Many ways to do this. I'm going to read the text file into a data structure that mirrors the output, rather than mirroring what the input looks like. This way, I'm going to parse directly into the records. What I could have done was to read into a series of hashes instead. This just proves it's doable:

use strict;
use Text::CSV;

my @data;
my %header;

{
    my $idx = 0;
    my @single_row;

    sub eos
    {
        if (@single_row)
        {
            push @data, [ @single_row ];
            @single_row = ();
        }
    }

    sub index_for
    {
        my $key = shift;
        unless (exists $header{$key})
        {
            $header{$key} = $idx++;
        }
        $header{$key};
    }

    while(<DATA>)
    {
        chomp;
        if (/^$/)
        {
            # end of section
            eos();
        }
        else
        {
            my ($key, $value) = split /\s*:\s*/, $_, 2;
            $single_row[index_for($key)] = $value;
        }
    }
    eos();
}

my $csv = Text::CSV->new();

my @h = sort { $header{$a} <=> $header{$b} } keys %header;
$csv->combine(@h);
print $csv->string(), $/;

$csv->combine(@$_), print $csv->string(), $/ foreach @data;


__END__
a: a1
b: b1
c: c1

a: a2
b: b2
c: c2

a: a3
b: b3
c: c3

a: a4
b: b4
c: c4
d: d1
[download]

Excellent, that works out very well! Just needed a couple modifications to handle input/output files and dos<-->unix filetype conversion, and now it's rocking. Thanks a lot!

The Fallen Monkey