Running multiple statecharts

It is not unusual to have to deal with multiple distinct components in which the behavior of a component is driven by things that happen in the other components. One can model such a situation using a single statechart with parallel states, or by plugging several statecharts into one main statechart (see sismic.model.Statechart.copy_from_statechart()). The communication and synchronization between the components can be done either by using active(state_name) in guards, or by sending internal events that address other components.

However, we believe that this approach is not very convenient:

  • all the components must be defined in a single statechart;

  • state name collision could occur;

  • components must share a single execution context;

  • component composition is not easy to achieve

Sismic allows to define multiple components in multiple statecharts, and brings a way for those statecharts to communicate and synchronize via events.

Communicating statecharts

Every instance of Interpreter exposes a bind() method which allows to bind statecharts.


Bind an interpreter (or a callable) to the current interpreter.

Internal events sent by this interpreter will be propagated as external events. If interpreter_or_callable is an Interpreter instance, its queue method is called. This is, if i1 and i2 are interpreters, i1.bind(i2) is equivalent to i1.bind(i2.queue).

This method is a higher-level interface for self.attach. If x = interpreter.bind(...), use interpreter.detach(x) to unbind a previously bound interpreter.


interpreter_or_callable (Union[Interpreter, Callable[[Event], Any]]) – interpreter or callable to bind.

Return type:

Callable[[MetaEvent], Any]


the resulting attached listener.

When an interpreter interpreter_1 is bound to an interpreter interpreter_2 using interpreter_1.bind(interpreter_2), the internal events that are sent by interpreter_1 are automatically propagated as external events to interpreter_2. The binding is not restricted to only two statecharts. For example, assume we have three instances of Interpreter:

assert isinstance(interpreter_1, Interpreter)
assert isinstance(interpreter_2, Interpreter)
assert isinstance(interpreter_3, Interpreter)

We define a bidirectional communication between the two first interpreters:


We also bind the third interpreters with the two first ones.


When an internal event is sent by an interpreter, the bound interpreters also receive this event as an external event. In the last example, when an internal event is sent by interpreter_3, then a corresponding external event is sent both to interpreter_1 and interpreter_2.


Practically, unless you subclassed Interpreter, the only difference between internal and external events is the priority order in which they are processed by the interpreter.

from sismic.interpreter import InternalEvent

# Manually create and raise an internal event

print('Events for interpreter_1:', interpreter_1._select_event(consume=False))
print('Events for interpreter_2:', interpreter_2._select_event(consume=False))
print('Events for interpreter_3:', interpreter_3._select_event(consume=False))
Events for interpreter_1: Event('test')
Events for interpreter_2: Event('test')
Events for interpreter_3: InternalEvent('test')


The bind() method is a high-level interface for attach(). Internally, the former wraps given interpreter or callable with an appropriate listener before calling attach(). You can unbound a previously bound interpreter with detach() method. This method accepts a previously attached listener, so you’ll need to keep track of the listener returned by the initial call to bind().

Example of communicating statecharts

Consider our running example, the elevator statechart. This statechart expects to receive floorSelected events (with a floor parameter representing the selected floor). The statechart operates autonomously, provided that we send such events.

Let us define a new statechart that models a panel of buttons for our elevator. For example, we consider that our panel has 4 buttons numbered 0 to 3.

  name: Elevator buttons
  description: |
    Buttons that remotely control the elevator.
  root state:
    name: active
    parallel states:
      - name: button_0
          - event: button_0_pushed
            action: send('floorSelected', floor= 0)
      - name: button_1
          - event: button_1_pushed
            action: send('floorSelected', floor= 1)
      - name: button_2
          - event: button_2_pushed
            action: send('floorSelected', floor= 2)
      - name: button_3
          - event: button_3_pushed
            action: send('floorSelected', floor= 3)

As you can see in the YAML version of this statechart, the panel expects an event for each button: button_0_pushed, button_1_pushed, button_2_pushed and button_3_pushed. Each of those event causes the execution of a transition which, in turn, creates and sends a floorSelected event. The floor parameter of this event corresponds to the button number.

We bind our panel with our elevator, such that the panel can control the elevator:

from import import_from_yaml
from sismic.interpreter import Interpreter

elevator = Interpreter(import_from_yaml(filepath='examples/elevator/elevator.yaml'))
buttons = Interpreter(import_from_yaml(filepath='examples/elevator/elevator_buttons.yaml'))

# Elevator will receive events from buttons

Events that are sent to buttons are not propagated, but events that are sent by buttons are automatically propagated to elevator:

print('Awaiting event in buttons:', buttons._select_event())  # None

print('Awaiting event in buttons:', buttons._select_event())  # External event
print('Awaiting event in elevator:', elevator._select_event())  # None

buttons.execute(max_steps=2)  # (1) initialize buttons, and (2) consume button_2_pushed
print('Awaiting event in buttons:', buttons._select_event())  # Internal event
print('Awaiting event in elevator:', elevator._select_event())  # External event
Awaiting event in buttons: None
Awaiting event in buttons: Event('button_2_pushed')
Awaiting event in elevator: None
Awaiting event in buttons: InternalEvent('floorSelected', floor=2)
Awaiting event in elevator: Event('floorSelected', floor=2)

The execution of bound statecharts does not differ from the execution of unbound statecharts:

print('Current floor:', elevator.context.get('current'))
Current floor: 2

Synchronizing the clock

Each interpreter in Sismic has its own clock to deal with time (see Dealing with time). When creating an interpreter, it is possible to specify which clock should be used to compute the time attribute of the interpreter. When multiple statecharts have to be run concurrently, it is often convenient to have their time synchronized. This can be achieved (to some extent) by providing a shared instance of a clock to their interpreter.

from import import_from_yaml

elevator_sc = import_from_yaml(filepath='examples/elevator/elevator.yaml')
buttons_sc = import_from_yaml(filepath='examples/elevator/elevator_buttons.yaml')

from sismic.clock import SimulatedClock
from sismic.interpreter import Interpreter

# Create the clock and share its instance with all interpreters
clock = SimulatedClock()
elevator = Interpreter(elevator_sc, clock=clock)
buttons = Interpreter(buttons_sc, clock=clock)


As SimulatedClock is the default clock used in Sismic, we could have written the three last lines of this example as follow:

elevator = Interpreter(elevator_sc)
buttons = Interpreter(buttons_sc, clock=elevator.clock)

We can now execute the statecharts and check their time value.


elevator_step = elevator.execute_once()
buttons_step = buttons.execute_once()


As a single instance of a clock is used by both interpreter, the values exposed by their clocks are obviously the same:

assert elevator.clock.time == buttons.clock.time

However, even if the clock is the same for all interpreters, this does not always mean that the calls to execute_once() are all performed at the same time. Depending on the time required to process the first execute_once, the second one will be called with a delay of (at least) a few milliseconds.

We can check this by looking at the time attribute of the returned steps, or by looking at the time attribute of the interpreter that corresponds to the time of the last executed step:

assert elevator_step.time != buttons_step.time
assert elevator.time != buttons.time

To avoid these slight variations between different calls to execute_once(), Sismic offers a SynchronizedClock whose value is based on another interpreter’s time.

from sismic.clock import SynchronizedClock

elevator = Interpreter(elevator_sc)
buttons = Interpreter(buttons_sc, clock=SynchronizedClock(elevator))

With the help of this SynchronizedClock, it is possible to perfectly “align” the time of several interpreters. Obviously, in this context, we first need to execute the interpreter that “drives” the time:


elevator_step = elevator.execute_once()
buttons_step = buttons.execute_once()


Now we can check that the time of the last executed steps are the same:

assert elevator_step.time == buttons_step.time
assert elevator.time == buttons.time


While the two interpreters were virtually executed at the same time value, their clocks still have different values as a SynchronizedClock is based on the time attribute of given interpreter and not on its internal clock.

assert elevator.clock.time != buttons.clock.time


Because the time of an interpreter is set by the clock each time execute_once() is called, you should avoid using execute() (that repeatedly calls execute_once()) if you want a perfect synchronization between two or more interpreters. In our example, a call to execute() instead of execute_once() for the first interpreter implies that the time value of the second interpreter will equal the time value of the first interpreter after having executed all its macro steps. In other words, the execution of the second interpreter will be synchronized with the execution of the last macro step of the first interpreter in that case.