Test coverage missing for starting the actor when it is in stopped state

[thedrow]

Currently, it is possible to transition from the stopped state back to the starting state using the start trigger. However, there is currently no test that verifies that such a transition is possible.

The test should start() the actor, stop() it and then start() it again.

[ahopkins]

Would there be a use case to add a .restart()?

[ahopkins]

I am not so sure this makes sense:

await fake_actor.start()
assert fake_actor.state == ActorState.started

await fake_actor.stop()
assert fake_actor.state == ActorState.stopped

await fake_actor.start()
assert fake_actor.state == ActorState.starting

This is not intuitive that it would be in starting and not mirror the first transition. What am I missing?

[thedrow]

Would there be a use case to add a .restart()?

Yes there is but there's no issue about it yet other than a TODO task in #3.

[thedrow]

I am not so sure this makes sense:

await fake_actor.start()
assert fake_actor.state == ActorState.started

await fake_actor.stop()
assert fake_actor.state == ActorState.stopped

await fake_actor.start()
assert fake_actor.state == ActorState.starting

This is not intuitive that it would be in starting and not mirror the first transition. What am I missing?

See the description of how the transitions work in #3. Essentially, after each transition, we call start() or stop() again to progress to the next state.

[ahopkins]

Right, but my point is this:

fake_actor = FakeActor()

assert fake_actor.state == ActorState.initializing

await fake_actor.start()
assert fake_actor.state == ActorState.started

await fake_actor.stop()
assert fake_actor.state == ActorState.stopped

await fake_actor.start()
assert fake_actor.state == ActorState.starting

await fake_actor.start()
assert fake_actor.state == ActorState.started

There is not a consistent progression of states from -ing to -ed

[thedrow]

The progression is internal. You will be eventually able to hook callbacks on those states (and transitions occasionally) to perform custom startup/shutdown activities.

[ahopkins]

A more intuitive API would be this:

fake_actor = FakeActor()

assert fake_actor.state == ActorState.initialized

await fake_actor.start()
assert fake_actor.state == ActorState.started

await fake_actor.stop()
assert fake_actor.state == ActorState.stopped

await fake_actor.start()
assert fake_actor.state == ActorState.started

With the initializing state hidden inside internals. And, the same for the second .start() >> starting.

Have you considered something like this:

fake_actor = FakeActor()

assert fake_actor.state == ActorState.initializing

await fake_actor.init()
assert fake_actor.state == ActorState.initialized

async with fake_actor.start():
    assert fake_actor.state == ActorState.starting

assert fake_actor.state == ActorState.started

async with fake_actor.stop():
    assert fake_actor.state == ActorState.stopping

assert fake_actor.state == ActorState.stopped

await fake_actor.start()
assert fake_actor.state == ActorState.started

await fake_actor.stop()

[thedrow]

We can't do that since the transitions library doesn't allow us to do so. The start() and stop() triggers are automatically attached to the Actor when we cls._state_machine.add_model(self) and they are defined by the state machine itself.

In any case, this is not needed and incorrect. The assertion would fail. This is the correct assertion with your API:

async with fake_actor.start():
    assert fake_actor.state == ActorSartingState.tasks_started

assert fake_actor.state == ActorState.started

Only the initial transition from initialized to starting sets the state to starting. Afterward, we transition into a sub-state of starting. See diagram at #3 or refer to the table with the description of states.