I've hit a performance issue with using | in regexes. It seems that in some (not-so-degenerated, actually) cases it loses significantly to looping over simple regexes i.e. $str =~ /$_// for @rx is much faster than $str =~ /$rx[0]|$rx[1]|$rx[2]/, which is rather counter-intuitive. Basically, for general cases, it would mean that alterations with grouping should be avoided at all, which is a strong statement and I wouldn't like it that way.

Is this a recognized problem? Is it a problem at all? Does it look like it needs to be reported as a bug? I can't decide myself.

Here's the test code:

use strict;
use warnings;
use Benchmark qw(:all);

my $str = 'a' x 100;

my @matchwords = qw(
aol aachen aaliyah aaron abbas abbasid abbott abby abdul abe abel abel
+ard abelson
aberdeen abernathy abidjan abigail abilene abner abraham abram abrams 
+absalom abuja
abyssinia abyssinian ac acadia acapulco accra acevedo achaean
);

my $q1s = join('|', map { "$_\\s*\\w+" } @matchwords );
my $q1  = qr/$q1s/;

my @q2s = map { "$_\\s*\\w+" } @matchwords;
my @q2  = map { qr/$_/ } @q2s;

my $q3s = join('|', map { "($_)\\s*\\w+" } @matchwords );
my $q3  = qr/$q3s/;

my @q4s = map { "($_)\\s*\\w+" } @matchwords;
my @q4  = map { qr/$_/ } @q4s;

timethese( 100000, {
    'alternation, no grouping' => sub {
        $str =~ /$q1/;
    },
    'loop, no grouping' => sub {
        for my $qr ( @q2 ) {
            $str =~ /$qr/;
        }
    },
    'alternation, grouping' => sub {
        $str =~ /$q3/;
    },
    'loop, grouping' => sub {
        for my $qr ( @q4 ) {
            $str =~ /$qr/;
        }
    },
});
[download]

Here's the output:

Benchmark: timing 100000 iterations ...
alternation, grouping: 12 wallclock secs (11.92 usr +  0.00 sys = 11.9
+2 CPU) @ 8389.26/s (n=100000)
alternation, no grouping:  0 wallclock secs ( 0.19 usr +  0.00 sys =  
+0.19 CPU) @ 526315.79/s (n=100000)
            (warning: too few iterations for a reliable count)
loop, grouping:  2 wallclock secs ( 1.33 usr +  0.00 sys =  1.33 CPU) 
+@ 75187.97/s (n=100000)
loop, no grouping:  1 wallclock secs ( 1.33 usr +  0.00 sys =  1.33 CP
+U) @ 75187.97/s (n=100000)
[download]

Update: got same results on perls 5.10.1, 5.16.0, and 5.17.6