Awesome

Lys, a language that compiles to WebAssembly.

Where to start?

To learn what can be used so far: browse the standard library
To learn how real code looks like: browse the execution tests
To learn how high level constructs get compiled: browse the sugar syntax tests
To start developing it locally, I do make watch and then I run the tests in other terminal with make snapshot
To see an example project: browse the keccak repo

Getting started

For the time being I'll use npm to distribute the language.

npm i -g lys

Create a folder and a file main.lys

import support::env

#[export]
fun test(): void = {
  support::test::START("This is a test suite")

  printf("Hello %X", 0xDEADBEEF)
  support::test::mustEqual(3 as u8, 3 as u16, "assertion name")

  support::test::END()
}

Run lys main.lys --test --wast. It will create main.wasm main.wast and will run the exported function named test.

How does it look?

Structs & Implementing operators

struct Vector3(x: f32, y: f32, z: f32)

impl Vector3 {
  fun -(lhs: Vector3, rhs: Vector3): Vector3 =
    Vector3(
      lhs.x - rhs.x,
      lhs.y - rhs.y,
      lhs.z - rhs.z
    )

  #[getter]
  fun length(this: Vector3): f32 =
    f32.sqrt(
      this.x * this.x +
      this.y * this.y +
      this.z * this.z
    )
}

fun distance(from: Vector3, to: Vector3): f32 = {
  (from - to).length
}

Pattern matching

// this snippet is an actual unit test

import support::test

enum Color {
  Red
  Green
  Blue
  Custom(r: i32, g: i32, b: i32)
}

fun isRed(color: Color): boolean = {
  match color {
    case is Red -> true
    case is Custom(r, g, b) -> r == 255 && g == 0 && b == 0
    else -> false
  }
}

#[export]
fun main(): void = {
  mustEqual(isRed(Red), true, "isRed(Red)")
  mustEqual(isRed(Green), false, "isRed(Green)")
  mustEqual(isRed(Blue), false, "isRed(Blue)")

  mustEqual(isRed(Custom(255,0,0)), true, "isRed(Custom(255,0,0))")
  mustEqual(isRed(Custom(0,1,3)), false, "isRed(Custom(0,1,3))")
  mustEqual(isRed(Custom(255,1,3)), false, "isRed(Custom(255,1,3))")
}

Algebraic data types

// this snippet is an actual unit test

enum Tree {
  Node(value: i32, left: Tree, right: Tree)
  Empty
}

fun sum(arg: Tree): i32 = {
  match arg {
    case is Empty -> 0
    case is Node(value, left, right) -> value + sum(left) + sum(right)
  }
}

#[export]
fun main(): void = {
  val tree = Node(42, Node(3, Empty, Empty), Empty)

  support::test::mustEqual(sum(tree), 45, "sum(tree) returns 45")
}

Types and overloads are created in the language itself

The compiler only knows how to emit functions and how to link function names. I did that so I had fewer things hardcoded into the compiler and allows me to write the language in the language.

To do that, I had to add either a %wasm { ... } code block, and a %stack { ... } type.

%wasm { ... }: can only be used as a function body, not as an expression. It is literally the code that will be emited to WAST. The parameter names remain the same (prefixed with $ as WAST indicates). Other symbols can be resolved with fully::qualified::names.
%stack { wasm="i32", size=4 }: it is a type literal, it indicates how much memory should be allocated in structs (size) and what type to use in locals and function parameters (wasm, it needs a better name).

/** We first define the type `int` */
type int = %stack { wasm="i32", size=4 }

/** Implement some operators for the type `int` */
impl int {
  fun +(lhs: int, rhs: int): int = %wasm {
    (i32.add (local.get $lhs) (local.get $rhs))
  }
  fun -(lhs: int, rhs: int): int = %wasm {
    (i32.sub (local.get $lhs) (local.get $rhs))
  }
  fun >(lhs: int, rhs: int): boolean = %wasm {
    (i32.gt_s (local.get $lhs) (local.get $rhs))
  }
}

fun fibo(n: int, x1: int, x2: int): int = {
  if (n > 0) {
    fibo(n - 1, x2, x1 + x2)
  } else {
    x1
  }
}

#[export "fibonacci"] // "fibonacci" is the name of the exported function
fun fib(n: int): int = fibo(n, 0, 1)

Some sugar

Enum types

enum Tree {
  Node(value: i32, left: Tree, right: Tree)
  Empty
}

Is the sugar syntax for

type Tree = Node | Empty

struct Node(value: i32, left: Tree, right: Tree)
struct Empty()

impl Tree {
  fun is(lhs: Tree): boolean = lhs is Node || lhs is Empty
  // ...
}

impl Node {
  fun as(lhs: Node): Tree = %wasm { (local.get $lhs) }

  // ... many methods were removed for clarity ..
}

impl Empty {
  fun as(lhs: Node): Tree = %wasm { (local.get $lhs) }
  // ...
}

`is` and `as` operators are just functions

impl u8 {
  /**
   * Given an expression with the shape:
   *
   *   something as Type
   *   ^^^^^^^^^    ^^^^
   *        $lhs    $rhs
   *
   * A function with the signature:
   *     fun as($lhs: LHSType): $rhs = ???
   *
   * Will be searched in the impl of LHSType
   *
   */


  fun as(lhs: u8): f32 = %wasm { (f32.convert_i32_u (local.get $lhs)) }
}

fun byteAsFloat(value: u8): f32 = value as f32

struct CustomColor(rgb: i32)

type Red = void
impl Red {
  fun is(lhs: CustomColor): boolean = match lhs {
    case is Custom(rgb) -> (rgb & 0xFF0000) == 0xFF0000
    else -> false
  }
}

var x = CustomColor(0xFF0000) is Red

// this may not be a good thing, but you get the idea

There are no dragons behind the structs

The struct keyword is only a high level construct that creats a type and base implementation of something that behaves like a data type, normally in the heap.

struct Node(value: i32, left: Tree, right: Tree)