Rust: controlling side effects from the test.

Hello and welcome to the newest episode on testing in Rust. Imagine you want to write a timestamping repository of some sorts, that will associate the timestamp of when the storage operation was invoked with the stored value. How to write it in Rust ? And more importantly - how to test it ? I would like to share a solution I found and talk a bit about how it works.

Please note that this solution can be used anywhere where you need to pass a handle that is remembered by the production code, and that thing it points to - you then want to change from the test.

trait Clock {
    fn now(&self) -> Instant;
}

struct SystemClock;

impl SystemClock {
    fn new() -> Self {
        SystemClock {}
    }
}

impl Clock for SystemClock {
    fn now(&self) -> Instant {
        Instant::now()
    }
}

struct TimestampingRepository<'a, ClockType>
where
    ClockType: Clock + 'a,
{
    clock: &'a ClockType,
    storage: Vec<(Instant, u32)>, // (timestamp, value)
}

impl<'a, ClockType> TimestampingRepository<'a, ClockType>
where
    ClockType: Clock + 'a,
{
    fn with_clock(clock: &'a ClockType) -> Self {
        TimestampingRepository {
            clock,
            storage: vec![],
        }
    }

    fn store(&mut self, value: u32) {
        self.storage.push((self.clock.now(), value));
    }

    fn all_stored(&self) -> Vec<(Instant, u32)> {
        self.storage.clone()
    }
}

#[cfg(test)]
mod should {

    #[test]
    fn handle_seconds() {
        let clock = FakeClock::with_time(Instant::now());
        let mut repository = TimestampingRepository::with_clock(&clock);

        repository.store(1);
        clock.move_by(Duration::from_secs(32));
        repository.store(2);

        let time_difference = time_difference_between_two_stored(repository);

        assert_eq!(32, time_difference.as_secs());
    }

    struct FakeClock {
        now: Instant,
        move_by_secs: AtomicUsize,
    }

    impl FakeClock {
        fn with_time(now: Instant) -> Self {
            FakeClock {
                now,
                move_by_secs: AtomicUsize::new(0),
            }
        }

        // WAT no `mut`
        fn move_by(&self, duration: Duration) {
            self.move_by_secs
                .store(duration.as_secs() as usize, Ordering::SeqCst);
        }
    }

    impl Clock for FakeClock {
        fn now(&self) -> Instant {
            let move_by_secs = self.move_by_secs.load(Ordering::SeqCst) as u64;
            self.now + Duration::from_secs(move_by_secs)
        }
    }

}

That's a lot of code. And I already skipped uses and some definitions to make it less. If you want to get the full source code that to follow along - try this playground or this repo for the full project including production code usage.

Let's start with the test itself. The clock appears to be immutable (immovable) in the test, yet we call move_by on it and the whole thing appears to be working somehow. First question: can't we just make the clock mutable and skip all this ? It appears that sadly (but fortunately) Rust prevents us from doing so. We cannot both have a immutable and mutable borrow of the clock in the same scope. For the full example with an error go here.

What is this sorcery then ? We use a type that provides Interior Mutability, namely AtomicUsize. On the outside - it look immutable, yet it provides a thread-safe and very narrow method of mutating the underlying state. As we trust AtomicUsize to be written correctly, we can then proceed and write our Rust code as usual, relying fully on the borrow checker. Rust compiler is happy and our test code is happy.

I wouldn't use this as a pattern in production code - the borrow checker rules are there for a reason. Please treat it as an escape hatch to be used in specific situations, situations like this.

Happy Rusting !

p.s. if you'd like to chat about Rust - ping me an email !

4.6 KiB Raw Blame History

Rust: controlling side effects from the test.

4.6 KiB

Raw Blame History