Re: global match except last one

[jpl]

Always curious, I wondered if the apparently complex lookahead logic proposed by ikegami and happy.barney would actually save time over your original approach, ignoring, as we should not, that your original approach made some bold assumptions about where the final underscore would appear. I factored out the final uppercasing to make the code more nearly comparable, and came up with

use Benchmark('countit');

$code = '$str =~ s/(?<=[a-z])(?=[A-Z])/_/g';
$t = countit(5, '$str="BuyACaseOfCamels";' . $code);
$count = $t->iters ;
print "$count loops of $code\n";

$code = '$str =~ s/([a-z]+)(?=(.?))/ $1 . (length($2) ? "_" : "") /ge'
+;
$t = countit(5, '$str="BuyACaseOfCamels";' . $code);
$count = $t->iters ;
print "$count loops of $code\n";

$code = '$str =~ s/([a-z]+)/$1_/g; $str =~ s/_$//';
$t = countit(5, '$str="BuyACaseOfCamels";' . $code);
$count = $t->iters ;
print "$count loops of $code\n";

I was a bit surprised at the results.

  perl ccase.pl
1686588 loops of $str =~ s/(?<=[a-z])(?=[A-Z])/_/g
1194666 loops of $str =~ s/([a-z]+)(?=(.?))/ $1 . (length($2) ? "_" : 
+"") /ge
1520477 loops of $str =~ s/([a-z]+)/$1_/g; $str =~ s/_$//

  perl ccase.pl
1793776 loops of $str =~ s/(?<=[a-z])(?=[A-Z])/_/g
1286561 loops of $str =~ s/([a-z]+)(?=(.?))/ $1 . (length($2) ? "_" : 
+"") /ge
1466174 loops of $str =~ s/([a-z]+)/$1_/g; $str =~ s/_$//

  perl ccase.pl
1760558 loops of $str =~ s/(?<=[a-z])(?=[A-Z])/_/g
1336044 loops of $str =~ s/([a-z]+)(?=(.?))/ $1 . (length($2) ? "_" : 
+"") /ge
1492832 loops of $str =~ s/([a-z]+)/$1_/g; $str =~ s/_$//

Over three runs, the counts varied slightly, but happy.barney's code consistently outperformed your original code, and ikegami's code was only slightly less peppy than yours, a fair tradeoff for doing a better job of trimming the final underscore. I, for one, am impressed at how well the regular expression engine can perform.