Fantas, Eel, and Specification 8: Apply

Aaand we’re back - hello, everyone! Today, we’re going to take another look at those mystical Functor types. We said a couple weeks ago that functors encapsulate a little world (context) with some sort of language extension. Well, what happens when worlds collide? Let’s talk about Apply.

All Apply types are Functor types by requirement, so we know they’re definitely “containers” for other types. The exciting new feature here is this ap function:

ap :: Apply f => f a ~> f (a -> b) -> f b
--                 a ->   (a -> b) ->   b

If we ignore the fs, we get the second line, which is our basic function application: we apply a value of type a to a function of type a -> b, and we get a value of type b. Woo! What’s the difference with ap? All those bits are wrapped in the context of our f functor!

That’s the, uh, “grand reveal”. Ta-da. I’m not really sure that, in isolation, this particularly helps our intuition, though, so let’s instead look at ap within the lift2 function:

// Remember: `f` MUST be curried!
// lift2 :: Applicative f
//       =>  (a ->   b ->   c)
//       -> f a -> f b -> f c
const lift2 = f => a => b =>

For me, this is much clearer. lift2 lets us combine two separate wrapped values into one with a given function.

lift1, if you think about it, is just The lift2 pattern actually works for any number of arguments; once you finish the article, why not try to write lift3? Or lift4?

Wait, combine? Do I sense a Semigroup?

Sort of! You can think of it this way: a Semigroup type allows us to merge values. An Apply type allows us to merge contexts. Neat, huh? Now, how could we forget the laws?!

// compose :: (b -> c) -> (a -> b) -> a -> c
const compose = f => g => x => f(g(x))

x.ap(g.ap( === x.ap(g).ap(f)

// But, if we write lift3...
const lift3 =
  f => a => b => c =>

// Now our law looks like this!
lift3(compose)(f)(g)(x) === x.ap(g).ap(f)

// Remember:
//       ===

By introducing some little helper functions, our law seems much clearer, and a little more familiar. It says that, just as map could only apply a function to the wrapped value, ap can only apply a wrapped function to the wrapped value. No magic tricks!

Before we go any further, I challenge you take a moment to try to build lift2 without ap. Just think about why we couldn’t do this with a plain old Functor. If we tried to write lift2, we might end up here:

// lift2F :: Functor f
//        => (  a ->   b ->      c)
//        ->  f a -> f b -> f (f c)
const lift2F = f => as => bs => => => f(a)(b)))

So, we can apply the inner values to our function - hooray! - but look at the type here. We’re doing a map inside a map, so we’ve ended up with two levels of our functor type! It’s clear that we can’t write a generic lift2 to work with any Functor, and ap is what’s missing.

With all that out the way, let’s look at some examples, shall we? We’ll start with the Identity type’s ap from our beloved spec:

const Identity = daggy.tagged('x')

// map :: Identity a ~> (a -> b)
//                   -> Identity b = function (f) {
  return new Identity(f(this.x))

// ap :: Identity a ~> Identity (a -> b)
//                  -> Identity b
Identity.prototype.ap = function (b) {
  return new Identity(b.x(this.x))

// Identity(5)
lift2(x => y => x + y)

No frills, no magic. Identity.ap takes the function from b, the value from this, and returns the wrapped-up result. Did you spot the similarity in type between map and ap, by the way? Moving on, here’s the slightly more complex implementation for Array:

// Our implementation of ap.
// ap :: Array a ~> Array (a -> b) -> Array b
Array.prototype.ap = function (fs) {
  return [].concat(...
    f =>

// 3 x 0 elements
// []
[2, 3, 4].ap([])

// 3 x 1 elements
// [ '2!', '3!', '4!' ]
[2, 3, 4]
.ap([x => x + '!'])

// 3 x 2 elements
// [ '2!', '2?'
// , '3!', '3?'
// , '4!', '4?' ]
[2, 3, 4]
.ap([ x => x + '!'
    , x => x + '?' ])

I’ve put a little note with the answers so we can see what’s happening: we get every a and b pair. This is called the cartesian product of the two arrays. On top of that, when we lift2 an f over two Array types, we’re actually doing something quite familiar

return lift2(x => y => x + y)(array1)(array2)

// ... is the same as...

const result = []

for (x in array1)
  for (y in array2)
    result.push(x + y)

return result

We get a really pretty shorthand for multi-dimensional loops. Flattening a loop within a loop gives us every possible pair of elements, and that’s what ap is for! If this feels weird, just think of the types. We have to use Array (a -> b) and Array a to get to Array b without violating the composition law; there aren’t many possibilities!

There are loads of types with ap instances. Most, we’ll see, implement ap in terms of chain; we’ll look at the Chain spec in a week or two, so don’t worry too much. Most of them are fairly intuitive anyway:

  • Maybe combines possible failures. If either of the two Maybe values are Nothing, the result is Nothing.
Just(2).ap(Just(x => -x)) // Just(-2)
Nothing.ap(Just(x => -x)) // Nothing
Just(2).ap(Nothing)       // Nothing
Nothing.ap(Nothing)       // Nothing
  • Either combines possible failures with exceptions. If either of the two are Left, the result is the first Left.
Right(2)    .ap(Right(x => -x)) // Right(-2)
Left('halp').ap(Right(x => -x)) // Left('halp')
Right(2)    .ap(Left('eek'))    // Left('eek')
Left('halp').ap(Left('eek'))    // Left('halp')

At some point, I’d like to write a follow up to the Functor post to give some more practical examples, but, for now, this is hopefully understandable (please tweet me if I’m wrong!). Whatever your Functor trickery, ap is map with a wrapped function. Before we go, though, I’d like to talk about one last trick up Apply’s sleeve…

A type we haven’t talked about before is Task. This is similar to Either - it represents either an error or a value - but the difference is that Task’s value is the result of a possibly-asynchronous computation. They look a lot like Promise types:

const Task = require('data.task')

// Convert a fetch promise to a Task.
const getJSON = url => new Task((rej, res) =>

We can see that it holds a function that will eventually call a resolver. Task, just like Promise, sorts out all the async wiring for us. However, an interesting feature of Task is that it implements Apply. Let’s take a look:

const renderPage = users => posts =>
  /* Write some HTML with this data... */

// A Promise of a web page.
// page :: Task e HTML
const page =

Just as we’d expect: we get the two results, and combine them into one using renderPage as the “glue”. However, we can see that lift2’s second and third arguments have no dependencies on one another. Because of this, the arguments to lift2 can always be calculated in parallel. Do you hear that? These AJAX requests are automatically parallelised! Ooer!

You can see Task.ap’s implementation for an exact explanation, but isn’t this great? We can abstract parallelism and never have to worry about it! When we have two parallel Tasks and finally want to glue them back together, we just use lift2! Parallelism becomes an implementation detail. Out of sight, out of mind!

I think Task gives a really strong case for Apply and why it’s immediately useful. When we look at Traversable in a few weeks, we’ll come back to ap and see just how powerful it is. Until then, don’t overthink ap - it’s just a mechanism for combining contexts (worlds!) together without unwrapping them.

I had originally intended to mention of in this post and cover the full Applicative. However, it’s already quite a long post, so I’ll write up that post some time this week! I might even throw in some bigger practical examples for good measure.

If you’re still with me, hooray! I hope that wasn’t too full-on. We’re definitely wading in deeper waters now, getting to the more advanced parts of the spec. All the more reason to keep asking questions, though! I want to make this as clear as possible, so don’t hesitate to get in touch.

For now until we talk about Applicative, though, it’s goodbye from me! Keep at it, Apply yourself (zing - this blog has jokes now!), and take care ♥