Stage::run::<StateTransitionStage>() causes the initial state to exit and enter again on first run

[tjamaan]

By setting NextState on first run in line 122 in state.rs the initial state will be switched to itself, thus triggering the exit and enter stages. If exit and enter are expensive (e.g. procedural generation) it will cause needless processing.

Link:

iyes_loopless/src/state.rs

Line 122 in cf8e61b

world.insert_resource(NextState(self.init_state.clone()));

[tjamaan]

Ways to fix this:

1. Treat switching a state to itself as an edge-case and require users to set another resource (e.g. insert_resource(NextState(same_state)) + insert_resource(SameStateSwitch(true))) in order to execute exit and enter stages.
   1.1. Pros: Only need to check for resource SameStateSwitch(true) if current and new state are equal.
   1.2. Cons: Clunky if you need to switch state to itself intentionally. New users need to discover the need for SameStateSwitch resource.
2. Turn NextState<T>(pub T) into NextState<T>(pub Option<T>).
   2.1. Pros: Simple to use.
   2.2. Cons: Matching an Option is not free. (It doesn't cost much though)

[thebluefish]

If we're asking the user to insert a resource to transition, I would personally expect the existence of the NextState resource to indicate that a transition should occur. IMO we should simply not insert this resource by default, and remove it once the transition has processed.

[inodentry]

That was the behavior in 0.1. I changed it, because "repeatedly inserting and removing a resource" (effectively "using a resource as an event") felt weird to me. But you are right, why not, it makes sense. In retrospect, if we are gonna use a resource to trigger a transition, that makes more sense. I guess I should bring it back.

[inodentry]

OK, reverted back to the 0.1 states behavior, except now without the next != current check (transitioning to the same state as the current is allowed). The NextState behaves as you described: insert it to trigger a transition, and it gets auto-removed when the transition is performed.

I consider this bug important enough that I decided to release a 0.3.0 immediately, just for this.

[tjamaan]

I confirmed that the fix works. Only the initial state's enter stage is run at the start, and it only ran once.