You are viewing the documentation for the 2.4.x release series. The latest stable release series is 2.5.3.

§Handling data streams reactively

Progressive Stream Processing and manipulation is an important task in modern Web Programming, starting from chunked upload/download to Live Data Streams consumption, creation, composition and publishing through different technologies including Comet and WebSockets.

Iteratees provide a paradigm and an API allowing this manipulation, while focusing on several important aspects:


An Iteratee is a consumer - it describes the way input will be consumed to produce some value. An Iteratee is a consumer that returns a value it computes after being fed enough input.

// an iteratee that consumes String chunks and produces an Int

The Iteratee interface Iteratee[E,A] takes two type parameters: E, representing the type of the Input it accepts, and A, the type of the calculated result.

An iteratee has one of three states: Cont meaning accepting more input, Error to indicate an error state, and Done which carries the calculated result. These three states are defined by the fold method of an Iteratee[E,A] interface:

def fold[B](folder: Step[E, A] => Future[B]): Future[B]

where the Step object has 3 states :

object Step {
  case class Done[+A, E](a: A, remaining: Input[E]) extends Step[E, A]
  case class Cont[E, +A](k: Input[E] => Iteratee[E, A]) extends Step[E, A]
  case class Error[E](msg: String, input: Input[E]) extends Step[E, Nothing]

The fold method defines an iteratee as one of the three mentioned states. It accepts three callback functions and will call the appropriate one depending on its state to eventually extract a required value. When calling fold on an iteratee you are basically saying:

Depending on the state of the iteratee, fold will produce the appropriate B using the corresponding passed-in function.

To sum up, an iteratee consists of 3 states, and fold provides the means to do something useful with the state of the iteratee.

§Some important types in the Iteratee definition:

Before providing some concrete examples of iteratees, let’s clarify two important types we mentioned above:

§Some primitive iteratees:

By implementing the iteratee, and more specifically its fold method, we can now create some primitive iteratees that we can use later on.

val doneIteratee = new Iteratee[String,Int] {
  def fold[B](folder: Step[String,Int] => Future[B])(implicit ec: ExecutionContext) : Future[B] = 
    folder(Step.Done(1, Input.Empty))

As shown above, this is easily done by calling the appropriate apply function, in our case that of Done, with the necessary information.

To use this iteratee we will make use of the Future that holds a promised value.

def folder(step: Step[String,Int]):Future[Option[Int]] = step match {
  case Step.Done(a, e) => future(Some(a))
  case Step.Cont(k) => future(None)
  case Step.Error(msg,e) => future(None)

val eventuallyMaybeResult: Future[Option[Int]] = doneIteratee.fold(folder)

eventuallyMaybeResult.onComplete(i => println(i))

of course to see what is inside the Future when it is redeemed we use onComplete

// will eventually print 1
eventuallyMaybeResult.onComplete(i => println(i))

There is already a built-in way allowing us to create an iteratee in the Done state by providing a result and input, generalizing what is implemented above:

val doneIteratee = Done[String,Int](1, Input.Empty)

Creating a Done iteratee is simple, and sometimes useful, but it does not consume any input. Let’s create an iteratee that consumes one chunk and eventually returns it as the computed result:

val consumeOneInputAndEventuallyReturnIt = new Iteratee[String,Int] {
def fold[B](folder: Step[String,Int] => Future[B])(implicit ec: ExecutionContext): Future[B] = {
     folder(Step.Cont {
       case Input.EOF => Done(0, Input.EOF) //Assuming 0 for default value
       case Input.Empty => this
       case Input.El(e) => Done(e.toInt,Input.EOF) 

def folder(step: Step[String,Int]):Future[Int] = step match {
  case Step.Done(a, _) => future(a)
  case Step.Cont(k) => k(Input.EOF).fold({
    case Step.Done(a1, _) => Future.successful(a1)
    case _ => throw new Exception("Erroneous or diverging iteratee")
  case _ => throw new Exception("Erroneous iteratee")

As for Done, there is a built-in way to define an iteratee in the Cont state by providing a function that takes Input[E] and returns a state of Iteratee[E,A] :

val consumeOneInputAndEventuallyReturnIt = {
  Cont[String,Int](in => Done(100,Input.Empty))

In the same manner there is a built-in way to create an iteratee in the Error state by providing an error message and an Input[E]

Back to the consumeOneInputAndEventuallyReturnIt, it is possible to create a two-step simple iteratee manually, but it becomes harder and cumbersome to create any real-world iteratee capable of consuming a lot of chunks before, possibly conditionally, it eventually returns a result. Luckily there are some built-in methods to create common iteratee shapes in the Iteratee object.

§Folding input:

One common task when using iteratees is maintaining some state and altering it each time input is pushed. This type of iteratee can be easily created using the Iteratee.fold which has the signature:

def fold[E, A](state: A)(f: (A, E) => A): Iteratee[E, A]

Reading the signature one can realize that this fold takes an initial state A, a function that takes the state and an input chunk (A, E) => A and returns an Iteratee[E,A] capable of consuming Es and eventually returning an A. The created iteratee will return Done with the computed A when an input EOF is pushed.

One example would be creating an iteratee that counts the number of bytes pushed in:

val inputLength: Iteratee[Array[Byte],Int] = {
  Iteratee.fold[Array[Byte],Int](0) { (length, bytes) => length + bytes.size }

Another would be consuming all input and eventually returning it:

val consume: Iteratee[String,String] = {
  Iteratee.fold[String,String]("") { (result, chunk) => result ++ chunk }

There is actually already a method in the Iteratee object that does exactly this for any scala TraversableLike, called consume, so our example becomes:

val consume = Iteratee.consume[String]()

One common case is to create an iteratee that does some imperative operation for each chunk of input:

val printlnIteratee = Iteratee.foreach[String](s => println(s))

More interesting methods exist like repeat, ignore, and fold1 - which is different from the preceding fold in that it gives one the opportunity to treat input chunks asynchronously.

Of course one should be worried now about how hard it would be to manually push input into an iteratee by folding over iteratee states over and over again. Indeed each time one has to push input into an iteratee, one has to use the fold function to check on its state, if it is a Cont then push the input and get the new state, or otherwise return the computed result. That’s when Enumerators come in handy.

Next: Enumerators

Dokümantasyonun bu çevirisi Play ekibi tarafından yapılmamaktadır. Eğer bir hata bulduysanız, bu sayfanın kaynak kodu burada bulunmaktadır. Dokümantasyon yönergelerini okuduktan sonra lütfen katkı yapmaktan çekinmeyin.