Tag Archives: typeclass

Scala Cats Typeclasses At A Glance

Scala Cats comes with a rich set of typeclasses, each of which “owns” a well-defined autonomous problem space. Many of those typeclasses are correlated and some are extended from others.

In this blog post, we’re going to give an at-a-glance hierarchical view of some of the most common Cats typeclasses. For brevity, we’ll skip discussions re: their corresponding mathematical laws, which can be found in many relevant tech docs. Our focus will be more on highlighting the correlations among these typeclasses.

Common typeclass hierarchy

For the impatient, below is a diagram highlighting the hierarchical correlation.

Scala Cats Common Typeclass Hierarchy

Semigroup and Monoid

Let’s start with the simplest ones, Semigroup and Monoid.

Semigroup comes with the abstract method combine to be implemented with the specific “combine” computational logic such as the addition of integers, union of sets, etc.

trait Semigroup[A] {
  def combine(x: A, y: A): A
}

trait Monoid[A] extends Semigroup[A] {
  def combine(x: A, y: A): A
  def empty: A
}

Note that Monoid simply supplements Semigroup with empty as the “zero” or “identity” element, allowing aggregating operations of arbitrarily many elements (e.g. summation of numbers from an initial 0).

Example:

implicit def setSemigroup[A]: Semigroup[Set[A]] =
  new Semigroup[Set[A]] {
    def combine(s1: Set[A], s2: Set[A]) = s1 union s2
  }

// Or, using SAM for brevity
// implicit def setSemigroup[A]: Semigroup[Set[A]] = _ union _

val setSG = implicitly[Semigroup[Set[Char]]]
setSG.combine(Set('a', 'b'), Set('c'))
// Set('a', 'b', 'c')

implicit def setMonoid[A](implicit sg: Semigroup[Set[A]]): Monoid[Set[A]] =
  new Monoid[Set[A]] {
    def combine(s1: Set[A], s2: Set[A]) = sg.combine(s1, s2)
    def empty = Set()
  }

val setM = implicitly[Monoid[Set[Char]]]
List(Set('a','b'),Set('c'),Set('d','e')).
  foldLeft(setM.empty)(setM.combine(_, _))
// HashSet('e', 'a', 'b', 'c', 'd')

SemigroupK and MonoidK

With a similar correlation, SemigroupK and MonoidK are the higher-kinded version of Semigroup and Monoid, respectively. SemigroupK combines values within a given context and MonoidK ensures the existence of an “empty” context.

trait SemigroupK[F[_]] {
  def combineK[A](x: F[A], y: F[A]): F[A]
}

trait MonoidK[F[_]] extends SemigroupK[F] {
  def combineK[A](x: F[A], y: F[A]): F[A]
  def empty[A]: F[A]
}

Example:

implicit val listSemigroupK: SemigroupK[List] =
  new SemigroupK[List] {
    def combineK[A](ls1: List[A], ls2: List[A]) = ls1 ::: ls2
  }

val listSGK = implicitly[SemigroupK[List]]
listSGK.combineK(List(1,2), List(3))
// List(1, 2, 3)

implicit def listMonoidK(implicit sgk: SemigroupK[List]): MonoidK[List] =
  new MonoidK[List] {
    def combineK[A](ls1: List[A], ls2: List[A]) = sgk.combineK(ls1, ls2)
    def empty[A] = List.empty[A]
  }

val listMK = implicitly[MonoidK[List]]
List(List(1,2),List(3),List(4,5)).foldLeft[List[Int]](listMK.empty)(listMK.combineK(_, _))
// List(1, 2, 3, 4, 5)

Functor

Functor is a higher-kinded typeclass characterized by its method map which transforms some value within a given context F via a function.

trait Functor[F[_]] {
  def map[A, B](fa: F[A])(f: A => B): F[B]
}

Example:

implicit val listFunctor: Functor[List] =
  new Functor[List] {
    def map[A, B](ls: List[A])(f: A => B) = ls.map(f)
  }

val listF = implicitly[Functor[List]]
listF.map(List(1,2,3))(i => s"#$i!")
// List("#1!", "#2!", "#3!")

Monad

Monad enables sequencing of operations in which resulting values from an operation can be utilized in the subsequent one.

But first, let’s look at typeclass FlatMap.

trait FlatMap[F[_]] extends Apply[F] {
  def flatMap[A, B](fa: F[A])(f: A => F[B]): F[B]
  def map[A, B](fa: F[A])(f: A => B): F[B]
  @tailrec
  def tailRecM[A, B](init: A)(f: A => F[Either[A, B]]): F[B]
}

FlatMap extends Apply whose key methods aren’t what we would like to focus on at the moment. Rather, we’re more interested in method flatMap which enables sequential chaining of operations.

In addition, method tailRecM is a required implementation for stack-safe recursions on the JVM (which doesn’t natively support tail call optimization).

Monad inherits almost all its signature methods from FlatMap.

trait Monad[F[_]] extends FlatMap[F] with Applicative[F] {
  def flatMap[A, B](fa: F[A])(f: A => F[B]): F[B]
  def pure[A](a: A): F[A]
  def map[A, B](fa: F[A])(f: A => B): F[B]
  @tailrec
  def tailRecM[A, B](init: A)(f: A => F[Either[A, B]]): F[B]
}

Monad also extends Applicative which we’ll get to (along with Apply) in a bit. For now, it suffices to note that Monad inherits pure from Applicative.

Even without realizing that Monad extends Functor (indirectly through FlatMap and Apply), one could conclude that Monads are inherently Functors by implementing map using flatMap and pure.

def map[A, B](fa: F[A])(f: A => B): F[B] = flatMap(fa)(a => pure(f(a)))

Example:

trait Monad[F[_]] {  // Skipping dependent classes
  def flatMap[A, B](fa: F[A])(f: A => F[B]): F[B]
  def pure[A](a: A): F[A]
  def map[A, B](fa: F[A])(f: A => B): F[B]
  def tailRecM[A, B](init: A)(f: A => F[Either[A, B]]): F[B]
}

implicit val optionMonad: Monad[Option] =
  new Monad[Option] {
    def flatMap[A, B](opt: Option[A])(f: A => Option[B]) = opt.flatMap(f)
    def pure[A](a: A) = Option(a)
    def map[A, B](opt: Option[A])(f: A => B): Option[B] = flatMap(opt)(a => pure(f(a)))
    @scala.annotation.tailrec
    def tailRecM[A, B](a: A)(f: A => Option[Either[A, B]]): Option[B] = f(a) match {
      case None => None
      case Some(leftOrRight) => leftOrRight match {
        case Left(a1) => tailRecM(a1)(f)
        case Right(b1) => Option(b1)
      }
    }
  }

val optMonad = implicitly[Monad[Option]]
optMonad.flatMap(Option(3))(i => if (i > 0) Some(s"#$i!") else None)
// Some("#3!")

Semigroupal and Apply

A higher-kinded typeclass, Semigroupal conceptually deviates from SemiGroup’s values combining operation to joining independent contexts in a tupled form “product”.

trait Semigroupal[F[_]] {
  def product[A, B](fa: F[A], fb: F[B]): F[(A, B)]
}

Despite the simplicity of method product (which is the only class method), Semigroupal lays out the skeletal foundation for the problem space of concurrency of independent operations, as opposed to Monad’s sequential chaining.

Next, Apply brings together the goodies of Semigroupal and Functor. Its main method ap has a rather peculiar signature that doesn’t look intuitively meaningful.

trait Apply[F[_]] extends Semigroupal[F] with Functor[F] {
  def ap[A, B](ff: F[A => B])(fa: F[A]): F[B]
  def map[A, B](fa: F[A])(f: A => B): F[B]
  def product[A, B](fa: F[A], fb: F[B]): F[(A, B)]
}

Conceptually, it can be viewed as a specialized map in which the transformation function is “wrapped” in the context.

By restructuring the type parameters in ap[A, B] and map[A, B], method product can be implemented in terms of ap and map.

// 1: Substitute `B` with `B => (A, B)`
def map[A, B](fa: F[A])(f: A => B => (A, B)): F[B => (A, B)]

// 2: Substitute `A` with `B` and `B` with `(A, B)`
def ap[A, B](ff: F[B => (A, B)])(fb: F[B]): F[(A, B)]

// Applying 1 and 2:
def product[A, B](fa: F[A], fb: F[B]): F[(A, B)] =
  ap(map(fa)(a => (b: B) => (a, b)))(fb)

Applicative

trait Applicative[F[_]] extends Apply[F] {
  def pure[A](a: A): F[A]
}

Like how Monoid supplements SemiGroup with the empty element to form a more “self-contained” typeclass, Applicative extends Apply and adds method pure which wraps a value in a context. The seemingly insignificant inclusion makes Applicative a typeclass capable of addressing problems within a particular problem space.

Similarly, Monad takes pure from Applicative along with the core methods from FlatMap to become another “self-contained” typeclass to master a different computational problem space.

Contrary to Monad’s chaining of dependent operations, Applicative embodies concurrent operations, allowing independent computations to be done in parallel.

We’ll defer examples for Applicative to a later section.

Foldable

Foldable offers fold methods that go over (from left to right or vice versa) some contextual value (oftentimes a collection) and aggregate via a binary function starting from an initial value. It also provides method foldMap that maps to a Monoid using an unary function.

trait Foldable[F[_]] {
  def foldLeft[A, B](fa: F[A], b: B)(f: (B, A) => B): B
  def foldRight[A, B](fa: F[A], lb: Eval[B])(f: (A, Eval[B]) => Eval[B]): Eval[B]
  def foldMap[A, B: Monoid](fa: F[A])(f: A => B): B
}

Note that the well known foldRight method in some Scala collections may not be stack-safe (especially in older versions). Cats uses a data type Eval in its foldRight method to ensure stack-safety.

Traverse

Traverse extends Functor and Foldable and provides method traverse. The method traverses and transforms some contextual value using a function that wraps the transformed value within the destination context, which as a requirement, is bound to an Applicative.

trait Traverse[F[_]] extends Functor[F] with Foldable[F] {
  def traverse[G[_]: Applicative, A, B](fa: F[A])(ff: A => G[B]): G[F[B]]
}

If you’ve used Scala Futures, method traverse (and the sequence method) might look familiar.

def sequence[G[_]: Applicative, A](fg: F[G[A]]): G[F[A]] =
  traverse(fg)(identity)

Method sequence has the effect of turning a nested context “inside out” and is just a special case of traverse by substituting A with G[B] (i.e. making ff an identity function).

Example: Applicative and Traverse

To avoid going into a full-on implementation of Traverse in its general form that would, in turn, require laborious implementations of all the dependent typeclasses, we’ll trivialize our example to cover only the case for Futures (i.e. type G = Future).

First, we come up with a specialized Traverse as follows:

import scala.concurrent.{ExecutionContext, Future}

trait FutureTraverse[F[_]] {  // Skipping dependent classes
  def traverse[A, B](fa: F[A])(ff: A => Future[B]): Future[F[B]]
}

For similar reasons, let’s also “repurpose” Applicative to include only the methods we need. In particular, we include method map2 which will prove handy for implementing the traverse method for FutureTraverse.

trait Applicative[F[_]] {  // Skipping dependent classes
  def map[A, B](fa: F[A])(f: A => B): F[B]
  def map2[A, B, Z](fa: F[A], fb: F[B])(f: (A, B) => Z): F[Z]
  def pure[A](a: A): F[A]
}

implicit val futureApplicative: Applicative[Future] =
  new Applicative[Future] {
    implicit val ec = ExecutionContext.Implicits.global
    def map[A, B](fa: Future[A])(f: A => B): Future[B] = fa.map(f)
    def map2[A, B, Z](fa: Future[A], fb: Future[B])(f: (A, B) => Z): Future[Z] =
      (fa zip fb).map(f.tupled)
    def pure[A](a: A): Future[A] = Future.successful(a)
  }

We implement map2 by tuple-ing the Futures and binary function via zip and tupled, respectively. With the implicit Applicative[Future] in place, we’re ready to implement FutureTraverse[List].

implicit val listFutureTraverse: FutureTraverse[List] =
  new FutureTraverse[List] {
    implicit val ec = ExecutionContext.Implicits.global
    implicit val appF = implicitly[Applicative[Future]]
    def traverse[A, B](ls: List[A])(ff: A => Future[B]): Future[List[B]] = {
      ls.foldRight[Future[List[B]]](Future.successful(List.empty[B])){ (a, acc) =>
        appF.map2(ff(a), acc)(_ :: _)
      }
    }
  }

import scala.concurrent.ExecutionContext.Implicits.global

val lsFutTraverse = implicitly[FutureTraverse[List]]
lsFutTraverse.traverse(List(1,2,3)){ i =>
  if (i > 0) Future.successful(s"#$i!") else Future.failed(new Exception())
}
// Future(Success(List("#1!", "#2!", "#3!")))

As a side note, we could implement traverse without using Applicative. Below is an implementation leveraging Future’s flatMap method along with a helper function (as demonstrated in a previous blog post about Scala collection traversal).

implicit val listFutureTraverse: FutureTraverse[List] =
  new FutureTraverse[List] {
    implicit val ec = ExecutionContext.Implicits.global
    def pushToList[A](a: A)(as: List[A]): List[A] = a :: as
    def traverse[A, B](ls: List[A])(ff: A => Future[B]): Future[List[B]] = {
      ls.foldRight[Future[List[B]]](Future.successful(List.empty[B])){ (a, acc) =>
        ff(a).map(pushToList).flatMap(acc.map)
      }
    }
  }

Orthogonal Typeclass In Scala

As an addendum to a previous blog post on the topic of ad-hoc polymorphism in Scala, I’m adding another common typeclass pattern as a separate post. The term “orthogonal” refers to a pattern that selected class attributes are taken out from the base class to form an independent typeclass.

Using an ADT similar to the Car/Sedan/SUV example used in that previous post, we first define `trait Car` as follows:

trait Car {
  def vin: String
  def model: String
  def price: Int
}

Unlike how the base trait was set up as a typeclass in the ad-hoc polymorphism example, `trait Car` is now an ordinary trait. But the more significant difference is that method `setPrice()` is no longer in the base class. It’s being constructed “orthogonally” in a designated typeclass:

trait Settable[T] {
  def setPrice(t: T, amount: Int): T
}

Similar to how implicit conversions are set up for ad-hoc polymorphism, implicit values are defined within the companion objects for the individual child classes to implement method `setPrice()` for specific car types.

import scala.language.implicitConversions

case class Sedan(vin: String, model: String, price: Int)
object Sedan {
  implicit val carSedan = new Settable[Sedan] {
    def setPrice(t: Sedan, amount: Int) = t.copy(price = amount)
  }
}

case class Sports(vin: String, model: String, price: Int)
object Sports {
  implicit val carSports = new Settable[Sports] {
    def setPrice(t: Sports, amount: Int) = t.copy(price = amount)
  }
}

The specific method implementations are then abstracted into a “unified” method, `setNewPrice()`, via an implicit constructor argument by passing the `Settable` typeclass into the `CarOps` implicit class:

implicit class CarOps[T](t: T)(implicit ev: Settable[T]) {
  def setNewPrice(amount: Int) = ev.setPrice(t, amount)
}

Testing it out:

val sedan1 = Sedan("1ABC*234", "Honda Accord", 20500)

sedan1.setNewPrice(19500)
// res1: Sedan = Sedan("1ABC*234", "Honda Accord", 19500.0)

Putting all method implementations in one place

It’s worth noting that having the implicit values for method implementations defined in the companion objects for the individual classes is just one convenient way. Alternatively, these implicit values could all be defined in one place:

trait Car {
  def vin: String
  def model: String
  def price: Int
}

case class Sedan(vin: String, model: String, price: Int)

case class Sports(vin: String, model: String, price: Int)

trait Settable[T] {
  def setPrice(t: T, amount: Int): T
}

object CarOps {
  implicit val carSedan = new Settable[Sedan] {
    def setPrice(t: Sedan, amount: Int) = t.copy(price = amount)
  }

  implicit val carSports = new Settable[Sports] {
    def setPrice(t: Sports, amount: Int) = t.copy(price = amount)
  }
}

import CarOps._

implicit class CarOps[T](t: T)(implicit ev: Settable[T]) {
  def setNewPrice(amount: Int) = ev.setPrice(t, amount)
}

A benefit of putting all method implementations in one place is that new methods can be added without touching the base classes – especially useful in situations where those case classes cannot be altered.

For instance, if `color` is also an attribute of `trait Car` and its child case classes, adding a new color setting method will be a trivial exercise by simply adding a `setColor()` method signature in `trait Settable` and its specific method implementations as well as `setNewColor()` within class `CarOps`.

Orthogonal type collection

Let’s see what a collection of cars looks like:

val sedan1 = Sedan("1ABC*234", "Honda Accord", 20500)
val sedan2 = Sedan("2DEF*345", "BMW 330i", 38000)
val sports1 = Sports("5MNO*678", "Ford Mustang", 34000)

val cars = List(sedan1, sedan2, sports1)
// cars: List[Product with java.io.Serializable] = List(
//   Sedan("1ABC*234", "Honda Accord", 20500.0),
//   Sedan("2DEF*345", "BMW 330i", 38000.0),
//   Sports("5MNO*678", "Ford Mustang", 34000.0)
// )

To refine the inferred `List[Product with java.io.Serializable]` collection type, we could provide some type hints as shown below:

// Existential type hints
import scala.language.existentials

val cars = List[(T, Settable[T]) forSome { type T }](
    (sedan1, implicitly[Settable[Sedan]]),
    (sedan2, implicitly[Settable[Sedan]]),
    (sports1, implicitly[Settable[Sports]])
  )

Ad-hoc Polymorphism In Scala

Over the past few years, there seems to be a subtle trend of software engineers favoring typeclass patterns that implement polymorphism in an ad-hoc fashion, namely, Ad-hoc Polymorphism. To see the benefits of such kind of polymorphism, let’s first look at what F-bounded polymorphism, a subtype polymorphism, has to offer.

F-bounded polymorphism

// F-bounded polymorphism
trait Car[T <: Car[T]] { self: T =>
  def vin: String
  def model: String
  def price: Double
  def setPrice(newPrice: Double): T
}

Next, a couple of child classes are defined:

// F-bounded polymorphism continued
case class Sedan(vin: String, model: String, price: Double) extends Car[Sedan] {
  def setPrice(newPrice: Double) = copy(price = newPrice)
}

case class Sports(vin: String, model: String, price: Double) extends Car[Sports] {
  def setPrice(newPrice: Double) = copy(price = newPrice)
}

A F-bounded type has a peculiar signature of the self-recursive `A[T <: A[T]]` which mandates the given type `T` itself a sub-type of `A[T]`, like how type `Sedan` is defined (Sedan <: Car[Sedan]). Note that the self-type annotation used in the trait isn’t requirement for F-bounded type. Rather, it’s a common practice for safeguarding against undesirable mix-up of sub-classes like below:

// F-bounded polymorphism with self-type
case class Sports(vin: String, model: String, price: Double) extends Car[Sedan] {
  def setPrice(newPrice: Double) = Sedan(vin, model, newPrice)
}
// ERROR: illegal inheritance; self-type Sports does not conform to Car[Sedan]'s selftype Sedan ...

“Type argument” versus “Type member”

Rather than a `type argument`, a F-bounded type could also be expressed as a `type member` which needs to be defined in its child classes.:

// F-bounded polymorphism expressed as type member
trait Car {
  type T <: Car
  def vin: String
  def model: String
  def price: Double
  def setPrice(newPrice: Double): T
}

case class Sedan(...) extends Car {
  type T = Sedan
  ...
}

case class Sports(...) extends Car {
  type T = Sports
  ...
}

It should be noted that with the `type member` approach, self-type would not be applicable, hence mix-up of sub-classes mentioned above is possible.

Let’s define a sedan and test out method `setPrice`:

// F-bounded polymorphism example continued
val sedan1 = Sedan("1ABC*234", "Honda Accord", 20500)

sedan1.setPrice(19500)
// res1: Sedan = Sedan("1ABC*234", "Honda Accord", 19500.0)

Under the F-bounded type’s “contract”, a method such as the following would work as intended to return the specified sub-type:

Had the Car/Sedan hierarchy been set up as the less specific `T <: Car`, the corresponding method:

// Problem with simple subtype
def setSalePrice[T <: Car](car: T, discount: Double): T =
  car.setPrice(car.price * (1.0 - discount max 0.0))

would fail as it couldn’t guarantee the returning type is the exact type of the input.

F-bounded type collection

Next, let’s look at a collection of cars.

// Collection of F-bounded elements
val sedan2 = Sedan("2DEF*345", "BMW 330i", 38000)
val sports1 = Sports("5MNO*678", "Ford Mustang", 34000)

val cars = List(sedan1, sedan2, sports1)
// cars: List[Product with Serializable with Car[_ >: Sports with SUV with Sedan <: Product with Serializable with Car[_ >: Sports with SUV with Sedan <: Product with Serializable]]] = ...

The resulting type is a rather ugly sequence of gibberish. To help the compiler a little, give it some hints about `T <: Car[T]` as shown below:

// Existential type hints
import scala.language.existentials

val cars = List[T forSome { type T <: Car[T] }](sedan1, sedan2, sports1)
// cars: List[T forSome { type T <: Car[T] }] = List(...)

Ad-hoc polymorphism

Contrary to subtype polymorphism which orients around a supertype with a rigid subtype structure, let’s explore a different approach using typeclasses, known as Ad-hoc polymorphism.

// Ad-hoc polymorphism example
trait Car[T] {
  def vin(car: T): String
  def model(car: T): String
  def price(car: T): Double
  def setPrice(car: T, newPrice: Double): T
}

case class Sedan(vin: String, model: String, price: Double)
case class Sports(vin: String, model: String, price: Double)

Next, a couple of “ad-hoc” implicit objects are created to implement the trait methods.

// Ad-hoc polymorphism example continued
import scala.language.implicitConversions

implicit object SedanCar extends Car[Sedan] {
  def vin(car: Sedan) = car.vin
  def model(car: Sedan) = car.model
  def price(car: Sedan) = car.price
  def setPrice(car: Sedan, newPrice: Double): Sedan = car.copy(price = newPrice)
}

implicit object SportsCar extends Car[Sports] {
  def vin(car: Sports) = car.vin
  def model(car: Sports) = car.model
  def price(car: Sports) = car.price
  def setPrice(car: Sports, newPrice: Double): Sports = car.copy(price = newPrice)
}

Note that alternatively, the implicit objects could be set up as ordinary companion objects of the case classes with implicit anonymous classes:

// Alternative - defining implicits in companion objects
case class Sedan(vin: String, model: String, price: Double)
object Sedan {
  implicit val SedanCar = new Car[Sedan] {
    def vin(car: Sedan) = car.vin
    def model(car: Sedan) = car.model
    def price(car: Sedan) = car.price
    def setPrice(car: Sedan, newPrice: Double): Sedan = car.copy(price = newPrice)
  }
}

case class Sports(vin: String, model: String, price: Double)
object Sports {
  implicit val SportsCar = new Car[Sports] {
    def vin(car: Sports) = car.vin
    def model(car: Sports) = car.model
    def price(car: Sports) = car.price
    def setPrice(car: Sports, newPrice: Double): Sports = car.copy(price = newPrice)
  }
}

Unifying implemented methods

Finally, an implicit conversion for cars of type `T` is provided by means of an implicit class to create a “unified” method that takes the corresponding method implementations from the provided implicit `Car[T]` parameter.

// Ad-hoc polymorphism example continued
implicit class CarOps[T](car: T)(implicit ev: Car[T]) {
  def setPrice(newPrice: Double): T = ev.setPrice(car: T, newPrice: Double)
}

Testing it out:

// Ad-hoc polymorphism example continued
val sedan1 = Sedan("1ABC*234", "Honda Accord", 20500)

sedan1.setPrice(19500)
// res1: Sedan = Sedan("1ABC*234", "Honda Accord", 19500.0)

New methods, like `setSalePrice`, can be added as needed in the implicit objects:

// Ad-hoc polymorphism example continued
implicit object SedanCar extends Car[Sedan] {
  ...
  def setSalePrice(car: Sedan, discount: Double): Sedan =
    car.setPrice(car.price * (1.0 - discount max 0.0))
}

implicit object SportsCar extends Car[Sports] {
    ...
  def setSalePrice(car: Sports, discount: Double): Sports =
    car.setPrice(car.price * (1.0 - discount max 0.0))
}

Ad-hoc type collection

Next, a collection of cars:

// Collection of elements of ad-hoc type
val sedan2 = Sedan("2DEF*345", "BMW 330i", 38000)
val sports1 = Sports("5MNO*678", "Ford Mustang", 34000)

val cars = List(sedan1, sedan2, sports1)
// cars: List[Product with java.io.Serializable] = List(
//   Sedan("1ABC*234", "Honda Accord", 20500.0),
//   Sedan("2DEF*345", "BMW 330i", 38000.0),
//   Sports("5MNO*678", "Ford Mustang", 34000.0)
// )

Similar to the F-bounded collection, the inferred resulting type isn’t very helpful. Unlike in the F-bounded case, we do not have a `T <: Car[T]` contract. Using an approach illustrated in this blog post, we could assemble the collection as a list of `(car, type)` tuples:

// Existential type hints
import scala.language.existentials

val cars = List[(T, Car[T]) forSome { type T }](
  (sedan1, implicitly[Car[Sedan]]),
  (sedan2, implicitly[Car[Sedan]]), 
  (sports1, implicitly[Car[Sports]])
)
// cars: List[(T, Car[T] forSome { type T })] = List(...)

By means of a simple example, we’ve now got a sense of how Ad-hoc polymorphism works. The F-bounded example serves as a contrasting reference of how the polymorphism bound by a more “strict” contract plays out in comparison. Given the flexibility of not having to bind the base classes into a stringent subtype relationship upfront, the rising popularity of Ad-hoc polymorphism certainly has its merits.

That said, lots of class models in real-world applications still fits perfectly well into a subtype relationship. In suitable use cases, F-bounded polymorphism generally imposes less boilerplate code. In addition, Ad-hoc polymorphism typically involves using of implicits that may impact code maintainability.