Applicative Functors

The Applicative typeclass extends Functors with additional methods. “pure” takes a value and wraps it in a Functor instance, while "<*>" applies a function found in one functor to a value in another functor. The “<$>” operator is just an infix version of fmap (we’ll see how that’s useful later).

class (Functor f) => Applicative f where
  pure :: a -> f a
  (<*>) :: f (a -> b) -> f a -> f b
  (<$>) :: (a -> b) -> f a -> f b
  (<$>) = fmap

Here’s how the Maybe type is defined as an Applicative Functor:

instance Applicative Maybe where
  pure = Just
  Nothing <*> _ = Nothing
  Just f <*> Nothing = Nothing
  Just f <*> Just x = Just $ f x

This time around, we can only carry out function application if both Maybe values are Justs. If either is a Nothing, we are either lacking a function to apply or a value to apply it to!

Now, we can do:

_ = pure (*2) <*> Just 5

Here, the <*> “operator” is just like function application, except lifted to functors.

More importantly, due to the currying of functions, we can also do:

_ = pure (*) <*> Just 2 <*> Just 5

Or, equivalently:

_ = fmap (*) (Just 2) <*> Just 5

Which allows us to simply write:

_ = (*) <$> Just 2 <*> Just 5

If we wanted to, we could easily implement arbitrary arity lifts, e.g.,

liftA2 :: Applicative f => (a -> b -> c) -> f a -> f b -> f c
liftA2 f a1 a2 = f <$> a1 <*> a2

liftA3 :: Applicative f => (a -> b -> c -> d) -> f a -> f b -> f c -> f d
liftA3 f a1 a2 a3 = f <$> a1 <*> a2 <*> a3

_ = let f = liftA3 (\x y z -> x + y - z)
    in f (Just 1) (Just 3) (Just 5)

But this isn’t necessary if we’re willing to just write in the “applicative style”, i.e., using <*> for function application.

Functors as Computational Contexts

We can think of Maybe qua Functor as providing a “computational context” for values. “Just” represents a successful context, while “Nothing” represents a failure. “pure” takes a value and gives us an initial (successful) context, and the <*> operator allows us to combine it with other contexts, while automatically detecting and propagating failure.

E.g., consider:

_ = (\x y z -> x++y++z) <$> Just "Hello" <*> Just "Hola" <*> Just "Hi"
_ = (\x y z -> x++y++z) <$> Just "Hello" <*> Just "Hola" <*> Nothing
_ = (\x y z -> x++y++z) <$> Just "Hello" <*> Nothing     <*> Just "Hi"
_ = (\x y z -> x++y++z) <$> Nothing      <*> Just "Hola" <*> Just "Hi"

The first expression is the only to succeed (with the result “HelloHolaHi”), because all contexts being combined are “Just” values. In subsequent expressions, the applicative machinery takes care of detecting the “Nothing” and causing the entire expression to fail. Note that the original function (a lambda, in this case) knows nothing about the possibility of failure.

Lists as Applicative Functors

Here’s how we make lists applicative functors.

instance Applicative [] where
  pure x = [x]
  fs <*> xs = [f x | f <- fs, x <- xs]

The <*> operator for lists will go through all the different ways of combining elements from its two arguments. When used as follows:

_ = pure (+2) <*> [1..5] -- => [3,4,5,6,7]

I.e., where the first list only contains one value, this behaves just like fmap (and map). When applied to multi-valued lists, however:

_ = [(*2), (*3)] <*> [1..5] -- => [2,4,6,8,10,3,6,9,12,15]

or, equivalently:

_ = pure (*) <*> [2,3] <*> [1..5]

We get the results of all the combinations of functions and arguments drawn from the lists in the expression.

The computational context represented by a non-empty list can be thought of as one where there are many different “answers” to the expression being evaluated. The applicative machinery automatically combines list functors to yield all these answers. This is sometimes referred to as non-deterministic programming — i.e., where we come up with not one, but all possible answers to a particular problem.

Here’s the cartesian product function implemented trivially with lifting:

cartesianProd :: [a] -> [b] -> [(a,b)]
cartesianProd = liftA2 (,)

But what happens when we insert an empty list into an applicative application?

_ = pure (*) <*> [] <*> [1..10] -- => []

Just as with Nothing for the Maybe functor, empty lists represent “failure” — i.e., no possible way of computing a result.

The road forward

Applicatives are useful when we want to apply pure functions to functors, but what happens when the functions we want to apply can themselves fail (i.e., produce some computational context)?

E.g., consider a collection of binary, numeric functions which each may either fail or compute a result:

binFuncA :: Num a => a -> a -> Maybe a
binFuncA = undefined

binFuncB :: Num a => a -> a -> Maybe a
binFuncB = undefined

binFuncC :: Num a => a -> a -> Maybe a
binFuncC = undefined

Having used one of the functions, we’d like to use its result in another function of the collection, and so on — each time taking into account the possibility of failure, e.g.,

let r1 = binFuncA a b
    r2 = binFuncB r1 c
in binFuncC r2 d

This code is incorrect, of course, because the results (r1, r2) of the functions are Maybe values, while the inputs are just numbers.

The applicative style is also kind of awkward here:

binFuncC <$> (binFuncB <$> (binFuncA <$> Just a <*> Just b) <*> Just c) Just d

So … what we need is another type of functor that accept functions which produce functors as output — these outputs must somehow be combined with the original functors in order to preserve all computational context.