Skip to content
This repository was archived by the owner on Dec 22, 2021. It is now read-only.

FAQ

Jump to bottom
Julien Richard-Foy edited this page Apr 25, 2018 · 90 revisions

This page should evolve over time, but already be useful now. When more official documentation is available, content can be replaced by links.

Collection Users

What are the breaking changes?

The following table summarizes the main changes:

Description Old Code New Code Automatic Migration Rule
Method to[C[_]] has been removed (it might be reintroduced but deprecated, though) xs.to[List] xs.to(List) NewCollections
SortedSet: the to, until and from methods are now called rangeTo, rangeUntil and rangeFrom, respectively xs.until(42) xs.rangeUntil(42)
immutable.Set/Map: the + operation no longer has an overload accepting multiple values Set(1) + (2, 3) Set(1) + 2 + 3
mutable.Set/Map no longer have a + operation mutable.Set(1) + 2 mutable.Set(1).clone() += 2
mutable.Set/Map no longer have an updated method mutable.Map(1 -> 2).updated(1, 3) mutable.Map(1 -> 2).clone() += 1 -> 3
collection.Set/Map no longer have + and - operations xs + 1 - 2 xs ++ Set(1) -- Set(2)
collection.Map no longer have -- operation
mapValues and filterKeys now return a MapView instead of a Map kvs.mapValues(f) kvs.mapValues(f).toMap
Traversable and TraversableOnce are replaced with Iterable and IterableOnce, respectively def f(xs: Traversable[Int]): Unit def f(xs: Iterable[Int]): Unit NewCollections
Stream is replaced with LazyList Stream.from(1) LazyList.from(1) NewCollections
Seq#union is replaced with concat xs.union(ys) xs.concat(ys)
Stream#append is replaced with lazyAppendAll xs.append(ys) xs.lazyAppendAll(ys) NewCollections
IterableOnce#toIterator is replaced with IterableOnce#iterator() xs.toIterator xs.iterator() NewCollections
copyToBuffer has been deprecated xs.copyToBuffer(buffer) buffer ++= xs NewCollections
Iterable no longer has a sameElements operation xs1.sameElements(xs2) xs1.iterator().sameElements(xs2)
TupleNZipped has been replaced with LazyZipN (xs, ys).zipped xs.lazyZip(ys) NewCollections
collection.breakOut no longer exists val xs: List[Int] = ys.map(f)(collection.breakOut) val xs = ys.view.map(f).to(List)
LinearSeq no longer has a companion-apply LinearSeq(1, 2, 3) List(1, 2, 3)
MapProxy, SetProxy, SynchronizedX no longer exist
SeqForwarder, MutableList, StreamView no longer exist

The “Automatic Migration Rule” column gives the name of the migration rule that can be used to automatically update old code to the new expected form. See https://github.com/scala/scala-collection-compat/ for more details on how to use it.

Other notable changes are:

  • Iterable.partition invokes iterator twice on non-strict collections and assumes it gets two iterators over the same elements. Strict subclasses override partition do perform only a single traversal

  • scala.Seq[+A] is now scala.collection.immutable.Seq[A] (this also affects varargs methods).

  • The new collections makes more use of overloading. You can find more information about the motivation behind this choice here. For instance, Map.map is overloaded:

    scala> Map(1 -> "a").map
  def map[B](f: ((Int, String)) => B): scala.collection.immutable.Iterable[B]
  def map[K2, V2](f: ((Int, String)) => (K2, V2)): scala.collection.immutable.Map[K2,V2]

    Type inference has been improved so that Map(1 -> "a").map(x => (x._1 + 1, x._2)) works, the compiler can infer the parameter type for the function literal. However, using a method reference in 2.13.0-M4 (improvement are on the way for 2.13.0) does not work, and an explicit eta-expansion is necessary:

    scala> def f(t: (Int, String)) = (t._1 + 1, t._2)
scala> Map(1 -> "a").map(f)
                        ^
      error: missing argument list for method f
      Unapplied methods are only converted to functions when a function type is expected.
      You can make this conversion explicit by writing `f _` or `f(_)` instead of `f`.
scala> Map(1 -> "a").map(f _)
res10: scala.collection.immutable.Map[Int,String] = ChampHashMap(2 -> a)

  • Views have been completely redesigned and we expect their usage to have a more predictable evaluation model. You can read more about the new design here.

Are there new collection types?

scala.collection.immutable.ImmutableArray is an immutable sequence backed by an array. It is used to pass varargs parameters.

The scala-collection-contrib module provides decorators enriching the collections with new operations. You can think of this artifact as an incubator: if we get evidence that these operations should be part of the core, we might eventually move them.

The following collections are provided:

  • MultiSet (both mutable and immutable)
  • SortedMultiSet (both mutable and immutable)
  • MultiDict (both mutable and immutable)
  • SortedMultiDict (both mutable and immutable)

Are there new operations on collections?

Partitionning operations

The following new operations are available:

def groupMap[K, B](key: A => K)(f: A => B): Map[K, CC[B]] // (Where `CC` can be `List`, for instance)
def groupMapReduce[K, B](key: A => K)(f: A => B)(g: (B, B) => B): Map[K, B]

groupMap is equivalent to groupBy(key).mapValues(_.map(f)). groupMapReduce is equivalent to groupBy(key).mapValues(_.map(f).reduce(g)).

Other operations

Additional operations are provided by the scala-collection-contrib module. You can think of this artifact as an incubator: if we get evidence that these operations should be part of the core, we might eventually move them.

The new operations are provided via an implicit enrichment. You need to add the following import to make them available:

import strawman.collection.decorators._

The following operations are provided:

  • Seq
    • intersperse
  • Map
    • zipByKey / join / zipByKeyWith
    • mergeByKey / fullOuterJoin / mergeByKeyWith / leftOuterJoin / rightOuterJoin

Are there new implementations of existing collection types (changes in performance characteristics)?

The default Set and Map are backed by a ChampHashSet and a ChampHashMap, respectively. The performance characteristics are the same but the operation implementations are faster. These data structures also have a lower memory footprint.

mutable.Queue and mutable.Stack now use mutable.ArrayDeque. This data structure supports constant time index access, and amortized constant time insert and remove operations.

How do I cross-build my project against Scala 2.12 and Scala 2.13?

We will eventually provide a CrossCompat migration rule (see scala/scala-collection-compat#19) that will rewrite old code into code that cross-compiles with 2.13 and with older Scala versions.

That being said, we expect most usages to be compatible with 2.13 (at the cost of some warnings, sometimes).

If your code does not compile with 2.13, you can migrate it and use the scala-collection-compat module to make the new code compile with older Scala versions.

Finally, if your code uses features that are not supported by the scala-collection-compat library (typically, if you implement a custom collection type), you will have to use different source directories per Scala version:

// Adds a `src/main/scala-2.13+` source directory for Scala 2.13 and newer
// and a `src/main/scala-2.13-` source directory for Scala version older than 2.13
unmanagedSourceDirectories in Compile += {
  val sourceDir = (sourceDirectory in Compile).value
  CrossVersion.partialVersion(scalaVersion.value) match {
    case Some((2, n)) if n >= 13 => sourceDir / "scala-2.13+"
    case _                       => sourceDir / "scala-2.13-"
  }
}

Examples

New deprecations

  • ArraySeq is deprecated. 2.12 had both WrappedArray and ArraySeq, where WrappedArray has specialized subclasses for primitive arrays. ArraySeq is backed by an object array, primitives are boxed. In 2.13, WrappedArray can be used either way for primitive arrays (TODO: document how).

Collection Implementers

CanBuildFrom no longer exists - what were its uses and how can they be replaced?

CanBuildFrom was used to:

  • generically implement transformation operations whose return type could vary according to the type of collection elements (ie mapping a Char to a Char in a String returns a String, but mapping a Char to an Int returns a IndexedSeq[Int]),
  • abstract over the arity of collection type constructors (ie List[_] vs Map[_, _]),
  • abstract over implicit parameters required to perform a transformation operations (ie mapping an A to a B in SortedSet[A] requires an implicit Ordering[B] to return a SortedSet[B]),
  • provide type-driven builders to implement generic transformation methods (eg Future.traverse).

The first three points are now handled using overloading. This means that Map[K, V] has two overloads of map, one that takes a (K, V) => (K', V') mapping function and returns a Map[K', V'], and one that takes a (K, V) => A mapping function and returns an Iterable[A].

You can find more information about the design here.

To address the last point, there is a BuildFrom typeclass that works exactly like the former CanBuildFrom. See the next section for examples of use.

How can I write generic extension methods?

Depending on the level of desired genericity, several solutions can apply, with different complexity.

Write an operation that can consume any collection

Take an IterableOnce[A] as parameter, or an Iterable[A] if you need more than one traversals:

implicit class SumByOperation[A](coll: IterableOnce[A]) {
  def sumBy[B](f: A => B)(implicit num: Numeric[B]): B = {
    val it = coll.iterator()
    var result = f(it.next())
    while (it.hasNext()) {
      result = num.plus(result, it.next())
    }
    result
  }
}

Write an operation that returns another collection

Use BuildFrom:

def optionSequence[CC[X] <: Iterable[X], A, To](xs: CC[Option[A]])(implicit bf: BuildFrom[CC[Option[A]], A, To]): Option[To] =
  xs.foldLeft[Option[Builder[A, To]]](Some(bf.newBuilder(xs))) {
    case (Some(builder), Some(a)) => Some(builder += a)
    case _ => None
  }.map(_.result())

The optionSequence operation can be used on any collection (but not on an Array or a String):

scala> optionSequence(List[Option[Int]](Some(1), Some(2), Some(3)))
res1: Option[List[Int]] = Some(List(1, 2, 3))

scala> optionSequence(Set[Option[(Int, String)]](Some(1 -> "foo"), Some(2 -> "bar")))(TreeMap) // Force the result type
res4: Option[scala.collection.immutable.TreeMap[Int,String]] = Some(TreeMap(1 -> foo, 2 -> bar))

Write an operation that can also be applied to a View, an Array or String

For more advanced cases, or your operation should also work with Arrays, Strings and Views, the pattern given in the previous section is not enough.

The scala-collection-contrib module provides a more advanced machinery that handles that:

class IntersperseOperation[C, S <: HasSeqOps[C]](coll: C)(implicit val seq: S) {
  def intersperse[B >: seq.A, That](sep: B)(implicit bf: BuildFrom[C, B, That]): That =
    bf.fromSpecificIterable(coll)(new View.Intersperse(seq(coll), sep)) // (Assume that there is a `View.Intersperse` implemented somewhere)
}

implicit def IntersperseOperation[C](coll: C)(implicit seq: HasSeqOps[C]): SeqDecorator[C, seq.type] =
    new IntersperseOperation(coll)(seq)

This pattern makes the intersperse operation available to any Seq like type (eg. a SeqView, an Array or a String).

How do I integrate my collection in the new design?

Read the draft document on the design of the Scala 2.13 collections: https://github.com/scalacenter/docs.scala-lang/blob/collection-strawman/_overviews/core/architecture-of-scala-collections.md

Which methods should I overload to support the “same result type” principle?

You want to add overloads to specialize a transformation operations such that they return a more specific result. Examples are:

  • map, on StringOps, when the mapping function returns a Char, should return a String (instead of an IndexedSeq),
  • map, on Map, when the mapping function returns a pair, should return a Map (instead of an Iterable),
  • map, on SortedSet, when an implicit Ordering is available for the resulting element type, should return a SortedSet (instead of a Set).

The following table lists transformation operations that might return a too wide type. You might want to overload these operations to return a more specific type.

Collection Operations
Iterable map, flatMap, collect, scanLeft, scanRight, groupMap, concat, zip, zipAll, unzip
Seq prepended, appended, prependedAll, appendedAll, padTo, patch
immutable.Seq updated
SortedSet map, flatMap, collect, zip
Map map, flatMap, collect, concat
immutable.Map updated, transform
SortedMap map, flatMap, collect, concat
immutable.SortedMap updated
Clone this wiki locally