RewindableStateMachine
Rollback-aware state machine implementation.
State machines are often used in games to implement different behaviors. However, most implementations are not prepared for rollbacks. This class provides an extensible implementation that can be used alongside a RollbackSynchronizer.
For a full example, see multiplayer-state-machine.
Creating a state machine
The first step is to add the RewindableStateMachine to your scene. It also
requires a RollbackSynchronizer that manages its state
property. Unless these
conditions are satisfied, an editor warning will be displayed.
Note
Editor warnings are only updated when the node tree changes. Configuration changes don't trigger an update. You may need to reload the scene after fixing a warning, or make a tree change, like deleting and re-adding a node by cutting and pasting.
Notice the RollbackSynchronizer added as a sibling to the
RewindableStateMachine, and having its state
property configured.
Implementing states
States are where the custom gameplay logic can be implemented. Each state must be an extension of the RewindableState class, and added as a child to the RewindableStateMachine.
States react to the game world using the following callbacks:
tick(delta, tick, is_fresh)
is called for every rollback tick.enter(previous_state, tick)
is called when entering the state.exit(next_state, tick)
is called when exiting the state.can_enter(previous_state)
is called before entering the state. The state is only entered if this method returns true.display_enter(previous_state, tick)
is called before displaying the state.display_exit(next_state, tick)
is called before displaying a different state.
You can override any of these callbacks to implement your custom behaviors.
For example, the snippet below implements an idle state, that transitions to other states based on movement inputs:
extends RewindableState
@export var input: PlayerInputStateMachine
func tick(delta, tick, is_fresh):
if input.movement != Vector3.ZERO:
state_machine.transition(&"Move")
elif input.jump:
state_machine.transition(&"Jump")
Transitions are based on node names, i.e. calling transition(&"Move")
will
transition to a state node called Move.
States must be added as children under a RewindableStateMachine to work.
Display State vs State
There's two sets of callbacks for state transition - enter()
/exit()
and
display_enter()
/display_exit()
.
The enter()
/exit()
callbacks are intended for implementing game logic. The
display_enter()
/display_exit()
are intended for implementing presentation
logic - visuals, animations, sound effects, etc.
The same applies to on_state_changed
vs. on_display_state_changed
.
Let's take an example. The game is currently on tick @8. It needs to re-run ticks @0 to @8 during rollback. In these ticks, the player moves a bit, performs a jump, and then stops after moving a bit more:
This will trigger the following state changes:
- Tick@1: Idle -> Moving
- Tick@3: Moving -> Jumping
- Tick@5: Jumping -> Moving
- Tick@8: Moving -> Idle
For each of the above, the on_state_changed
signal will be emitted, and the
enter()
/exit()
callbacks will be triggered.
This makes the above callbacks ideal for game logic, e.g. adding an upward
velocity to the player when they enter the Jumping
state.
Note that the displayed state does not change. Before the rollback loop, the
player's state was Idle
. After the rollback loop, the player's state is also
Idle
. Even though the player has ran and performed a jump, it wouldn't make
sense to change their animation or play any sound effect.
Let's take a different rollback example:
In this case, the display state did change. Before the rollback loop, the
player's state was Moving
. After the rollback loop, the player's state is
Jumping
. It would make sense to change the player's animation and play a
jumping sound effect.
This can be done by using the display state callbacks - the
on_display_state_changed
signal, and the display_enter()
/display_exit()
methods.
Caveats
RewindableStateMachine runs in the rollback tick loop, which means that all the Rollback Caveats apply.
In addition, rollback ticks are only ran for nodes that have known inputs for the given tick, and need to be simulated - either on the server to determine the new state, or on the client to predict. In practice, ticks are usually only ran on the host owning state and the client owning inputs. The rest of the peers use the state broadcast by the host.
This means that transition callbacks are not always ran. This is by design and expected ( see #327 ).
As a best practice, in the enter()
, exit()
callbacks and the
on_state_changed
signal, only change game state - i.e. properties that are
configured as state in RollbackSynchronizer.
To update visuals - e.g. change animation, spawn effects, etc. -, use either
the on_display_state_changed
signal, or the display_enter()
and
display_exit()
callbacks to react to state transitions.