Run your CLI programs as state machines with persistence and recovery abilities. When such a program breaks you'll have opportunity to change the external world (create a missing folder, change a file permissions or something) and continue the program from the step it was interrupted on.
Let's toss two coins and make sure they both landed on the same side. We express the behaviour
as two states of our machine. Step logic is implemented in State::next()
methods which
return the next state or None
for the last step (the full code is in examples/coin.rs
).
#[derive(Debug, Serialize, Deserialize, From)]
enum Machine {
FirstToss(FirstToss),
SecondToss(SecondToss),
}
#[derive(Debug, Serialize, Deserialize)]
struct FirstToss;
impl FirstToss {
fn next(self) -> StepResult {
let first_coin = Coin::toss();
println!("First coin: {:?}", first_coin);
Ok(Some(SecondToss { first_coin }.into()))
}
}
#[derive(Debug, Serialize, Deserialize)]
struct SecondToss {
first_coin: Coin,
}
impl SecondToss {
fn next(self) -> StepResult {
let second_coin = Coin::toss();
println!("Second coin: {:?}", second_coin);
ensure!(second_coin == self.first_coin, "Coins landed differently");
println!("Coins match");
Ok(None)
}
}
Then we start our machine like this:
let init_state = FirstToss.into();
let mut engine = Engine::<Machine>::new(init_state)?.restore()?;
engine.drop_error()?;
engine.run()?;
We initialize the Engine
with the first step. Then we restore the previous state if the
process was interrupted (e.g. by an error). Then we drop a possible error and run all the steps
to completion.
Let's run it now:
$ cargo run --example coin
First coin: Heads
Second coin: Tails
Error: Coins landed differently
We weren't lucky this time and the program resulted in an error. Let's run it again:
$ cargo run --example coin
Second coin: Heads
Coins match
Notice that, thanks to the restore()
, our machine run from the step it was interrupted,
knowing about the first coin landed on heads.