Monads do not Compose

Monads do not Compose
@philip_schwarzslides by
inspired by (and with excerpts from)
Functional Programming in Scala A companion booklet to
Functional Programming in Scala
not in a generic way - there is no general way of composing monads

A Monad is both a Functor and an Applicative.
If you want to know more about Applicative then see the following
@philip_schwarz
https://www.slideshare.net/pjschwarz/ https://www.slideshare.net/pjschwarz/applicative-functor-116035644

Functor trait Functor[F[_]] {
map def map[A, B](m: F[A], f: A ⇒ B): F[B]
Applicative trait Applicative[F[_]] extends Functor[F] {
unit def unit[A](a: ⇒ A): F[A]
map2 def map2[A,B,C](fa: F[A], fb: F[B])(f: (A, B) ⇒ C): F[C]
override def map[A,B](fa: F[A])(f: A ⇒ B): F[B] = map2(fa,unit(()))((a,_) ⇒ f(a))
traverse def traverse[A,B](as: List[A])(f: A ⇒ F[B]): F[List[B]] = as.foldRight(unit(List[B]()))((a, fbs) ⇒ map2(f(a), fbs)(_::_)
sequence def sequence[A](lfa: List[F[A]]): F[List[A]] = traverse(lfa)(fa ⇒ fa)
Monad trait Monad[F[_]] extends Applicative[F]
flatMap def flatMap[A,B](ma: F[A])(f: A ⇒ F[B]): F[B]
override def map[A,B](m: F[A])(f: A ⇒ B): F[B] = flatMap(m)(a ⇒ unit(f(a)))
override def map2[A,B,C](ma: F[A], mb: F[B])(f: (A, B) ⇒ C): F[C] = flatMap(ma)(a ⇒ map(mb)(b ⇒ f(a, b)))
compose def compose[A,B,C](f: A ⇒ F[B], g: B ⇒ F[C]): A ⇒ F[C] = a ⇒ flatMap(f(a))(g)
join def join[A](mma: F[F[A]]): F[A] = flatMap(mma)(ma => ma)
listMonad val listMonad = new Monad[List] {
override def unit[A](a: ⇒ A)= List(a)
override def flatMap[A,B](ma: List[A])(f: A ⇒ List[B]) = ma flatMap f
A Monad is both a Functor and an Applicative
Here we define a Monad in terms of unit and flatMap
Yes, sequence and traverse really belong on
a Traverse trait, but they are not the main
focus here, just examples of functions using
Applicative functions unit and map2.

Functors compose.
On the next slide we look at a couple of examples.
If you would like to know more about Functor composition then see the following
@philip_schwarz
https://www.slideshare.net/pjschwarz/functor-composition

trait Functor[F[_]] {
def map[A,B](fa: F[A])(f: A => B): F[B]
def compose[G[_]](G:Functor[G]):Functor[λ[α=>F[G[α]]]] = {
val self = this
new Functor[λ[α => F[G[α]]]] {
override def map[A, B](fga:F[G[A]])(f:A=>B):F[G[B]] =
self.map(fga)(ga => G.map(ga)(f))
}
}
}
// Functor[List[Option]] = Functor[List] compose Functor[Option]
val optionListFunctor = listFunctor compose optionFunctor
assert(optionListFunctor.map(List(Some(1),Some(2),Some(3)))(double) == List(Some(2),Some(4),Some(6)))
// Functor[Option[List]] = Functor[Option] compose Functor[List]
val listOptionFunctor = optionFunctor compose listFunctor
assert(listOptionFunctor.map(Some(List(1,2,3)))(double) == Some(List(2,4,6)))
val double: Int => Int = _ * 2
implicit val listFunctor = new Functor[List] {
def map[A, B](fa: List[A])(f: A => B): List[B] = fa map f
}
implicit val optionFunctor = new Functor[Option] {
def map[A, B](fa: Option[A])(f: A => B): Option[B] = fa map f
}
We first compose the List Functor with the Option Functor.
This allows us to map a function over lists of options. Then
we do the opposite: we compose the Option Functor with
the List Functor. This allows us to map a function over an
optional list.
using https://github.com/non/kind-projector
allows us to simplify type lambda ({type f[α] = F[G[α]]})#f
to this: λ[α => F[G[α]]]
Functors compose
The map function of the composite
Functor is the composition of the map
functions of the functors being composed.

Applicatives also compose.
We can compose Applicatives using a similar technique to the
one we used to compose Functors.
Let’s look at FPiS to see how it is done.
@philip_schwarz

EXERCISE 12.9
Hard: Applicative functors also compose another way! If F[_] and G[_] are applicative functors, then so is F[G[_]].
Implement this function:
def compose[G[_]](G: Applicative[G]): Applicative[({type f[x] = F[G[x]]})#f]
ANSWER TO EXERCISE 12.9
def compose[G[_]](G: Applicative[G]): Applicative[({type f[x] = F[G[x]]})#f] = {
val self = this
new Applicative[({type f[x] = F[G[x]]})#f] {
def unit[A](a: => A) = self.unit(G.unit(a))
override def map2[A,B,C](fga: F[G[A]], fgb: F[G[B]])(f: (A,B) => C) =
self.map2(fga, fgb)(G.map2(_,_)(f))
}
}
(by Paul Chiusano and Runar Bjarnason)
@pchiusano @runarorama
(by Runar Bjarnason)
@runarorama

In the next two slides we look at a couple of
examples of composing Applicatives.

trait Applicative[F[_]] extends Functor[F] {
def unit[A](a: => A): F[A]
def map2[A,B,C](fa: F[A], fb: F[B])(f: (A, B) => C): F[C]
def map[A,B](fa: F[A])(f: A => B): F[B] =
map2(fa, unit(()))((a, _) => f(a))
def compose[G[_]](G: Applicative[G]): Applicative[λ[α => F[G[α]]]] = {
val self = this
new Applicative[λ[α => F[G[α]]]] {
def unit[A](a: => A): F[G[A]] = self.unit(G.unit(a))
override def map2[A,B,C](fga:F[G[A]],fgb:F[G[B]])(f:(A,B)=>C):F[G[C]] =
self.map2(fga, fgb)(G.map2(_,_)(f))
}
}
}
val optionApplicative = new Applicative[Option] {
def unit[A](a: => A): Option[A] = Some(a)
def map2[A,B,C](fa:Option[A],fb:Option[B])
(f:(A,B)=>C):Option[C] =
(fa, fb) match {
case (Some(a), Some(b)) => Some(f(a,b))
case _ => None
}
}
val listApplicative = new Applicative[List] {
def unit[A](a: => A): List[A] = List(a)
def map2[A, B, C](fa:List[A],fb:List[B])
(f:(A,B)=>C):List[C] =
for {
a <- fa
b <- fb
} yield f(a, b)
}
Applicatives compose
The unit function of the composite Applicative first lifts
its parameter into inner Applicative G and then lifts the
result into outer Applicative F.
The map2 function of the composite Applicative is the
composition of the map2 functions of the applicatives
being composed. It uses the map2 function of F to
break through the outer Applicative and the map2
function of G to break through the inner Applicative.
This is how it manages to break through the two layers
of Applicative in the composite.
Let’s create an Applicative instance for Option and one for List.

// Applicative[Option] compose Applicative[List] = Applicative[Option[List]]
val listOptionApplicative = optionApplicative compose listApplicative
assert( optionApplicative.map2( Option(1), Option(2) )(add)
== Option(3) )
assert( listApplicative.map2( List(1,2,3), List(4,5,6))(add)
== List(5,6,7,6,7,8,7,8,9))
assert( listOptionApplicative.map2( Option( List( 1,3,5 ) ), Option( List(2,4,6) ) )(add)
== Option( List(3,5,7,5,7,9,7,9,11) ) )
assert( listOptionApplicative.map( Option( List(1,2,3) ) )(double)
== Option( List(2,4,6) ) )
// Applicative[List] compose Applicative[Option] = Applicative[List[Option]
val optionListApplicative = listApplicative compose optionApplicative
assert(
(optionListApplicative map2(
List( Option(1), Option(2) ),
List( Option(3), Option(4) )
)(add)
== List( Option(4), Option(5), Option(5), Option(6) ) )
assert( optionListApplicative.map2( List( Option(1), Option(2) ), List( Option(3), Option(4) ) )(add)
== List( Option(4), Option(5), Option(5), Option(6) ) )
assert( optionListApplicative.map( List( Option(1), Option(2) ) )(double)
== List( Option(2), Option(4) ) )
Let’s create an Applicative that is the
composition of our List Applicative
and our Option Applicative. Let’s
then have a go at mapping a binary
function over two Lists of Options.
Now lets create the opposite
composite Applicative and have a go
at mapping a binary function over two
Optional Lists.
val add: (Int,Int) => Int = _ + _ val double: Int => Int = _ * 2
Applicatives compose

What about a Monad? Is it possible for the Monad
trait to have a compose function that takes any other
Monad and returns a composite Monad?
Let’s try implementing such a compose function.

trait Monad[F[_]] extends Applicative[F] {
def flatMap[A, B](ma: F[A])(f: A => F[B]): F[B]
override def map[A, B](m: F[A])(f: A => B): F[B] =
flatMap(m)(a => unit(f(a)))
override def map2[A, B, C](ma: F[A], mb: F[B])(f: (A, B) => C): F[C] =
flatMap(ma)(a => map(mb)(b => f(a, b)))
def compose[G[_]](G: Monad[G]): Monad[λ[α => F[G[α]]]] = {
val self = this
new Monad[λ[α => F[G[α]]]] {
def flatMap[A, B](fga: F[G[A]])(f: A => F[G[B]]): F[G[B]] = {
self.flatMap(fga) { ga =>
G.flatMap(ga) { a =>
val fgb: F[G[B]] = f(a)
??? // this inner flatMap must return G[_] but all we have is an F[G[_]
??? // to obtain a G[_] we'd have to swap F and G and return G[F[_]]
}
??? // this outer flatMap must return F[G[_]] but all we have is a G[_]
??? // had we been able to swap F and G in the inner flatMap we would have a G[F[_]]
??? // so we would have to swap G and F again to get an F[G[_]]
}
}
}
}
The unit function of the composite Monad first lifts its
parameter into inner Monad G and then lifts the result into
outer Monad F.
The flatMap function of the composite Monad needs to be
the composition of the flatMap functions of the monads
being composed. It needs to use the flatMap function of F
to break through the outer Monad and the flatMap function
of G to break through the inner Monad. It would then be able
to break through the two layers of Monad in the composite.
But we are not able to write suitable functions to pass to
the flatMap functions of the inner and outer Monads.
Do Monads compose ?
@philip_schwarz

def join[A](mma: F[F[A]]): F[A]
def flatMap[A, B](ma: F[A])(f: A => F[B]): F[B] = join(map(ma)(f))
val self = this
def join[A](fgfga: F[G[F[G[A]]]]): F[G[A]] = {
self.join(G.join(fgfga)) // does not compile - it is impossible to flatten fgfga
}
}
}
Do Monads compose ?
It is easier to see what the problem is if we change the Monad
trait so that it is defined in terms of map, join and unit rather
than in terms of flatMap and unit.
We then have to implement a join function for the composite
Monad (rather than a flatMap function).
The join function has to turn an F[G[F[G[A]]]]]]]]):
into an F[G[A]], which is not possible using the join
functions of the Monads being composed.

def join[A](mma: F[F[A]]): F[A]
def flatMap[A, B](ma: F[A])(f: A => F[B]): F[B] = join(map(ma)(f))
val self = this
def join[A](fgfga: F[G[F[G[A]]]]): F[G[A]] = {
self.join(G.join(fgfga)) // does not compile - it is impossible to flatten fgfga
val ffgga: F[F[G[G[A]]]] = ??? // if it were possible to rearrange fgfga into ffgga
val fgga: F[G[G[A]]] = self.join(ffgga) // then we could flatten ffgga to fgga
val fga: F[G[A]] = self.map(fgga)(gga => G.join(gga)) // and then flatten fgga to fga
fga
}
}
}
Do Monads compose ?
If join were able to turn its parameter F[G[F[G[A]]]] into
F[F[G[G[A]]]] then it would be able to flatten the two Fs
and the 2 Gs and be left with the desired F[G[A]].
But is doing something like that possible and in which cases?
@philip_schwarz

In the next two slides we look at what FPiS says about
Monad composition, and we’ll see that it discusses turning
F[G[F[G[A]]]] into F[F[G[G[A]]]].

Answer to Exercise 12.11
You want to try writing flatMap in terms of Monad[F] and Monad[G].
def flatMap[A,B](mna: F[G[A]])(f: A => F[G[B]]): F[G[B]] =
self.flatMap(na => G.flatMap(na)(a => ???))
Here all you have is f, which returns an F[G[B]]. For it to have the appropriate type to return from the argument
to G.flatMap, you’d need to be able to “swap” the F and G types. In other words, you’d need a distributive
law. Such an operation is not part of the Monad interface.
EXERCISE 12.11
Try to write compose on Monad. It’s not possible, but it is instructive to attempt it and understand why this is the case.
def compose[G[_]](G: Monad[G]): Monad[({type f[x] = F[G[x]]})#f]
@runarorama

Many Concepts Go Well Together
Before we delve into monads, let us consider those cases where there are no problems. If we know that both f and
g are functors, we can make a new functor out of their composition.
…
Traversables provide a similar interface to functors but work with functions of the form a -> f b, where f is an
applicative functor.
…
Given these similarities with Functor, we can reuse the idea of mapping twice to obtain an instance for the
composition of two, traversable functors
...
The applicative functor is another structure that works well under composition.
…
…
It seems like the promise of composition is achieved: start with a set of primitive functors — which might also be
traversables, applicatives, or alternatives — and compose them as desired. The resulting combination is
guaranteed to support at least the same operations as its constituents. If only that were true of monads.
But Monads Do Not
As you might have already guessed, it is not possible to take two monads f and g and compose them into a new
monad f :.: g in a generic way. By generic way, we mean a single recipe that works for every pair of monads. Of
course, there are some pairs of monads that can be combined easily, like two Readers, and others that need a bit
more work, like lists and optionals, as shown at the beginning of the chapter. But, stressing the point once again,
there is no uniform way to combine any two of them.
The Book of Monads: Master the theory and practice
of monads, applied to solve real world problems
Alejandro Serrano Mena
@trupill

In order to understand why, we are going to consider in greater detail the idea of monads as boxes
class Monad m where
return :: a -> m a
join :: m (m a) -> m
We have just seen how to combine the fmap operations of two functors and the pure operations — return for
monads — of two applicative functors. The latter method, return, poses no problem for composition: just take
the definition of pure we described for the composition of two applicative functors. Therefore, we must
conclude that join is the one to blame. The join operation for the composition of two monads has the following
type:
join :: (f :.: g) (( f :.: g) a) -> (f :.: g) a
If we leave out for a moment the newtype, this type amounts to:
join :: f (g (f (g a))) -> f (g a)
In a monad, we only have methods that add layers of monads — return and fmap — and a method that flattens
two consecutive layers of the same monad. Alas, f (g (f (g a))) has interleaved layers, so there is no way to use
join to reduce them to f (g a). As you can see, the reason why we cannot combine two monads is not very deep.
It is just a simple matter of types that do not match. But the consequences are profound, since it tells us that
monads cannot be freely composed.
@trupill
map fmap
flatMap bind
flatten/join join
unit pure/return

Distributive Laws for Monads
One way to work around this problem is to provide a function that swaps the middle layers:
swap :: g (f a) -> f (g a)
This way, we can first turn f (g (f (g a))) into f (f (g (g a))) by running swap under the first layer. Then we can join
the two outer layers, obtaining f (g (g a)) as a result. Finally, we join the two inner layers by applying the fmap
operation under the functor f.
…
When such a function, swap, exists for two monads f and g, we say that there is a distributive law for those
monads. In other words, if f and g have a distributive relationship, then they can be combined into a new
monad.
…
Some monads even admit a distributive law with any other monad. The simplest example is given by Maybe.
…
@trupill

The list monad
The discussion above contains a small lie. We definitely need a swap function if we want to combine two
monads, but this is not enough. The reason is that a well-typed implementation may lead to a combined monad
that violates one of the monad laws. The list monad is a well-known example of this.
…
For the specific situation of the list monad, the cases for which its combination with another monad m lead to a
new monad have been described. In particular, for the swap procedure to be correct, m needs to be a
commutative monad, that is, the following blocks must be equal. The difference lies in the distinct order in which
xs and ys are bound:
do x <- xs do y <- ys
y <- ys ≡ x <- xs
return (x, y) return (x, y)
The Maybe monad is commutative, since the order of failure does not matter for a final absent value, and in the
case in which both elements are Just values, the result is the same. On the other hand, the list monad is not
commutative: both blocks will ultimately result in the same elements, but the order in which they are produced
will be different.
…
@trupill

12.7.6 Monad composition
Let’s now return to the issue of composing monads. As we saw earlier in this chapter, Applicative instances
always compose, but Monad instances do not. If you tried before to implement general monad composition, then you
would have found that in order to implement join for nested monads F and G, you’d have to write something of a
type like F[G[F[G[A]]]] => F[G[A]]. And that can’t be written generally.
But if G also happens to have a Traverse instance, we can sequence to turn G[F[_]] into F[G[_]], leading to
F[F[G[G[A]]]]. Then we can join the adjacent G layers using their respective Monad instances.
EXERCISE 12.20
Hard: implement the composition of two monads where one of them is traversable.
def composeM[F[_],G[_]](F: Monad[F], G: Monad [G], T: Traverse[G]):
Monad[({type f[x] = F[G[x]]})#f]
Answer to Exercise 12.20
def composeM[G[_],H[_]](implicit G: Monad[G], H: Monad[H], T: Traverse[H]):
Monad[({type f[x] = G[H[x]]})#f] = new Monad[({type f[x] = G[H[x]]})#f] {
def unit[A](a: => A): G[H[A]] = G.unit(H.unit(a))
override def flatMap[A,B](mna: G[H[A]])(f: A => G[H[B]]): G[H[B]] =
G.flatMap(mna)(na => G.map(T.traverse(na)(f))(H.join))
}
@runarorama

6 The name Traversable is already taken by an unrelated trait in the Scala standard library.
The interesting operation here is sequence. Look at its signature closely. It takes F[G[A]] and swaps the order of F and G, so
long as G is an applicative functor. Now, this is a rather abstract, algebraic notion. We’ll get to what it all means in a minute, but
first, let’s look at a few instances of Traverse.
trait Traverse[F[_]] {
def traverse[M[_]:Applicative,A,B](fa: F[A])(f: A => M[B]): M[F[B]]
sequence(map(fa)(f))
def sequence[M[_]:Applicative,A](fma: F[M[A]]): M[F[A]] =
traverse(fma)(ma => ma)
}
If you are interested in Traverse, then see the following for an in depth explanation.
@philip_schwarz
https://www.slideshare.net/pjschwarz/sequence-and-traverse-part-1
As you can see on the previous slide, composeM uses Traverse.traverse
There is a lot to say about the Traverse trait that is out of scope for this slide deck.
The following excerpt from FPiS is sufficient for our current purposes.
The next slide is just a teaser to whet your appetite.@philip_schwarz

def map[B](fa: F[A])(f: A => B): F[B] =
map2(fa, unit(()))((a, _) => f(a))
def sequence[A](fas: List[F[A]]): F[List[A]] =
traverse(fas)(fa => fa)
def traverse[A,B](as: List[A])(f: A => F[B]): F[List[B]]
as.foldRight(unit(List[B]()))((a, fbs) => map2(f(a), fbs)(_ :: _))
…
}
trait Foldable[F[_]] {
import Monoid._
def foldRight[A, B](as: F[A])(z: B)(f: (A, B) => B): B =
foldMap(as)(f.curried)(endoMonoid[B])(z)
def foldLeft[A, B](as: F[A])(z: B)(f: (B, A) => B): B =
foldMap(as)(a => (b: B) => f(b, a))(dual(endoMonoid[B]))(z)
def foldMap[A,B](as:F[A])(f:A=>B)(implicit mb: Monoid[B]):B =
foldRight(as)(mb.zero)((a, b) => mb.op(f(a), b))
def concatenate[A](as: F[A])(implicit m: Monoid[A]): A =
foldLeft(as)(m.zero)(m.op)
}
}
def flatMap[A,B](ma: F[A])(f: A => F[B]): F[B]
override def map[A,B](m: F[A])(f: A => B): F[B] =
override def map2[A,B,C](ma:F[A], mb:F[B])(f:(A, B) => C): F[C] =
}
trait Traverse[F[_]] extends Functor[F] with Foldable[F] { self =>
def traverse[M[_]:Applicative,A,B](fa:F[A])(f:A=>M[B]):M[F[B]]
def sequence[M[_] : Applicative, A](fma: F[M[A]]): M[F[A]] =
type Id[A] = A
val idMonad = new Monad[Id] {
def unit[A](a: => A) = a
override def flatMap[A, B](a: A)(f: A => B): B = f(a)
}
def map[A, B](fa: F[A])(f: A => B): F[B] =
traverse[Id, A, B](fa)(f)(idMonad)
import Applicative._
override def foldMap[A,B](as: F[A])(f: A => B)
(implicit mb: Monoid[B]): B =
traverse[({type f[x] = Const[B,x]})#f,A,Nothing](
as)(f)(monoidApplicative(mb))
…
}
type Const[M, B] = M
implicit def monoidApplicative[M](M: Monoid[M]) =
new Applicative[({ type f[x] = Const[M, x] })#f] {
def unit[A](a: => A): M = M.zero
def map2[A,B,C](m1: M, m2: M)(f: (A,B) => C): M = M.op(m1,m2)
}
trait Monoid[A] {
def op(x: A, y: A): A
def zero: A
}

In the next two slides we look at an example of
composing a Monad with a traversable Monad.
@philip_schwarz

trait Traverse[F[_]] {
def traverse[M[_]:Applicative,A,B](fa: F[A])(f: A => M[B]): M[F[B]]
def sequence[M[_]:Applicative,A](fma: F[M[A]]): M[F[A]] =
}
def join[A](mma: F[F[A]]): F[A] = flatMap(mma)(ma => ma)
def flatMap[A, B](ma: F[A])(f: A => F[B]): F[B]
def composeM[G[_]](G: Monad[G], T: Traverse[G]): Monad[λ[α => F[G[α]]]] = {
val self = this
def flatMap[A, B](fga: F[G[A]])(f: A => F[G[B]]): F[G[B]] = {
self.flatMap(fga){ ga => self.map(T.traverse(ga)(f)(self))(G.join) }
}
}
}
def map[A,B](fa: F[A])(f: A => B): F[B] =
map2(fa, unit(()))((a, _) => f(a))
}
}
Example of composing a Monad with a traversable Monad

val optionMonad = new Monad[Option] {
def flatMap[A,B](ma: Option[A])(f: A => Option[B]): Option[B] =
ma match {
case Some(a) => f(a)
case None => None
}
}
assert( traversableListMonad.traverse(List("12","23"))(parseInt)(optionApplicative) == Some(List(12, 23)))
assert( traversableListMonad.traverse(List("12","2x"))(parseInt)(optionApplicative) == None)
val listOptionMonad = optionMonad.composeM(traversableListMonad,traversableListMonad)
assert( listOptionMonad.flatMap( Option( List( "12", "34" ) ) )(charInts) == Option( List( '1', '2', '3', '4' ) ) )
assert( listOptionMonad.flatMap( Option( List( "12", "3x" ) ) )(charInts) == None )
val optionApplicative = new Applicative[Option] {
def map2[A,B,C](fa:Option[A],fb:Option[B])(f:(A,B)=>C):Option[C] =
(fa, fb) match {
case (Some(a), Some(b)) => Some(f(a,b))
case _ => None
}
}
val traversableListMonad = new Monad[List] with Traverse[List] {
def unit[A](a: => A): List[A] = List(a)
def flatMap[A,B](ma: List[A])(f: A => List[B]): List[B] = {
ma.foldRight(List.empty[B])((a,bs) => f(a) ::: bs)
}
override def traverse[M[_],A,B](as: List[A])
(f: A => M[B])(implicit M: Applicative[M]): M[List[B]] =
as.foldRight(M.unit(List[B]()))((a, fbs) => M.map2(f(a), fbs)(_ :: _))
}
val parseInt: String => Option[Int] =
s => Try{ s.toInt }.toOption
val charInts: String => Option[List[Char]] =
s => parseInt(s).map(_.toString.toList)
assert( charInts("12") == Option( List('1', '2')) )
assert( charInts("1x") == None )
Example of composing a Monad with a traversable Monad

Expressivity and power sometimes come at the price of compositionality and modularity.
The issue of composing monads is often addressed with a custom-written version of each monad that’s specifically
constructed for composition. This kind of thing is called a monad transformer. For example, the OptionT monad
transformer composes Option with any other monad:
The flatMap definition here maps over both M and Option, and flattens structures like
M[Option[M[Option[A]]]] to just M[Option[A]]. But this particular implementation is specific to Option.
And the general strategy of taking advantage of Traverse works only with traversable functors. To compose with
State (which can’t be traversed), for example, a specialized StateT monad transformer has to be written.
There’s no generic composition strategy that works for every monad.
See the chapter notes for more information about monad transformers.
case class OptionT[M[_],A](value: M[Option[A]])(implicit M: Monad[M]) {
def flatMap[B](f: A => OptionT[M, B]): OptionT[M, B] =
OptionT(value flatMap {
case None => M.unit(None)
case Some(a) => f(a).value
})
}
There is no generic composition strategy that works for every monad The issue of composing monads is often addressed with monad transformers

Monad transformers
A monad transformer is a data type that composes a particular monad with any other monad, giving us a
composite monad that shares the behavior of both.
There is no general way of composing monads. Therefore we have to have a specific transformer for each monad.
For example, OptionT is a monad transformer that adds the behavior of Option to any other monad. The type
OptionT[M, A] behaves like the composite monad M[Option[_]]. Its flatMap method binds over both the M and the
Option inside, saving us from having to do the gymanstics of binding over both.
Scalaz provides many more monad transformers, including StateT, WriterT, EitherT, and ReaderT (also known
as Kleisli).
@runarorama

@trupill
A Solution: Monad Transformers
It would be impolite to thoroughly describe a problem — the composition of monads — and then not describe
at least one of the solutions. That is the goal of this chapter, to describe how monad transformers allow us to
combine the operations of several monads into one, single monad. It is not a complete solution, however, since
we need to change the building blocks: instead of composing several different monads into a new monad, we
actually enhance one monad with an extra set of operations via a transformer.
Alas, there is one significant downside to the naïve, transformers approach: we cannot abstract over monads
that provide the same functionality but are not identical. This hampers the maintainability of our code, as any
change in our monadic needs would lead to huge rewrites. The classic solution is to introduce type classes for
different sets of operations. Consider that solution carefully, as it forms the basis of one of the styles for
developing your own monads.
One word of warning before proceeding: monad transformers are a solution to the monad composition
problem. But they are not the solution. Another approach, effects, is gaining traction in the functional
programming community. The right way to design an effects system is an idea that is still in flux, however, as
witnessed by its many, different implementations.

If you are interested in knowing more about
Monad Transformers then see the following
@philip_schwarz
https://www.slideshare.net/pjschwarz/ https://www.slideshare.net/pjschwarz/monad-transformers-part-1

Monads do not Compose

More Related Content

What's hot (20)

Similar to Monads do not Compose (20)

More from Philip Schwarz (20)

Recently uploaded (20)

Monads do not Compose