#type #meta-programming #proofs #logic #proof

macro junglefowl-macros

Macros for type-level mathematical computation

2 releases

0.1.1 May 12, 2023
0.1.0 May 11, 2023

#39 in #meta-programming

24 downloads per month
Used in junglefowl

LGPL-3.0-or-later

8KB
77 lines

Junglefowl

Brutally murdering Rust's type system one proof at a time.

Junglefowl runs Peano arithmetic on Rust types, verified at compile time.

Why?

So we can do theoretically hard stuff, like these const-generic slices:

use junglefowl::*;

// Accept only `u8` arrays with exactly 3 elements:
fn picky<T: Nest<Element = u8, Length = peano!(3)>>(_: &T) {}

// Create an array with 5 elements:
let n12345 = nest![1, 2, 3, 4, 5];

// Split it after its second element without changing anything in memory:
let (left, right) = split!(n12345, 2);

// And we can prove that the second segment will have exactly two elements:
picky(&right);
// picky(&left); // won't compile!

// And know exactly what its elements are:
assert_eq!(nest![3, 4, 5], right);

How?

Here's our Peano encoding:

0 <-->                ()
1 <-->           ((), ())
2 <-->      ((), ((), ()))
3 <--> ((), ((), ((), ())))

Note that, thanks to a clever abuse of Rust's syntax, these are both types and values.

Next, there's a macro so you can forget what you just read:

peano!(0);
 --> ()
peano!(42);
 --> ((), ((), ((), ((), ((), ((), ((), ((), ((), ...)))))))))

Note that this macro expands to a type, so you would use it like this:

let x: peano!(42) = todo!();

instead of like this:

let x = peano!(42); // bad!

And next, there's a hell of a lot of other stuff, but instead of explaining it all, watch this compile:

static_assert_eq!(peano!(39), sub!(peano!(42), peano!(3)));

expands to

enum False {} // uninstantiable type

//      this part vvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvv evaluates to zero when the two sides are equal
const _: [False; (peano!(39) != sub!(peano!(42), peano!(3))) as usize] = [];
// ... which makes the list length zero, which matches the right-hand side (and couldn't be nonzero since its members are uninstantiable)
// learned the list length trick from the `static_assertions` crate, so all credit there!

Expanding the interesting part above (and inverting so != becomes ==):

peano!(39) == <         peano!(42)          as peano::Sub<     peano!(3)     >>::Difference;
peano!(39) == <((), ((), ((), peano!(39)))) as peano::Sub<((), ((), ((), ())))>::Difference;

Here's the definition of peano::Sub, pretty representative for most operations in this crate:

pub trait Sub<R: peano::N>: peano::N { type Difference: peano::N; } // sealed trait
impl<T: peano::N> Sub<()> for T { type Difference = Self; } // subtracting zero is our super-simple base case
impl<L: peano::N + Sub<R>, R: peano::N> Sub<((), R)> for ((), L) { type Difference = sub!(L, R); } // otherwise, reduce the problem until it's dividing by zero

Begin reduction!

peano!(39) == <((), ((), ((), peano!(39)))) as peano::Sub<((), ((), ((), ())))>::Difference;
peano!(39) == <     ((), ((), peano!(39)))  as peano::Sub<     ((), ((), ())) >::Difference;
peano!(39) == <          ((), peano!(39))   as peano::Sub<          ((), ())  >::Difference;
peano!(39) == <               peano!(39)    as peano::Sub<               ()   >::Difference;
peano!(39) ==                 peano!(39)                                                   ;

et voila!

What's with the name?

A certain well-known theorem prover is named after the French word for cock rooster (coq), so I Googled "rooster" and found (to my amazement!) that they belong to the junglefowl species. This name sounded suitably cool.

Dependencies

~230–670KB
~16K SLoC