So you're saying on two runs with the same source code and the same environment, it may produce different opcodes?