node package manager

fantasy-land

Specification for interoperability of common algebraic structures in JavaScript

Fantasy Land Specification

(aka "Algebraic JavaScript Specification")

This project specifies interoperability of common algebraic structures:

An algebra is a set of values, a set of operators that it is closed under and some laws it must obey.

Each Fantasy Land algebra is a separate specification. An algebra may have dependencies on other algebras which must be implemented.

  1. "value" is any JavaScript value, including any which have the structures defined below.
  2. "equivalent" is an appropriate definition of equivalence for the given value. The definition should ensure that the two values can be safely swapped out in a program that respects abstractions. For example:
    • Two lists are equivalent if they are equivalent at all indices.
    • Two plain old JavaScript objects, interpreted as dictionaries, are equivalent when they are equivalent for all keys.
    • Two promises are equivalent when they yield equivalent values.
    • Two functions are equivalent if they yield equivalent outputs for equivalent inputs.

In order to add compatibility with Fantasy Land to your library, you need to add methods that you want to support with fantasy-land/ prefix. For example if a type implements Functors' map, you need to add fantasy-land/map method to it. The code may look something like this:

MyType.prototype['fantasy-land/map'] = MyType.prototype.map

It's not required to add unprefixed methods (e.g. MyType.prototype.map) for compatibility with Fantasy Land, but you're free to do so of course.

Further in this document unprefixed names are used just to reduce noise.

For convenience you can use fantasy-land package:

var fl = require('fantasy-land')
 
// ... 
 
MyType.prototype[fl.map] = MyType.prototype.map
 
// ... 
 
var foo = bar[fl.map](x => x + 1)
  1. a.equals(a) === true (reflexivity)
  2. a.equals(b) === b.equals(a) (symmetry)
  3. If a.equals(b) and b.equals(c), then a.equals(c) (transitivity)
equals :: Setoid a => a ~> a -> Boolean

A value which has a Setoid must provide an equals method. The equals method takes one argument:

a.equals(b)
  1. b must be a value of the same Setoid

    1. If b is not the same Setoid, behaviour of equals is unspecified (returning false is recommended).
  2. equals must return a boolean (true or false).

  1. a.concat(b).concat(c) is equivalent to a.concat(b.concat(c)) (associativity)
concat :: Semigroup a => a ~> a -> a

A value which has a Semigroup must provide a concat method. The concat method takes one argument:

s.concat(b)
  1. b must be a value of the same Semigroup

    1. If b is not the same semigroup, behaviour of concat is unspecified.
  2. concat must return a value of the same Semigroup.

A value that implements the Monoid specification must also implement the Semigroup specification.

  1. m.concat(m.empty()) is equivalent to m (right identity)
  2. m.empty().concat(m) is equivalent to m (left identity)
empty :: Monoid m => () -> m

A value which has a Monoid must provide an empty method on itself or its constructor object. The empty method takes no arguments:

m.empty()
m.constructor.empty()
  1. empty must return a value of the same Monoid
  1. u.map(a => a) is equivalent to u (identity)
  2. u.map(x => f(g(x))) is equivalent to u.map(g).map(f) (composition)
map :: Functor f => f a ~> (a -> b) -> f b

A value which has a Functor must provide a map method. The map method takes one argument:

u.map(f)
  1. f must be a function,

    1. If f is not a function, the behaviour of map is unspecified.
    2. f can return any value.
  2. map must return a value of the same Functor

A value that implements the Apply specification must also implement the Functor specification.

  1. v.ap(u.ap(a.map(f => g => x => f(g(x))))) is equivalent to v.ap(u).ap(a) (composition)
ap :: Apply f => f a ~> f (a -> b) -> f b

A value which has an Apply must provide an ap method. The ap method takes one argument:

a.ap(b)
  1. b must be an Apply of a function,

    1. If b does not represent a function, the behaviour of ap is unspecified.
  2. a must be an Apply of any value

  3. ap must apply the function in Apply b to the value in Apply a

A value that implements the Applicative specification must also implement the Apply specification.

  1. v.ap(a.of(x => x)) is equivalent to v (identity)
  2. a.of(x).ap(a.of(f)) is equivalent to a.of(f(x)) (homomorphism)
  3. a.of(y).ap(u) is equivalent to u.ap(a.of(f => f(y))) (interchange)
of :: Applicative f => a -> f a

A value which has an Applicative must provide an of method on itself or its constructor object. The of method takes one argument:

a.of(b)
a.constructor.of(b)
  1. of must provide a value of the same Applicative

    1. No parts of b should be checked
  1. u.reduce is equivalent to u.reduce((acc, x) => acc.concat([x]), []).reduce
reduce :: Foldable f => f a ~> (b -> a -> b) -> b -> b

A value which has a Foldable must provide a reduce method. The reduce method takes two arguments:

u.reduce(f, x)
  1. f must be a binary function

    1. if f is not a function, the behaviour of reduce is unspecified.
    2. The first argument to f must be the same type as x.
    3. f must return a value of the same type as x
  2. x is the initial accumulator value for the reduction

A value that implements the Traversable specification must also implement the Functor and Foldable specifications.

  1. t(u.traverse(x => x, F.of)) is equivalent to u.traverse(t, G.of) for any t such that t(a).map(f) is equivalent to t(a.map(f)) (naturality)

  2. u.traverse(F.of, F.of) is equivalent to F.of(u) for any Applicative F (identity)

  3. u.traverse(x => new Compose(x), Compose.of) is equivalent to new Compose(u.traverse(x => x, F.of).map(x => x.traverse(x => x, G.of))) for Compose defined below and any Applicatives F and G (composition)

var Compose = function(c) {
  this.c = c;
};
 
Compose.of = function(x) {
  return new Compose(F.of(G.of(x)));
};
 
Compose.prototype.ap = function(f) {
  return new Compose(this.c.ap(f.c.map(u => y => y.ap(u))));
};
 
Compose.prototype.map = function(f) {
  return new Compose(this.c.map(y => y.map(f)));
};
traverse :: Apply f, Traversable t => t a ~> ((a -> f b), (c -> f c)) -> f (t b)

A value which has a Traversable must provide a traverse method. The traverse method takes two arguments:

u.traverse(f, of)
  1. f must be a function which returns a value

    1. If f is not a function, the behaviour of traverse is unspecified.
    2. f must return a value of an Applicative
  2. of must be the of method of the Applicative that f returns

  3. traverse must return a value of the same Applicative that f returns

A value that implements the Chain specification must also implement the Apply specification.

  1. m.chain(f).chain(g) is equivalent to m.chain(x => f(x).chain(g)) (associativity)
chain :: Chain m => m a ~> (a -> m b) -> m b

A value which has a Chain must provide a chain method. The chain method takes one argument:

m.chain(f)
  1. f must be a function which returns a value

    1. If f is not a function, the behaviour of chain is unspecified.
    2. f must return a value of the same Chain
  2. chain must return a value of the same Chain

A value that implements the ChainRec specification must also implement the Chain specification.

  1. m.chainRec((next, done, v) => p(v) ? d(v).map(done) : n(v).map(next), i) is equivalent to (function step(v) { return p(v) ? d(v) : n(v).chain(step); }(i)) (equivalence)
  2. Stack usage of m.chainRec(f, i) must be at most a constant multiple of the stack usage of f itself.
chainRec :: ChainRec m => ((a -> c, b -> c, a) -> m c, a) -> m b

A Type which has a ChainRec must provide a chainRec method on itself or its constructor object. The chainRec method takes two arguments:

a.chainRec(f, i)
a.constructor.chainRec(f, i)
  1. f must be a function which returns a value
    1. If f is not a function, the behaviour of chainRec is unspecified.
    2. f takes three arguments next, done, value
      1. next is a function which takes one argument of same type as i and can return any value
      2. done is a function which takes one argument and returns the same type as the return value of next
      3. value is some value of the same type as i
    3. f must return a value of the same ChainRec which contains a value returned from either done or next
  2. chainRec must return a value of the same ChainRec which contains a value of same type as argument of done

A value that implements the Monad specification must also implement the Applicative and Chain specifications.

  1. m.of(a).chain(f) is equivalent to f(a) (left identity)
  2. m.chain(m.of) is equivalent to m (right identity)
  1. w.extend(g).extend(f) is equivalent to w.extend(_w => f(_w.extend(g)))
extend :: Extend w => w a ~> (w a -> b) -> w b

An Extend must provide an extend method. The extend method takes one argument:

 w.extend(f)
  1. f must be a function which returns a value

    1. If f is not a function, the behaviour of extend is unspecified.
    2. f must return a value of type v, for some variable v contained in w.
  2. extend must return a value of the same Extend.

A value that implements the Comonad specification must also implement the Functor and Extend specifications.

  1. w.extend(_w => _w.extract()) is equivalent to w
  2. w.extend(f).extract() is equivalent to f(w)
  3. w.extend(f) is equivalent to w.extend(x => x).map(f)
extract :: Comonad w => w a ~> () -> a

A value which has a Comonad must provide an extract method on itself. The extract method takes no arguments:

c.extract()
  1. extract must return a value of type v, for some variable v contained in w.
    1. v must have the same type that f returns in extend.

A value that implements the Bifunctor specification must also implement the Functor specification.

  1. p.bimap(a => a, b => b) is equivalent to p (identity)
  2. p.bimap(a => f(g(a)), b => h(i(b)) is equivalent to p.bimap(g, i).bimap(f, h) (composition)
bimap :: Bifunctor f => f a c ~> (a -> b) -> (c -> d) -> f b d

A value which has a Bifunctor must provide a bimap method. The bimap method takes two arguments:

c.bimap(f, g)
  1. f must be a function which returns a value

    1. If f is not a function, the behaviour of bimap is unspecified.
    2. f can return any value.
  2. g must be a function which returns a value

    1. If g is not a function, the behaviour of bimap is unspecified.
    2. g can return any value.
  3. bimap must return a value of the same Bifunctor.

A value that implements the Profunctor specification must also implement the Functor specification.

  1. p.promap(a => a, b => b) is equivalent to p (identity)
  2. p.promap(a => f(g(a)), b => h(i(b))) is equivalent to p.promap(f, i).promap(g, h) (composition)
promap :: Profunctor p => p b c ~> (a -> b) -> (c -> d) -> p a d

A value which has a Profunctor must provide a promap method.

The profunctor method takes two arguments:

c.promap(f, g)
  1. f must be a function which returns a value

    1. If f is not a function, the behaviour of promap is unspecified.
    2. f can return any value.
  2. g must be a function which returns a value

    1. If g is not a function, the behaviour of promap is unspecified.
    2. g can return any value.
  3. promap must return a value of the same Profunctor

When creating data types which satisfy multiple algebras, authors may choose to implement certain methods then derive the remaining methods. Derivations:

  • map may be derived from ap and of:

    function(f) { return this.ap(this.of(f)); }
  • map may be derived from chain and of:

    function(f) { return this.chain(a => this.of(f(a))); }
  • map may be derived from bimap:

    function(f) { return this.bimap(a => a, f); }
  • map may be derived from promap:

    function(f) { return this.promap(a => a, f); }
  • ap may be derived from chain:

    function(m) { return m.chain(f => this.map(f)); }
  • reduce may be derived as follows:

    function(f, acc) {
      function Const(value) {
        this.value = value;
      }
      Const.of = function(_) {
        return new Const(acc);
      };
      Const.prototype.map = function(_) {
        return this;
      };
      Const.prototype.ap = function(b) {
        return new Const(f(b.value, this.value));
      };
      return this.traverse(x => new Const(x), Const.of).value;
    }
  • map may be derived as follows:

    function(f) {
      function Id(value) {
        this.value = value;
      };
      Id.of = function(x) {
        return new Id(x);
      };
      Id.prototype.map = function(f) {
        return new Id(f(this.value));
      };
      Id.prototype.ap = function(b) {
        return new Id(this.value(b.value));
      };
      return this.traverse(x => Id.of(f(x)), Id.of).value;
    }

If a data type provides a method which could be derived, its behaviour must be equivalent to that of the derivation (or derivations).

  1. If there's more than a single way to implement the methods and laws, the implementation should choose one and provide wrappers for other uses.
  2. It's discouraged to overload the specified methods. It can easily result in broken and buggy behaviour.
  3. It is recommended to throw an exception on unspecified behaviour.
  4. An Id container which implements all methods is provided in id.js.

There also exists Static Land Specification with the exactly same ideas as Fantasy Land but based on static methods instead of instance methods.