I was curious about the efficiency of the generated regular expression from my previous post. I run some benchmarks, and the results are somewhat unexpected, at least for me!:

my $len = 300;

my @lines;
push @lines, join('', (map { 'f' . ('o' x rand $len * 1.5), (rand > .8
+ ? '. ' : ' ') } 0..rand 20 ), "\n")
    for 0 .. 1000;

sub make_re {
    my $len = shift;
    my $re = "\\b"
             . join('|', map("\\w(?:\\s[\\s\\w]{$_,}?", reverse 1 .. (
+$len - 3)), "\\w+\\s+")
             . (")" x ($len - 3))
             . "\\w+";
    qr/($re)/;
}

# match maximal length sentence
my $len_minus_two = $len - 2;
sub max { my @m = grep /\s*\b\w(?=\w*\s+\w+)[\w\s]{$len_minus_two,}\w/
+o, @lines }

# match minimal length sentence
my $re = make_re $len;
sub min { my @m = grep /$re/, @lines }

use Benchmark qw(cmpthese);
cmpthese(-1, { max => \&max,
               min => \&min } );

__OUTPUT__
      Rate  max  min
max 24.3/s   -- -26%
min 33.0/s  36%   --
[download]

Note that the two regexps used match different things.