. But somehow, I check the script and and changed this

...
        $str .= $1 . $/ while ( $ARGV[0] =~ m/(.{3})/g ); ## NOTE the 
+number 3
...

to this

...
        $str .= $1 . $/ while ( $ARGV[0] =~ m/(.{LEN})/g ); ## NOTE 3 
+is now LEN
...

Since the constant LEN is 3, then I got this surprise:

Benchmark: running regex, substring for at least 2 CPU seconds...
     regex:  4 wallclock secs ( 2.72 usr +  0.01 sys =  2.73 CPU) @ 74
+0317.95/s (n=2021068)
 substring:  2 wallclock secs ( 2.66 usr +  0.02 sys =  2.68 CPU) @ 17
+332.84/s (n=46452)
              Rate substring     regex
substring  17333/s        --      -98%
regex     740318/s     4171%        --

Common on?! Is the benchmark module broke or something? lol!!. Or I didn't write the benchmark script well? Anyway like am back in love with simple and precise over verbose. What do you think?
~ zadok_the_priest ~