Reputation: 2107
In the sample code below, why does the Iterable[String] test1
produce a Set after mapping?
val foo = Map("a" -> 1, "b" -> 1)
val test1: Iterable[String] = foo.keys
val test2: Iterator[String] = foo.keys.toIterator
println(test1.map(foo).size) // 1
println(test2.map(foo).size) // 2
I was puzzled by this because its entirely counter-intuitive when reading the code. Even though foo.keys
just returns an Iterable, it creates a Set when calling map
, as the reflection code shows:
println(test1.map(foo).getClass.getName) // immutable.Set.Set1
println(test2.map(foo).getClass.getName) // Iterator$$anon$11
How does the standard library determine that it should create an immutable.Set
here, even though the inferred type of the collection is just Iterable[String]
?
Upvotes: 4
Views: 621
Reputation: 39577
Excavating Kolmar's comment that, although an implicit argument is in play that determines how the result collection is built, in this case the source collection is simply queried for the builder to use.
Iterable.map
:
def map[B, That](f: (A) ⇒ B)(implicit bf: CanBuildFrom[Iterable[A], B, That]): That
Implicit scope includes types related to the type args, including Iterable
and Int
.
Iterable
defines a "generic" CanBuildFrom
that invokes genericBuilder
on the source collection. That's how the result type is tied to the source.
Conversely, the result collection is divorced from the source by taking a CanBuildFrom[From = Nothing, _, _]
. This is how cc.to[Set]
is expressed, where a Set
is built without regard for the source collection cc
. For operations such as map
, the method collection.breakOut
provides such a CanBuildFrom
, where the result type can be usefully inferred.
You can inject an arbitrary CanBuildFrom
for the desired behavior:
$ scala
Welcome to Scala 2.11.8 (Java HotSpot(TM) 64-Bit Server VM, Java 1.8.0_92).
Type in expressions for evaluation. Or try :help.
scala> val m = Map("a" -> 1, "b" -> 1)
m: scala.collection.immutable.Map[String,Int] = Map(a -> 1, b -> 1)
scala> val k = m.keys
k: Iterable[String] = Set(a, b)
scala> import collection.{generic, mutable}, generic.{CanBuildFrom => CBF}, mutable.ListBuffer
import collection.{generic, mutable}
import generic.{CanBuildFrom=>CBF}
import mutable.ListBuffer
scala> implicit def `as list`: CBF[Iterable[_], Int, List[Int]] =
| new CBF[Iterable[_], Int, List[Int]] {
| def apply() = new ListBuffer[Int]
| def apply(from: Iterable[_]) = apply()
| }
as$u0020list: scala.collection.generic.CanBuildFrom[Iterable[_],Int,List[Int]]
scala> k.map(m)
res0: List[Int] = List(1, 1)
Worth adding that completion can show types as of 2.11.8:
scala> k.map(m) //print<tab>
$line4.$read.$iw.$iw.k.map[Int, Iterable[Int]]($line3.$read.$iw.$iw.m)(scala.collection.Iterable.canBuildFrom[Int]) // : Iterable[Int]
Using breakOut
:
scala> k.map(m)(collection.breakOut)
res1: scala.collection.immutable.IndexedSeq[Int] = Vector(1, 1)
scala> k.map(m)(collection.breakOut) //print
$line4.$read.$iw.$iw.k.map[Int, scala.collection.immutable.IndexedSeq[Int]]($line3.$read.$iw.$iw.m)(scala.collection.`package`.breakOut[Any, Int, scala.collection.immutable.IndexedSeq[Int]](scala.Predef.fallbackStringCanBuildFrom[Int])) // : scala.collection.immutable.IndexedSeq[Int]
As shown, it actually picks up the CanBuildFrom
intended for operations such as:
scala> "abc".map(_ + 1)
res2: scala.collection.immutable.IndexedSeq[Int] = Vector(98, 99, 100)
scala> "abc".map(_ + 1) //print
scala.Predef.augmentString("abc").map[Int, scala.collection.immutable.IndexedSeq[Int]](((x$1: Char) => x$1.+(1)))(scala.Predef.fallbackStringCanBuildFrom[Int]) // : scala.collection.immutable.IndexedSeq[Int]
Compare:
scala> k.map(m)(collection.breakOut) : List[Int] //print
(($line6.$read.$iw.$iw.k.map[Int, List[Int]]($line5.$read.$iw.$iw.m)(scala.collection.`package`.breakOut[Iterable[String], Int, List[Int]](scala.collection.immutable.List.canBuildFrom[Int]))): scala.`package`.List[scala.Int]) // : List[Int]
The canonical Q&A on breakOut.
Upvotes: 2
Reputation: 3482
This is tricky implicit magic. Simplified answer: there exists CanBuildFrom
value, that is passed in implicit scope. When compiler is searching for most common type, it looks for implicits in arguments' scope.
In your example, compiler is able to figure out that most common type for foo.keys
is Set
. This sounds reasonable: Set can be viewed as Map with absent values (also java's HashMap/HashSet do that). When you convert to iterable, implicits are lost, and Set
is gone (as side note, those CanBuildFrom
hacks are not robust, and might get gone in future, because they really complicate extending of existing collections, you might also want to read this answer and comments).
Scala shares Java's concept of "de-jure and de-facto types". "de-jure" is one declared in method definition, but "de-facto" might be one of inheritors. That's why, for example, you see Map.keys
type as Iterable
, when de-facto it is Set
(produced from Map.keySet
, which has Set
de-jure type).
Last, 1
in first println is because in underlying map foo
all values are the same, and Set(1,1)
becomes Set(1)
.
Upvotes: 0
Reputation: 42047
foo.keys
returns a Set
(despite its return type being more general) and calling map on a Set
produces another Set
. The inferred or compile time type is not always the most precise.
You can see that the keys
method on a Set
returns a Set
even though the return type is Iterable[A]
:
scala> Map(1 -> 2).keys
res0: Iterable[Int] = Set(1)
Upvotes: 5