G'day Excalibor,

All the suggestions so far have been fantastic, and it sounds like all you really need now is a very-fast 'discard line' subroutine.

Be aware that regardless of how efficient your code may be, you'll be limited by the speed of the I/O operations provided by your operating system. If you've got to read 380Mb from disk, that's going to take some time regardless of how you process it.

If possible, set your program running and take a look at what your system is doing. If you're on a unix-flavoured system, then top and time can help a lot. If you're hitting 100% CPU usage, and a lot of that is in userland time, then a tigher reading-loop may help. If you're not seeing 100% CPU usage, or you're seeing a very high amount of system time, then you're probably I/O bound. You'll need faster disks, hardware, and/or filesystems for your program's performance to improve.

Assuming that you are CPU bound, you can potentially write your 'discard line' subroutine in C, which allows it to be very fast and compact. Here's an example using Inline::C

use Inline 'C';

# Example, skip a line of input from STDIN:

skip_line();

# Look!  The next line is read fine by Perl.

print scalar <STDIN>;

__END__
__C__

/* Read (and discard) until we find a newline */
/* NOTE: This will loop endlessly if it hits EOF
 *       before finding a newline. Caveat lector. */

void skip_line() {
    while( getchar() != '\n' ) { }
}
[download]

I haven't benchmarked that, but it should be both very memory efficient and fast. Be aware the of the problem that you will encounter if skip_line() hits EOF before a newline; unless you're very sure of your input file you'll want to improve upon the sample code provided here.

If you do benchmark, keep in mind that any caching by the CPU may make a significant difference to your end results.

All the very best,

How do you know if you've skipped over a newline?

Be well,
rir

If you're reading each line as soon as it is appended to the file, there won't be a newline.

But if there's a chance that another line will follow more quickly than your seek gets there, you could always read backwards $MAX bytes to see if there's a newline.

---

Caution: Contents may have been coded under pressure.

Maybe you could try replacing the output file with a named pipe (man mkfifo). The program outputs to the pipe, and you have a filter program read from the pipe.

Or depending on the predictability/frequency of the output, you could compare the file size once, then later if the file size isn't greater than $MAX_DIFFERENCE, you know you don't have to worry about it.

Or patch the program to not insert long lines.