SM serves as one of the building blocks for an open-source game about space engineering and exploration. This library is in active use and development.
SM aims to be a safe, fast and simple state machine library.
-
safe β Rust's type system, ownership model and exhaustive pattern matching prevent you from mis-using your state machines
-
fast β zero runtime overhead, the machine is 100% static, all validation happens at compile-time
-
simple β five traits, and one optional declarative macro, control-flow only, no business logic attached
Using this library, you declaratively define your state machines as as set of states, connected via transitions, triggered by events. You can query the current state of the machine, or pattern match against all possible machine variants.
The implementation ensures a zero-sized abstraction that uses Rust's type-system and ownership model to guarantee valid transitions between states using events, and makes sure previous states are no longer accessible after transitioning away to another state. Rust validates correct usage of the state machine at compile-time, no runtime checking occurs when using the library.
The library exposes the sm!
macro, which allows you to declaratively build
the state machine.
extern crate sm;
use sm::sm;
sm! {
Lock {
InitialStates { Locked, Unlocked }
TurnKey {
Locked => Unlocked
Unlocked => Locked
}
Break {
Locked, Unlocked => Broken
}
}
}
fn main() {
use Lock::*;
let lock = Machine::new(Locked);
let lock = lock.transition(TurnKey);
assert_eq!(lock.state(), Unlocked);
assert_eq!(lock.trigger().unwrap(), TurnKey);
}
The below example explains step-by-step how to create a new state machine using the provided macro, and then how to use the created machine in your code by querying states, and transitioning between states by triggering events.
First, we import the macro from the crate:
extern crate sm;
use sm::sm;
Next, we initiate the macro declaration:
sm! {
Then, provide a name for the machine, and declare a list of allowed initial states:
Lock {
InitialStates { Locked, Unlocked }
Finally, we declare one or more events and the associated transitions:
TurnKey {
Locked => Unlocked
Unlocked => Locked
}
Break {
Locked, Unlocked => Broken
}
}
}
And we're done. We've defined our state machine structure, and the valid transitions, and can now use this state machine in our code.
You can initialise the machine as follows:
let sm = Lock::Machine::new(Lock::Locked);
We can make this a bit less verbose by bringing our machine into scope:
use Lock::*;
let sm = Machine::new(Locked);
We've initialised our machine in the Locked
state. You can get the current
state of the machine by sending the state()
method to the machine:
let state = sm.state();
assert_eq!(state, Locked);
While you can use sm.state()
with conditional branching to execute your
code based on the current state, this can be a bit tedious, it's less
idiomatic, and it prevents you from using one extra compile-time validation
tool in our toolbox: using Rust's exhaustive pattern matching requirement to
ensure you've covered all possible state variants in your business logic.
While sm.state()
returns the state as a unit-like struct (which itself is
a ZST, or Zero Sized Type), you can use the sm.as_enum()
method to get
the state machine back as an enum variant.
Using the enum variant and pattern matching, you are able to do the following:
use Lock::Variant::*;
match sm.as_enum() {
InitialLocked(m) => {
assert_eq!(m.state(), Locked);
assert!(m.trigger().is_none());
}
InitialUnlocked(m) => {
assert_eq!(m.state(), Unlocked);
assert!(m.trigger().is_none());
}
LockedByTurnKey(m) => {
assert_eq!(m.state(), Locked);
assert_eq!(m.trigger().unwrap(), TurnKey);
}
UnlockedByTurnKey(m) => {
assert_eq!(m.state(), Unlocked);
assert_eq!(m.trigger().unwrap(), TurnKey);
}
BrokenByBreak(m) => {
assert_eq!(m.state(), Broken);
assert_eq!(m.trigger().unwrap(), Break);
}
}
Each state configured with InitialStates
has its own variant named
Initial<State>
. Next to those, each valid state + event combination also
has its own variant, named <state>By<event>
.
The compiler won't be satisfied until you've either exhausted all possible enum variants, or you explicitly opt-out of matching all variants, either way, you can be much more confident that your code won't break if you add a new state down the road, but forget to add it to a pattern match somewhere deep inside your code-base.
To transition this machine to the Unlocked
state, we send the transition
method, using the TurnKey
event:
let sm = sm.transition(TurnKey);
assert_eq!(sm.state(), Unlocked);
Because multiple events can lead to a single state, it's also important to
be able to determine what event caused the machine to transition to the
current state. We can ask this information using the trigger()
method:
assert_eq!(sm.trigger().unwrap(), TurnKey);
The trigger()
method returns None
if no state transition has taken place
yet (ie. the machine is still in its initial state), and Some(Event)
if
one or more transitions have taken place.
It's important to realise that we've consumed the original machine in the
above example when we transitioned the machine to a different state, and got
a newly initialised machine back in the Unlocked
state.
This allows us to safely use the machine without having to worry about multiple readers using the machine in different states.
All these checks are applied on compile-time, so the following example would fail to compile:
let sm2 = sm.transition(TurnKey);
assert_eq!(sm.state(), Locked);
This fails with the following compilation error:
error[E0382]: use of moved value: `sm`
--> src/lib.rs:315:12
|
22 | let sm2 = sm.transition(TurnKey);
| -- value moved here
23 | assert_eq!(sm.state(), Locked);
| ^^ value used here after move
|
= note: move occurs because `sm` has type `Lock::Machine<Lock::Locked>`, which does not implement the `Copy` trait
Similarly, we cannot execute undefined transitions, these are also caught by the compiler:
assert_eq!(sm.state(), Broken);
let sm = sm.transition(TurnKey);
This fails with the following compilation error:
error[E0599]: no method named `transition` found for type `Lock::Machine<Lock::Broken>` in the current scope
--> src/lib.rs:360:13
|
4 | sm! {
| --- method `transition` not found for this
...
25 | let sm = sm.transition(TurnKey);
| ^^^^^^^^^^
|
= help: items from traits can only be used if the trait is implemented and in scope
= note: the following trait defines an item `transition`, perhaps you need to implement it:
candidate #1: `sm::Transition`
The error message is not great (and can potentially be improved in the
future), but any error telling you transition
is not implemented, or the
passed in event type is invalid is an indication that you are trying to
execute an illegal state transition.
Finally, we are confined to initialising a new machine in only the states
that we defined in InitialStates
:
let sm = Machine::new(Broken);
This results in the following error:
error[E0277]: the trait bound `Lock::Broken: sm::InitialState` is not satisfied
--> src/lib.rs:417:10
|
21 | let sm = Machine::new(Broken);
| ^^^^^^^^^^^^ the trait `sm::InitialState` is not implemented for `Lock::Broken`
|
= note: required because of the requirements on the impl of `sm::NewMachine<Lock::Broken>` for `Lock::Machine<Lock::Broken>`
And that's it! There's nothing else to it, except a declarative β and easy to read β state machine construction macro, and a type-safe and ownership-focused way of dealing with states and transitions, without any runtime overhead.
Go forth and transition!
Licensed under either of
- Apache License, Version 2.0, (LICENSE-APACHE or https://www.apache.org/licenses/LICENSE-2.0)
- MIT license (LICENSE-MIT or https://opensource.org/licenses/MIT)
Unless you explicitly state otherwise, any contribution intentionally submitted for inclusion in the work by you, as defined in the Apache-2.0 license, shall be dual licensed as above, without any additional terms or conditions.