Algebraic data types and immutable structures for Javascript.

adt.js gives you the following for free:

• Immutablity
• Type-checking attributes
• Deep equality and cloning
• Curried constructors
• toString and toJSON implementations
• Enums
• Sweet.js macros

npm install adt

Let's start by creating a simple Maybe ADT for possible failure:

JS:

var adt = require('adt');
var Maybe = adt.data({
  Nothing: null,
  Just: { value: adt.any }
});

CS:

{data, any, only} = require 'adt'
Maybe = data
  Nothing : null
  Just :
    value : any

Macros:

data Maybe {
  Nothing,
  Just {
    value: *
  }
}

adt.any is a value constraint that will allow anything. If you wanted to restrict the type, you could use adt.only.

Here's how you might use our new data type:

var noth = Maybe.Nothing;
var just = Maybe.Just(42);

// Inheritance
(just instanceof Maybe.Just) === true;
(just instanceof Maybe) === true;

// Type-checking
just.isNothing === false;
just.isJust === true;

// Record attributes
just.value === 42;

// Immutablity: `set` returns a new instance
var just2 = just.set({ value: 43 });
just !== just2;

// Retrieve values by name or by index
just.get('value') === 42;
just.get(0) === 42;
just.get(1); // Error: Out of range

// `toString` implementation
just.toString() === 'Just(42)';

Since Nothing is not a record (it doesn't have any data attributes), it exists as a singleton instance and does not need to be instanciated.

Let's define a linked-list type:

JS:

var adt = require('adt');
var List = adt.data(function () {
  return {
    Nil: null,
    Cons: {
      head: adt.any,
      tail: adt.only(this)
    }
  };
});

CS:

{data, any, only} = require 'adt'
List = data ->
  Nil : null
  Cons :
    head : any
    tail : only this

Macros:

data List {
  Nil,
  Cons {
    head: *,
    tail: List
  }
}

Note that we've introduced a lambda to house our definition. With our Maybe type this wasn't necessary, because we didn't need to reference the ADT itself. But here, we want to use adt.only to put a constraint on the value of tail so it can only contain List types. If we left out the lambda and just used the object literal syntax, List wouldn't exist when we try to pass it to adt.only and we'd get a ReferenceError. See the end of this document for an alternative that does not require a lambda.

And now let's put it to good use:

var list = List.Cons(12, List.Cons(42, List.Nil));

// Record attributes
list.head === 12;
list.tail.toString() === 'Cons(42, Nil)';

// Deep equality
var list2 = List.Cons(42, List.Nil);
list.tail.equals(list2) === true;

// Instanciate with key/value pairs
List.Cons.create({
  head: 42,
  tail: List.Nil
});

// Curried constructor
var consPartial = List.Cons(12);
var list3 = consPartial(List.Nil);

// Constraints
List.Cons(42, 12) // TypeError!

Let's define a simple days-of-the-week enum using adt.enumeration or its alias adt.enum:

JS:

var Days = adt.enum('Sun', 'Mon', 'Tues', 'Wed', 'Thur', 'Fri', 'Sat');

Macros:

enum Days {
  Sun, Mon, Tues, Wed, Thur, Fri, Sat
}

Enums can be compared using lt, lte, eq, gte, and gt.

var day1 = Days.Tues;
var day2 = Days.Fri;

day1.lt(day2) === true;
day2.gt(day1) === true;
day1.eq(Days.Mon) === false;

Enums can also have constant values for JSON serialization:

JS:

var Days2 = adt.enum({
  Sun  : 1,
  Mon  : 2,
  Tues : 3,
  Wed  : 4,
  Thur : 5,
  Fri  : 6,
  Sat  : 7
});

// Our previous definition serializes everything to null.
Days.Mon.toJSON() === null;

// But our new one serializes to an integer.
Days2.Mon.toJSON() === 2;

Macros:

enum Days2 {
  Sun = 1,
  Mon = 2,
  Tues = 3,
  // ...etc
}

Note that the value you give it does not affect the comparison methods. That is determined solely by insertion order.

Enums aren't really special. They are just normal ADTs with some extra behavior. You are not restricted to only using singleton types like we did above. You could just as easily have an enum of record types too. Likewise, you can also give a value to any singleton type. null is just the default value and often times a good representation of the type (Nothing, Nil, Empty, etc).

Sometimes you just need a type that exists by itself. Use adt.newtype as a shortcut:

JS:

// Instead of this:
var Lonely = adt.data({
  Lonely: {
    value: adt.any
  }
});
Lonely = Lonely.Lonely;

// Do this:
var Lonely = adt.newtype('Lonely', {
  value: adt.any
});

Macros:

newtype Lonely {
  value: *
}

adt.js has two builtin value constraints: any, to represent the lack of a constraint, and only, to restrict a value to certain types.

// `any` is an id function
adt.any(12) === 12;
adt.any('Foo') === 'Foo';

// Only is a constraint factory
var onlyNumbers = adt.only(Number);
var onlyStrings = adt.only(String);
var onlyPrimitives = adt.only(Number, String, Boolean);

onlyNumbers(12) === 12;
onlyStrings('Foo') === 'Foo';
onlyPrimitives(/^$/); // TypeError!

Constraints are just functions that take a value and return another or throw an exception.

function toString (x) { 
  return x.toString();
};

var OnlyStrings = adt.newtype({
  value: toString
});

OnlyStrings(12).value === '12';

All ADTs are left "open" by default, meaning you can add types and fields to it at a later time. You can close your ADT by calling seal.

var Maybe = adt.data();
var Nothing = Maybe.type('Nothing');
var Just = Maybe.type('Just', { value: adt.any });

// Close it.
Maybe.seal();

// Calling `type` results in an error
Maybe.type('Foo'); // Error!

Astute readers might notice that adt.js relies on a controversial feature: the host engine maintaining insertion order of keys in object literals. It's true that the Javascript spec does not require this feature. However, it has become a defacto standard, and all engines implement this feature for the string keys we are using.

adt.js also offers a "safe" API that does not rely on this feature:

var List = adt.data(function (type, List) {
  type('Nil', null);
  type('Cons', adt.record(function (field) {
    field('head', adt.any);
    field('tail', adt.only(List));
  }));
});

In fact, this is just the desugared form of the terse API. See the end of this document for an alternative that uses chaining instead of lambdas and closures.

Javascript is inherently mutable, and so adt.js can't guarantee immutablity, only facilitate it. By using set instead of direct attribute assignment, we get safe, immutable structures. But if we were to store say an object literal as a value, we could certainly get a reference to it and mutate it, affecting any data that might be sharing it.

var obj = { foo: 'bar' };
var just1 = Just(obj);
var just2 = Just(obj);

// Bad!
just1.value.foo = 'baz';
just2.value.foo === 'baz';

adt.js only performs deep equality on adt.js types. It does not perform deep equality on native arrays or objects. Anything that is not an adt.js type is compared using strict equality (===).

var arr = [1, 2, 3];
var just1 = Just(arr);
var just2 = Just(arr);

just1.equals(just2) === true;
just1.equals(Just([1, 2, 3])) === false;

If you would like to extend this behavior, you can override the default method for equality on native JS types. For example, if you were using lodash:

// Deep equality on all native JS types (Objects, Arrays, RegExps, Dates, etc.)
adt.nativeEquals = _.isEqual;

adt.js types all have a clone method for returning a safe copy of a data structure. As with deep equality, it only clones adt.js types and copies arrays and objects by reference. Singleton instances will always return the same instance when copied.

var just1 = Just(42);
var just2 = just.clone();

just2.value === 42;
just1 !== just2;

As with equality, you can extend the default cloning behavior for native JS types. Using lodash:

adt.nativeClone = _.cloneDeep;

For some types, it can be nice to have some sugar on the parent type. For example, it would be nice if you could build a List like you would an Array:

var arr = Array(1, 2, 3);

// Wouldn't this be nice?
var list = List(1, 2, 3);
list.toString() === 'Cons(1, Cons(2, Cons(3, Nil)))';

adt.js detects when you override your apply method and can use that to create your types.

List.apply = function (ctx, args) {
  // Hypothetical `fromArray` function
  return List.fromArray(args);
};

Data types made with adt.js have builtin support for sparkler, a pattern matching engine for JavaScript:

data Tree {
  Empty,
  Node {
    value: *,
    left: Tree,
    right: Tree
  }
}

function treeFn {
  case Empty => 'empty'
  case Node(42, ...) => '42'
  case Node{ left: Node(12, ...) } => 'left 12'
}

Find out more about sparkler: https://github.com/natefaubion/sparkler

adt.js has a versatile API, so you can define your types in a way that suits you. Some ways are very terse, while others are "safer" (don't rely on object insert order).

If you don't like defining recursive types within a function, you might like:

var List = adt.data();
var Nil  = List.type('Nil');
var Cons = List.type('Cons', {
  head: adt.any,
  tail: adt.only(List)
});

This has the advantage of shaving off a few lines but requires some name duplication.

Another way of defining "safe" types is to use chaining instead of a closure:

var List = adt.data();
var Nil  = List.type('Nil');
var Cons = List.type('Cons', {})
             .field('head', adt.any)
             .field('tail', adt.only(List));

Depending on you needs, there should hopefully be an easy, terse way of defining your types.

npm install -g sweet.js
npm install adt
sjs -m adt/macros myfile.js

In your file you don't need to require('adt'). The macro will load it for you when you define a data type.

One nice property of the macros is that the data constructors are automatically brought into the surrounding scope:

data List {
  Nil,
  Cons {
    head: *,
    tail: List
  }
}

// Nil and Cons are in scope.
var list = Cons(42, Cons(12, Nil));

When declaring your constraints, the macros try to "do the right thing". If the identifier for the constraint starts with an upper-case letter, it will use an adt.only constraint. If it starts with a lower-case letter, it will use it as is. You can also inline a function literal as a constraint.

Nathan Faubion (@natefaubion)

MIT