Understanding Applicative functors would require some understanding about Functors and Monads. We shall try to look at brief definitions for the scope of this article.
According to Haskell docs, a functor is simply something that can be mapped over.
In other words, it is an abstraction for a context with the ability to apply a function to all the things inside the context. The context can be defined as container, computation etc. E.g.; a list or sequence is a container of homogenous elements. You can apply a function to each of the elements, which produces a new sequence of elements transformed by the function.
Let’s create a new functor, implementing
myfmap such that
Data.Map is an instance of the new functor typeclass.
myfmap applies a function
f to the value(s) inside functor’s context while preserving the context.
import Data.Map as DataMap import Data.List as DataList class MyFunctor f where myfmap :: (a -> b) -> f a -> f b instance (Ord k) => MyFunctor (DataMap.Map k) where myfmap f x = DataMap.fromList $ DataList.map (\(p,q) ->(p,f q)) $ DataMap.toList x
In computer science, applicative is an abstraction for a context, and it has the ability to apply functions in the same type of context to all elements in the context. For example, a sequence A which has homogenous elements, and sequence B which consists of functions that can be applied to sequence A , which produce a new sequence of elements transformed by all the functions. (Super confusing? Sorry 😕 . We will clear it up)
The Applicative typeclass in Haskell is located in Control.Applicative module and defines pure and
class (Functor f) => Applicative f where pure :: a -> f a (<*>) : f (a -> b) -> f a -> f b
An Applicative has to be Functor. This is a class constraint.
puretakes any context and returns an Applicative the value of context inside it.
<*>is a representation of
<*>takes a functor that has a function in it and another functor and run the function from first functor and maps it over the second functor. Whereas
fmapwhich takes a function and functor and applies the function inside the functor.
List an an Applicative!
List also instantiates an applicative typeclass, with implementation as :
instance Applicative  where pure x = [x] fs <*> xs = [f x | f <- fs, x <- xs]
Here, the implementation of
(<*>) is basically a list comprehension, where every function is applied to every value. For example :
ghci> (*) <$> [2, 3] <*> [4, 5] [8, 10, 12, 15]
If you want to apply each function in first list to the respective value in second list, the ZipList typeclass is very handy.
Curious case of Either
The base Monad instance for
Either is defined as follows.
instance Monad (Either e) where return = Right Left e >>= _ = Left e Right a >>= f = f a
This instance has inherent short-circuiting. But in case you would like to collect error messages which occur anywhere in the above computation, it goes against
Lazy evaluation is when we proceed from left to right, when a single computation “fails” into the Left then all the rest do as well.
(>>=)takes a function, maps it over an instance of a monad and then flattens the result.
(>>=) :: m a -> (a -> m b) -> m b
m aso long as it can run
(a -> m b). This demands that the value of a should ideally exists during the time of computation, and this is impossible for
So let’s try to solve the above problem by defining a Functor instance of Either .
instance Functor (Either a) where fmap f (Left x) = Left x fmap f (Right y) = Right (f y)
Things we can understand from the above definition :
We know the definition of
fmap :: (c -> d) -> f c -> f d.
If we replace
Either a, we get
fmap :: (c -> d) -> Either a c -> Either a d
The problem with this implementation of
Eitheris that we cannot map over Left .
To understand that, let
Either a b computation, which may succeed and return b or fail with error a , similar to monad instance. So the functor instance does not map over
Left values since you would want to map over the computation, if it fails, there is nothing to manipulate.
Implementing an Applicative instance
Applicative Monad instance cannot have a corresponding
As we saw the definition of Applicative, we will define pure and
(<*>) . Defining
pure is rather simple, as we want it return the Right element. Implementation of
(<*>) is little tricky . The following cases need to be considered for defining
instance Applicative (Either e) where pure = Right Right f <*> Right a = Right (f a) Left e <*> Right _ = Left e Right _ <*> Left e = Left e Left e1 <*> Left e2 = Left (e1 <> e2)
The first statement is the pure statement.
(<*>)allows evaluation in parallel instead of necessarily needing results from previous computation to compute present values.
Thus, we can use our purely
Applicative Eitherto collect errors, ignoring Right if any Left exist in the sequence.
As soon as it hits a Left, it aborts and returns that Left.
ghci> (++) <$> Left "Hello" <*> undefined Left "Hello" -- not undefined ghci> (++) <$> Right "Hello" <*> undefined *** Exception: Prelude.undefined -- undefined ghci> (++) <$> Right "Hello" <*> Left " World" Left " World" ghci> (++) <$> Right "Hello" <*> Right " World" Right "Hello World"
There’s some limitations to using purely applicative functor. As we saw in the definition of
(>>=) :: m a -> (a -> m b) -> m b ; which means that without
(>>=) you can’t pick “what to do next based on what came before”.
Also, if you take a generally pure function and feed the applicative arguments to it, like
> f :: a -> b -> c > f <$> getLine <*> getProcessID "chrome" <*> getFreeMemory
All the arguments will get evaluated, no matter what. That is, you cannot express — “if the second argument exits, abort the rest of the computation”. Due to this, anything recursive, as well as most interactive programs do not use Applicatives. Monads, on the other hand are a good choice in that case.
More Reading :