edfsm - Event Driven Finite State Machine

An Event Driven Finite State Machine is a useful formalism for control and monitoring applications. The key concepts are:

• The state maintained by the state machine represents aspects of its external environment and its history.
• An event represents a change in the environment and is one form of input to the state machine. Reception of an event may cause the state to be updated.
• A command is an input to the state machine that may cause it to perform an effect. An event may also be produced by the command which may update the state.
• An effect is some action that influences the environment.

The effect, if any, produced by a command will depend on both the command and the a-priori state. This models the proactive behaviour of a Mealy machine. An effect may also be produced after and event is processed and it will depend on the a-posteriori state. This models the reactive behaviour of a Moore machine.

Why edfsm?

edfsm, and its DSL in particular, help you identify the functions required to handle commands and events given declared states, and strongly type their declarations. In short, edfsm is designed to enhance the code quality of your state machine by leveraging the compiler to assert your declaration of its transitions.

DSL

An attribute macro has been provided that provides a Domain Specific Language (DSL) mapping directly from a Finite State Machine description to code. The goal of the DSL is to convey the Finite State Machine in a way that can closely match its design. Given the macro, compilation also ensures that the correct state, command and event types are handled by the developer.

Here is an example given the declaration of states, commands, events and an effect handler:

struct MyFsm;

#[impl_fsm]
impl Fsm for MyFsm {
    type S = State;
    type C = Command;
    type E = Event;
    type SE = EffectHandlers;

    state!(Running / entry);

    command!(Idle    => Start => Started => Running);
    command!(Running => Stop  => Stopped => Idle);

    ignore_command!(Idle    => Stop);
    ignore_command!(Running => Start);
}

The state! macro declares state-related attributes. At this time, entry handlers can be declared. In our example, the macro will ensure that a on_entry_running method will be called for MyFsm. The developer is then required to implement these methods e.g.:

fn on_entry_running(_old_s: &Running, _se: &mut EffectHandlers) {
    // Do something
}

The command! macro declares what should happen given a command using the form:

<from-state> => <given-command> [=> <yields-event> [=> <to-state>]]

When declaring states it is also possible to use a wildcard i.e. _ in place of <from-state> and <to-state>.

In our example, for the first step declaration, multiple methods will be called that the developer must provide e.g.:

fn for_idle_start(_s: &Idle, _c: Start, _se: &mut EffectHandlers) -> Option<Started> {
    // Perform some effect here if required. Effects are performed via the EffectHandler
    Some(Started)
}

fn on_idle_started(_s: &Idle, _e: &Started) -> Option<Running> {
    Some(Running)
}

Note that steps may also be declared for events using a event! macro (not shown). The form then becomes:

<from-state> => <given-event> [=> <to-state> [ / action]]

(/ action can be used to declare that a side-effect is to be performed)

The ignore_command! macro describes those states and commands that should be ignored given:

<from-state> => <given-command>

Note if no ignore_command! declarations are provided then exhaustive matching on states and commands is not enforced.

There is a ignore_event! macro available for ignoring events where events are providing the input.

State machines are then advanced given a mutable state and command. An optional event can be emitted along with a possible state transition e.g.:

let mut s = State::Idle(Idle);
let c = Command::Start(Start);
// Now step the state machine with the state and command,
// and, an (undeclared) effect handler.
let (e, t) = MyFsm::step(&mut s, Input::Command(c), &mut se);

State can also be re-constituted by replaying events. If there is no transition to an entirely new state then the existing state may still have been updated. Here is an example of applying an event to state with the update of state if necessary and an option of r indicating some type of change that occurred, or None otherwise.

let r = MyFsm::on_event(&mut s, &e);

Mutating state can be very useful where a state itself represents a finer granularity of state with its fields, and so we wish to update them directly. For example, given our previous representation of:

command!(Running => Stop  => Stopped => Idle);

...if we change it to:

command!(Running => Stop  => Stopped);

i.e. if we remove the target state, then the associated function will be able to mutate the state and no transition can be returned as they are mutually exclusive actions. Here is a sample signature in accordance with the above command!.

fn on_idle_started(s: &mut Idle, e: &Started) {
    // `s` can now be mutated given some `e`.
}

Please see the event_driven/tests folder for complete examples, including the ability to mutate the passed state in the absence of a target state i.e. when emitting an event but not transitioning.

edfsm-machine and edfsm-kv-store

These two crates provide a means to conveniently run an FSM of edfsm as a future, and also to use an FSM as a key/value store respectively.

no_std

The library is able to supportno_std and is designed for efficient usage with embedded targets.

Contribution policy

Contributions via GitHub pull requests are gladly accepted from their original author. Along with any pull requests, please state that the contribution is your original work and that you license the work to the project under the project's open source license. Whether or not you state this explicitly, by submitting any copyrighted material via pull request, email, or other means you agree to license the material under the project's open source license and warrant that you have the legal authority to do so.

License

This code is open source software licensed under the Apache-2.0 license.

© Copyright Titan Class P/L, 2022-2024