#logic #meta-programming #type #proofs

no-std rsmonad

Haskell-style monads for idiomatic Rust

7 releases

0.2.4 May 25, 2023
0.2.2 May 23, 2023
0.1.2 May 19, 2023

#377 in Math

Download history 24/week @ 2024-07-29 2/week @ 2024-09-23 5/week @ 2024-09-30

99 downloads per month

MPL-2.0 and LGPL-3.0-or-later

74KB
1.5K SLoC

Monads, Functors, & More in Stable Rust

Haskell-style monads with macro-free Rust syntax.

Types work with zero annotations

A fancy example (that compiles and runs as a test in gallery.rs):

assert_eq!(
    vec![
        || "this string isn't a number",
        || "67",
        || "that was, but not binary",
        || "1010101010",
        || "that was, but more than 8 bits",
        || "101010",
        || "that one should work!",
        || panic!("lazy evaluation!"),
    ]
    .fmap(|x| move || u8::from_str_radix(x(), 2).ok())
    .asum(),
    Some(42)
);

Desugared do-notation:

assert_eq!(
    list![1..20]
        >> |x| {
            list![{ x }..20]
                >> |y| {
                    list![{ y }..20]
                        >> |z| guard::<List<_>>(x * x + y * y == z * z) >> |_| list![(x, y, z)]
                }
        },
    list![
        // Including "non-primitive" (i.e. multiples of smaller) triples
        (3, 4, 5),
        (5, 12, 13),
        (6, 8, 10),
        (8, 15, 17),
        (9, 12, 15)
    ]
);

The logic of Haskell lists with the speed of Rust vectors:

use rsmonad::prelude::*;
let li = list![1, 2, 3, 4, 5];
fn and_ten(x: u8) -> List<u8> { list![x, 10 * x] }
assert_eq!(li >> and_ten, list![1, 10, 2, 20, 3, 30, 4, 40, 5, 50]);

N-fold bind without type annotations:

// from the wonderful Haskell docs: https://en.wikibooks.org/wiki/Haskell/Understanding_monads/List
fn bunny(s: &str) -> List<&str> {
    list![s, s, s]
}
assert_eq!(
    list!["bunny"] >> bunny,
    list!["bunny", "bunny", "bunny"],
);
assert_eq!(
    list!["bunny"] >> bunny >> bunny,
    list!["bunny", "bunny", "bunny", "bunny", "bunny", "bunny", "bunny", "bunny", "bunny"],
);

And even the notoriously tricky join-in-terms-of-bind with no type annotations necessary:

let li = list![list![0_u8]]; // List<List<u8>>
let joined = li.join();      // -->  List<u8>!
assert_eq!(joined, list![0]);

Syntactic sugar

  • Rust requires >>= to be self-modifying, so we use >> instead. return (keyword) becomes consume and Haskell's >> becomes &.
  • Function-application order is standardized across the board to match >>= instead of <$>: it's x.fmap(f) to mean f <$> x.
    • I think this makes it easier to read as a chain of computations, but I'm not dead-set on it, and the type system would work either way.
  • For functors, you can use fmap(f, x) or x.fmap(f), or you can pipe it: x % f % g % ....
  • Applicative uses * instead of <*> (or the explicit method tie).
  • Alternative uses | instead of <|> (or the explicit method either).
  • Monoids use + instead of <>.

Use

Just write a monad! { ... and you get all its superclasses (Functor, Applicative, Alternative, ...) for free, plus property-based tests of the monad laws:

use rsmonad::prelude::*;

enum Maybe<A> {
    Just(A),
    Nothing,
}

monad! {
    Maybe<A>:

    fn consume(a) {
        Just(a)
    }

    fn bind(self, f) {
        match self {
            Just(a) => f(a),
            Nothing => Nothing,
        }
    }
}

// And these just work:

// Monad
assert_eq(Just(4) >> |x| u8::checked_add(x, 1).into(), Just(5));
assert_eq(Nothing >> |x| u8::checked_add(x, 1).into(), Nothing);
assert_eq(Just(255) >> |x| u8::checked_add(x, 1).into(), Nothing);

// Functor
assert_eq!(Just(4) | u8::is_power_of_two, Just(true));
assert_eq!(Nothing | u8::is_power_of_two, Nothing);

Rust-specific monads

Catch panics without worrying about the details:

fn afraid_of_circles(x: u8) -> BlastDoor<()> {
    if x == 0 { panic!("aaaaaa!"); }
    Phew(())
}
assert_eq!(
    Phew(42) >> afraid_of_circles,
    Phew(())
);
assert_eq!(
    Phew(0) >> afraid_of_circles,
    Kaboom,
);

Sharp edges

Right now, you can use >> for bind only when you have a concrete instance of Monad like Maybe but not a general <M: Monad<A>>. The latter still works but requires an explicit call to m.bind(f) (or, if you don't use the trait, Monad::<A>::bind(m, f)). This should be fixed with the Rust's non-lifetime binder feature when it rolls out.

"Cannot find type ... in this scope" in a doctest

Doctests try to guess where to place a fn main { ... } if you don't provide one, and sometimes it reads an rsmonad macro as something that should be in a main block. Try adding an explicit fn main () { ... } around the statements you want to run. If you don't want fn main() { ... } to show up in documentation but can't fix this error, comment it out:

/// monad! { ... }
/// # fn main {
/// a();
/// b();
/// c();
/// # }

#![no_std]

Disable default features:

# Cargo.toml

[dependencies]
rsmonad = { version = "*", default-features = false }

Note that this will also disable List,though this is probably what you want: we can't know its length at compile time (that's the point of its bind implementation), so it requires a heap. An alloc feature is in the works for #![no_std] extern crate alloc; crates, but it's not finalized yet.

Dependencies

~1.4–2.3MB
~40K SLoC