CODEWITHSUNDEEP
scalagenericstypestype-systemstype-bounds
Yann Moisan
Yann Moisan

Reputation: 8281

How to avoid duplication of type bound in Scala

I have a class with a constrained type parameter.

I've tried with identity but the return type is not precise.

And in the method identityTP, I need to specify again the constraint

How to avoid to duplicate this constraint with methods that work on this type ?

Here is an example :

  sealed trait Location
  case object Single extends Location
  case object Multi extends Location

  final case class Log[L <: Location](l: L)

  def identity(log: Log[_]): Log[_] = log
  def identityTP[L<: Location](log: Log[L]): Log[L] = log

Upvotes: 2

Views: 199

Answers (1)

Dmytro Mitin
Dmytro Mitin

Reputation: 51703

Actually this is not a duplication. L in

final case class Log[L <: Location](l: L)

and L in

def identityTP[L <: Location](log: Log[L]): Log[L] = log

are two completely different type parameters. This could be more clear if you used different identifiers

final case class Log[L <: Location](l: L)
def identityTP[L1 <: Location](log: Log[L1]): Log[L1] = log

An upper bound for one type parameter doesn't duplicate an upper bound for other type parameter.

Also this is not a duplication because actually upper bounds can be different

sealed trait SubLocation extends Location
final case class Log[L <: Location](l: L)
def identityTP[L1 <: SubLocation](log: Log[L1]): Log[L1] = log

If you don't want to create the second type parameter you can make identityTP nested into Log (making it Log's method)

final case class Log[L <: Location](l: L) {
  def identityTP: Log[L] = this
}

Sometimes it can help if you make L a type member rather than type parameter

trait Log {
  type L <: Location
  val l: L
}
object Log {
//  def apply[_L <: Location](_l: _L): Log { type L = _L} = new Log {
//    override type L = _L
//    override val l: L = _l
//  }

  def apply[_L <: Location](_l: _L): Log = new Log {
    override type L = _L
    override val l: L = _l
  }
}

//  def identityTP(log: Log): Log = log
def identityTP(log: Log): Log { type L = log.L } = log

Notice that although we have to repeat upper bound in apply but we don't have in identityTP.

Normally it's not a big deal to repeat upper bound when necessary

class MyClass[A <: A1]
def foo[A <: A1](mc: MyClass[A]) = ???
def bar[A <: A1](mc: MyClass[A]) = ???

When this becomes cumbersome

class MyClass[A <: A1, B <: B1, C <: C1]
def foo[A <: A1, B <: B1, C <: C1](mc: MyClass[A, B, C]) = ???
def bar[A <: A1, B <: B1, C <: C1](mc: MyClass[A, B, C]) = ???

you should redesign your abstractions. For example

trait Tuple {
  type A <: A1
  type B <: B1
  type C <: C1
}
class MyClass[T <: Tuple]
def foo[T <: Tuple](mc: MyClass[T]) = {
  //T#A, T#B, T#C instead of A, B, C
  ???
}

or

class MyClass[T <: Tuple](val t: T)
//class MyClass(val t: Tuple)

def foo[T <: Tuple](mc: MyClass[T]) = {
//def foo(mc: MyClass) = {
  import mc.t
  //t.A, t.B, t.C instead of A, B, C
  ???
}

Also sometimes you can play with replacing type bounds with type constraints

final case class Log[L](l: L)(implicit ev: L <:< Location)
def identityTP[L](log: Log[L])(implicit ev: L <:< Location): Log[L] = log

Although this doesn't remove repetitions but there are ways to fight against repetitions among implicit parameters as well (type classes). See How to wrap a method having implicits with another method in Scala?

Upvotes: 4

Related Questions