And thank you again, ikegami.

The need for the synchonizing code with sysread came from graff's observation that he needed to add a sleep 1; the first line or so of output gets lost unless the main script waits a bit before it sends anything.

I hadn't thought about spaces embedded in the password, but it would only be a problem if the script read more than one byte at a time. Using getc, $_ will never contain embedded whitespace, because it would have been found on a previous pass through the loop.

Last night you pointed out that checking for /\s/ (or /\s$/) doesn't work if the password contains spaces. After I though about it for a while, I think it can be fixed by changing

next unless /\s/;
[download]

to see if $_ is the beginning of the password

next if $_ eq substr "' . (quotemeta $password) . '", 0, length;
[download]

when reading one character at a time.

If it was really necessary to read more than one character at a time and the passwords would contain spaces, it would probably work to check if any trailing substring of the buffer matches any leading substring of the password like this:

[[... beginning of the code as above...]]
print '
  use warnings;
  use strict;
  use List::Util qw(min);
  select STDOUT;
  $| = 1;
  my $chr;
  $_ = "";
CHAR:
  while (defined ($chr = getc DATA)) {
      $_ .= $chr;
      foreach my $n (1..min ' . (length $password) . ', length) {
          next CHAR if substr ("' . (quotemeta $password) . '", 0, $n)
            eq substr ($_, (length)-$n, $n);
      }
      s/' . (quotemeta $password) . '/removed/ig;
      print;
      $_ = "";
  }
  print;
  __DATA__
';
[[... the rest of the code is as above...]]
[download]

... but I think I prefer the simplicity of reading one character at a time.

Take care,

quester