I've come across a problem when using the perl redo statement. I created a small script (see below) which reproduces the problem.

On a Linux machine with perl 5.12.0, the script works; it has problems on an AIX machine with perl 5.8.2.

#!/usr/local/bin/perl
use strict;

my ($file) = @ARGV;

open( FH, "<", $file ) or die "$!";

while( my $line = <FH> ) {
        chomp( $line );

        if( $line =~ /^#/ ) {
                redo;
        }
        else {
                print "Doesn't begin with #\n";
        }
}
[download]

Note: the file that is opened and read contains a single '#' character in the first position.

The "correct" case: perl 5.12.0

When running in the perl debugger, and watching $line, after the redo statement executes, the code correctly jumps to the chomp statement, without reevaluating the while-condition, and without anything happening to the $line variable. (Note that, with this code sample, this will induce an endless loop. In the 'real' code, there are other things happening which would prevent this.)

The "weird" case: perl 5.8.2

When stepping through with the debugger here (same code, same input file), and watching $line, after the redo statement executes, and before executing the chomp statement, this displays:

main::(jt:12):                  redo;
  DB<2>
Watchpoint 0:   $line changed:
        old value:      '#'
        new value:      ''
main::(jt:9):           chomp( $line );
[download]

Notice how the $line just changed for no reason (from pound to undef).

I understand that the ideal thing here is to upgrade to 5.12 across the board, but there are many machines involved and that's not possible at the moment.

Am I hitting something known? I've Googled around for "perl redo bug", etc, but cannot locate anything.

UPDATE

By taking the my out of the while-condition, and declaring my $line just above the while condition, this seems to fix the problem. However, I still wonder why.