bluescreen's suggestion is a good one if your CSV is otherwise uncomplicated - i.e. no quoted fields.

But for complicated CSV, you really do want to use something like Text::CSV. It is a shame that, even setting Text::CSV's eol option, it can't handle this situation.

Best solution would probably be to pre-process each line, replacing the bare "\n" with another character, parse it with Text::CSV, and then replace it back again. Something like:

use Data::Dumper;
use IO::Scalar;
use Text::CSV;

my $data = "Alice,123\r\n"
    .    "Bob\nSmith,456\r\n";
my $io = IO::Scalar->new(\$data);

$/ = "\r\n";
my $CSV = Text::CSV->new({
    eol      => $/,
    sep_char => q{,},
    binary   => 1,
    });

while (defined(my $line = <$io>))
{
    $line =~ s/\n(.)/\xFF$1/g;
    $CSV->parse($line);
    my @row = map { s/\xFF/\n/g; $_ } $CSV->fields;
    print Dumper(\@row);
}
[download]