Difference between mapValues and transform in Map

Question

In Scala Map (see API) what is the difference in semantics and performance between mapValues and transform ?

For any given map

Answer 1

collection.Map doesn't provide transform: it has a different signature for mutable and immutable Maps.

$ scala
Welcome to Scala version 2.11.2 (Java HotSpot(TM) 64-Bit Server VM, Java 1.8.0_11).
Type in expressions to have them evaluated.
Type :help for more information.

scala> val im = Map('a -> 1, 'b -> 2, 'c -> 3)
im: scala.collection.immutable.Map[Symbol,Int] = Map('a -> 1, 'b -> 2, 'c -> 3)

scala> im.mapValues(_ * 7) eq im
res0: Boolean = false

scala> im.transform { case (k,v) => v*7 } eq im
res2: Boolean = false

scala> val mm = collection.mutable.Map('a -> 1, 'b -> 2, 'c -> 3)
mm: scala.collection.mutable.Map[Symbol,Int] = Map('b -> 2, 'a -> 1, 'c -> 3)

scala> mm.mapValues(_ * 7) eq mm
res3: Boolean = false

scala> mm.transform { case (k,v) => v*7 } eq mm
res5: Boolean = true

Mutable transform mutates in place:

scala> mm.transform { case (k,v) => v*7 }
res6: mm.type = Map('b -> 98, 'a -> 49, 'c -> 147)

scala> mm.transform { case (k,v) => v*7 }
res7: mm.type = Map('b -> 686, 'a -> 343, 'c -> 1029)

So mutable transform doesn't change the type of the map:

scala> im mapValues (_ => "hi")
res12: scala.collection.immutable.Map[Symbol,String] = Map('a -> hi, 'b -> hi, 'c -> hi)

scala> mm mapValues (_ => "hi")
res13: scala.collection.Map[Symbol,String] = Map('b -> hi, 'a -> hi, 'c -> hi)

scala> mm.transform { case (k,v) => "hi" }
<console>:9: error: type mismatch;
 found   : String("hi")
 required: Int
              mm.transform { case (k,v) => "hi" }
                                           ^

scala> im.transform { case (k,v) => "hi" }
res15: scala.collection.immutable.Map[Symbol,String] = Map('a -> hi, 'b -> hi, 'c -> hi)

...as can happen when constructing a new map.

Answer 2

Here's a couple of unmentioned differences:

• mapValues creates a Map that is NOT serializable, without any indication that it's just a view (the type is Map[_, _], but just try to send one across the wire).

• Since mapValues is just a view, every instance contains the real Map - which could be another result of mapValues. Imagine you have an actor with some state, and every mutation of the state sets the new state to be a mapValues on the previous state...in the end you have deeply nested maps with a copy of each previous state of the actor (and, yes, both of these are from experience).

Answer 3

Let's say we have a Map[A,B]. For clarification: I'm always referring to an immutable Map.

mapValues takes a function B => C, where C is the new type for the values.

transform takes a function (A, B) => C, where this C is also the type for the values.

So both will result in a Map[A,C].

However with the transform function you can influence the result of the new values by the value of their keys.

For example:

val m = Map( "a" -> 2, "b" -> 3 )
m.transform((key, value) => key + value) //Map[String, String](a -> a2, b -> b3)

Doing this with mapValues will be quite hard.

The next difference is that transform is strict, whereas mapValues will give you only a view, which will not store the updated elements. It looks like this:

protected class MappedValues[C](f: B => C) extends AbstractMap[A, C] with DefaultMap[A, C] {
  override def foreach[D](g: ((A, C)) => D): Unit = for ((k, v) <- self) g((k, f(v)))
  def iterator = for ((k, v) <- self.iterator) yield (k, f(v))
  override def size = self.size
  override def contains(key: A) = self.contains(key)
  def get(key: A) = self.get(key).map(f)
}

(taken from https://github.com/scala/scala/blob/v2.11.2/src/library/scala/collection/MapLike.scala#L244)

So performance-wise it depends what is more effective. If f is expensive and you only access a few elements of the resulting map, mapValues might be better, since f is only applied on demand. Otherwise I would stick to map or transform.

transform can also be expressed with map. Assume m: Map[A,B] and f: (A,B) => C, then

m.transform(f) is equivalent to m.map{case (a, b) => (a, f(a, b))}