Hashes do not preserve insertion order. In other words, it is simply a fluke that your examples #1 and #2 print out the hash keys in the correct order; they are the exception rather than the rule.

If you just want the keys sorted alphabetically, use

    for my $k (sort keys %GRAMMAR) {
[download]

But you may run into situations where you really do want to preserve insertion order, and it's not just a matter of sorting alphabetically.

If you want to preserve insertion order and still use a hash, take a look at Tie::IxHash. Or, if you're not wedded to a hash, you could always use an array, and push array refs onto your list of grammars:

   push @GRAMMAR, [ $1, $2 ];
[download]

And then when iterating:

    for my $rec (@GRAMMAR) {
        print $rec->[0], "\n";
    }
[download]

BTW, when you do a regex match w/ capturing parens, you really should check whether the regex successfully matches:

    next unless /([A-Z]) -> (.*)/;
    $GRAMMAR{$1} = $2;
[download]

That way, if the regex fails--ie. the line you're looking at doesn't match that regex--you'll just skip that line. You don't want to use $1 and $2 if the match was unsuccessful, because there's no telling what they could contain. :)