I needed to calculate the number of consonant bigrams in English monomorphemic words from CELEX. The hash ‘bigrams’ contains the 30^2 possible consonant clusters given the DISC transcriptions, the transcription has already had stress and syllabification stripped from it. The first version of the code did it this way:

foreach my $bg ( keys( %bigrams )) {
    $bigrams{ $bg }++ while ( $PhonStrsDISC =~ /$bg/g );
}

that loop looks for 900 unique strings, globally, in each of about 12,000 words and time looks like this:

real    0m58.995s
user    0m58.254s
sys     0m0.221s

the second version does it this way:

my @phones = split( //, $PhonStrsDISC );
for ( my $i = 0; $i < $#phones; $i++ ) {
    my $bg = $phones[ $i ] . $phones [ $i + 1 ];
    $bigrams{ $bg }++ if ( exists( $bigrams{ $bg } ));
}

I get exactly the same output. The time differences?!?!

real    0m0.730s
user    0m0.706s
sys     0m0.026s

moral of the story: if you’re going to be an idiot, at least be a moderately efficient idiot. What I find ironic here is that the first pass is about as idiomatic as perl comes whiles the second is slightly less perlish (apart from the heavy reliance on split() and hashes, I mean). Normally I think of these well-worn perl idioms as trading accessibility to outsiders for speed.