1. There is no overflow check (maximum 5-tuple should be 's8W-!', but there are more efficient means of checking for an overflow as the result is calculated).
Thanks for pointing this out. I will care about it in the new decoder implementation. 2. The memory handling is insufficient; in the current implementation it is trivial to overflow the allocated memory - for example, just make the input "zzz~>" - a perfectly valid stream. The old growing-buffers implementation is being replaced.
