This solution has been tested on both AIX and Win98. Its advantages are:

1. only holds one line in memory, memory usage is the minimum
2. only goes thru the file once, extreamly fast

One key point to help understand the source code: read from the beginning to the end, and write from the end to the beginning.

#usage perl -w test.pl oldfile newfile
use Fcntl qw(SEEK_SET SEEK_END);
use File::Copy;

use strict;
use constant NEWLINE_ADJUSTMENT => 0; #set this to 1 on Win98, 0 on un
+ix!

copy($ARGV[0], $ARGV[1]);
open(AFILE, "<", $ARGV[0]);
open(BFILE, "+<", $ARGV[1]);

seek(AFILE, 0, SEEK_END);
my $len = tell(AFILE);
my $written_len = NEWLINE_ADJUSTMENT;
seek(AFILE, 0, SEEK_SET);

while (<AFILE>) {
    seek(BFILE, $len - $written_len - length(), SEEK_SET);
    $written_len += length() + NEWLINE_ADJUSTMENT;
    print BFILE;
}

close(AFILE);
close(BFILE);
[download]