Language Thoughts

Target

• static/strong type
• structural and nominal(compiletime and runtime)
• reference type by default, has value type
• functional features
  • records
  • pipe
  • immutability by default

Type declaration

Structure and function member should be declared separately.

type Foo {
    foo: string,
    f: () => int
}

impl Foo {
    toString(): string {
        return Foo::type.name
    }
}

Conversion operator

Should happens between nominal types.

let foo = Foo {} as Bar

type Foo {
    static operator as(this: Foo): Bar {
        return this as Bar // recursive calling?
        return {} as Bar
    }
}

type Bar {
    
}

Compile time shape

Declare a compile time type to do strict structure typing. Will be erased after compilation.

• type for declaring a nominal type.
• shape for declaring a structure type.

shape Point {
    x, y: double
}

let p = { x = 1f, y = 2f } shape as Point // `statisfies` in typescript
let p = shape Point {  x = 1f, y = 2f  } // or reversed? which one is more intuitive?
let p: shape Point = {  x = 1f, y = 2f  } // even this? just like golang
let f = (obj: unknown) => {
    obj as Point  // type assertion upon unknown
}

Unit type

Mimicking abstraction upon primitive data types.

Motivation

A custom literal can be useful for data calculating. It should offer benefits:

• static type checking
• unit type composition inference for arithmetic operation
• default format string

Unit type mechanism

Problem: Should it be a operator like custom literals in cpp? Or just a wrapper type for the data?

• Cpp approach lacks of type info since it just simply returns a same type.
• If each unit type is nominal, we will have m*m conversion explosion.
• Another way is extension upon the data type, but it's just a problem of form.

let foo = 360deg
let bar = 6.28rad // for literal
let baz = Math.PI<rad> // for non literal, or (Math.PI)rad ? What if we need a tuple in future? this also conflicts with generic property?
let a = 2 * 3deg // evaluates to (2 * 3)deg, unit type as target type always
let b = 1 - 2deg // error: only scalars can be applied, should do (1 - 2)deg instead
let f = (): <rad> => 1rad // bracketed annotation for differ from normal types

// a nominal unit type can store state
type unit 'deg' for int | double {
    
}

type unit 'rad' for double | int {

}
// shape units are checked at compiletime, no state stored
shape unit 'cm' for Numeric
shape unit 's' for Numeric

Unit type manipulation

Inspired by fsharp - Units of Measure

TimeSpan or Interval can be a good example for unit type, but different unit type like ms, s or min should be evaluated to a same type.

let a = 1min
let b = 60s
let c = 60_000ms
// so it's a operator returns TimeSpan, not a type???
type unit 's' for Numeric as TimeSpan {

}

Questions

• How to do reflection? Is it necessary?
  • fsharp only do compiletime check, unit types are erased at runtime.

Compile-Time evaluation and flows

Type composition

type Foo {
    foo: int,
    bar: string
}
type Bar {
    ...Foo, // [!code highlight] 
    foo: double, // override foo property
    baz: any
}

Or use with for better semantic?

type Foo = {
    foo: int,
    bar: string
}
type Bar = Foo with { foo: double, baz: any } // [!code highlight]

Constant members

Constants as special members, including functions and fields.

// should be evaluated at compile-time
function foo(): const string { // implies the function returns const
    return "I am an constant"
}

type Foo {
    foo: const string = foo() // only constant can be assigned from an constant generator function
    bar: const string = "I am also an constant"
    fn = (): const string => "foo"
}

_ = Foo::foo // use :: to access constant members
_ = Foo::fn() // this evaluates to an constant too

Reflection constants

Some reflection properties should be just constants. Any type should have some default constant members like name, fullName, namespace(if we do plan to design the module system like this) Which means they're preserved constants.

type Bar {
    typeName: any  // preserved member name can not be set
}
// Use `::` as constant accessor.
_ = Foo::typeName // -> `Foo`
_ = Bar::name

Default parameter mutation

type Person {
    name: string
    age: uint
}
function foo(arg: Person = { name = "John", age = 18 }): void {

}
// use `default` to represent the default value of the parameter
foo(default with { name = "jane" })