Then I even looked at 'const36a', but it seems that this version doesn't have this bug (probably already fixed, because it's a 2 years later version) - the code (around 3C6E +/- a few bytes) is in fact almost identical to your fixed code.