We're claiming that the semantics of a Klister program doesn't depend on the order on which the tasks are executed. In the current implementation, that order is always the same. One good test of that claim would be to randomize the task queue at every iteration and see if the tests start failing.
For performance reasons, we should probably make this behaviour configurable and off by default, but we could turn it on in CI.
We're claiming that the semantics of a Klister program doesn't depend on the order on which the tasks are executed. In the current implementation, that order is always the same. One good test of that claim would be to randomize the task queue at every iteration and see if the tests start failing.
For performance reasons, we should probably make this behaviour configurable and off by default, but we could turn it on in CI.