The AoA's is not a fundemental part of the algorithm, The version below is nearly identical except fro it reads the offset pairs from <DATA>, one pair per line instead of from an AoA.

The only memory constraint then becomes the length of the string $result, which is defined (though not automatically detected) by the largest offset.

As the process uses just one pass of two simple loops--one to read the offsets and build the result string, one to read the results string and extract the combined offsets--it should be considerably faster than any algorithm that use multiply recursive passes. You don't even need to sort the input offsets, and it outputs the combined offsets already sorted.

Hope it proves of some use.

Slightly modified code

#! perl -slw
use strict;
my $result = "\0" x 100;
while (my ($start, $end) = split ' ', <DATA>) {
    substr($result, $start, $end - $start + 1) = 1 x ($end - $start + 
+1);
}
my (@results, $start);
for (0 .. length $result) {
    my $c = substr($result, $_, 1);
    next if not $start and $c ne '1';
    $start = $_ if $c eq '1' and not $start;
    next if $start and $c eq '1';
    push @results, [$start, $_-1];
    $start = undef;
}
print map{local $"='-'; "[@$_] "}@results; #!"

__END__
17 19
34 39
26 29
53 57
43 47
58 59
40 45
30 33
20 24
10 15
6 9
1 4
35 45
7 15
[download]