I hate it when ... invalid models

I hate it when my internal model of a program doesn't quite match its actual behavior.

So I'm tailing a file: "tail -f file".  Tail recognizes when a file's been truncated and starts rereading it.  BUT, even if you say "-n +0" it still starts rereading from the new end, not from the beginning.  So if you have

# Terminal 1
date > file
tail -f -n +0 file

# Terminal 2
date > file
date > file
date > file

You'll never see anything new from tail, because the length of the file never actually changes.

More subtly, say you have

# Terminal 1
date > file
date >> file
tail -f -n +0 file

# Terminal 2
date > file

Here the tail will see that you truncated the file, but it won't show you the new line, because the file length changed from 58 to 29, not from 58 to 0 to 29.

All that's fine and maybe even "obvious" when laid out like that.  But you can get the same behavior like this:

# Program 1, pseudo-Perl
open LOGFILE, ">file";
while (my $request = get_request()) {
    print LOGFILE "the request\n";
    process( $_ );
}

# Terminal 1
tail -f -n +0 file

When you restart Program 1, you won't see the first request in the log file, because just as above, the file length went from (say) 100 to 50, not 100 to 0 to 50.

But if you cat the file, there it is.

Anyway, I hate it when that happens.

(The answer, by the way, is to unlink the file before you start writing to it again, and tell tail "--follow=name".)