My Programming Escapades
Andraz Bajt
Pretty function composition in scala and asynchronous function composition on android

Surjective composition: the first function nee… Surjective composition: the first function need not be surjective. (Photo credit: Wikipedia)

Function composition is a nice way to sequence transformations on data. For example in a compiler you take your source, parse, check, optimize and generate code(in a nutshell). It's a linear series of transformations -> perfect for function composition In Haskell you can use this beautiful syntax

compile = codegen . optimize .check . parse

leaving out the parameters and noting composition as “.” which kinda looks like ° used in math(if you squint a bit). In scala you could do something like

val compile = codegen compose optimize compose check compose parse

Nearly there, I just want it to look a bit prettier. (Compose is a method defined on Function1 trait) So I define an implicit conversion from function to “composable”

implicit def function2composable[A,B](f: A=>B) = new AnyRef{
  def -->[C](g: B=>C) = v => g(f(c))

This creates an object and reimplements “compose” but I really like the syntax:

compile = parse --> check --> optimize --> codegen

Functional programming can be imagined as a waterfall of data, flowing from one function into the next, and the --> operator represents this nicely. Let's go a step further. If the composition is one-off, and this is a waterfall of DATA it could be nice to represent that. Something like

val result = source --> parse --> check --> optimize --> codegen

So now I'm taking a value and sending it through black boxes. Very nice. Apart from the fact it doesn't work(yet!).

implicit def any2waterfall[A](a: A) = new AnyRef{
  def -->[B](f: A=>B) = f(a)

Scala‘s awesome compiler can handle two implicit conversions with same method names. Nice. You can even mix and match

val result = source --> (parse --> check --> optimize) --> codegen

This does the composition of parse, check and optimize into an anonymous function, applies it to the source and then applies codegen to it's result.

Goin async

What about asynchronous calls? Can I compose those too? I think it's possible with Lift actors(or with scalaz's?), but I needed to integrate that into Android‘s activities quite recently. Well I did not need to do it, but it was quite a nice solution.

The usual way of doing things async in Android is with the conveniently named AsyncTask. The problem is - you can't subclass it in scala because of some compiler bug regarding varargs parameters. Silly.

So let's do a lightweight(in terms of code) substitution. We can “spawn a thread” using scala's actors. And activity can receive messages through a Handler.

import android.os.{Message, Handler}

trait MessageHandler {
  private val handler = new Handler {
   override def handleMessage(msg: Message) {

  def react(msg: AnyRef)

  def !(message: AnyRef) {
   val msg = new Message
   msg.obj = message

So in an activity that mixes in MessageHandler I can post messages to my self. And I can do it async since Handler is thread safe

def react(msg: AnyRef) = msg match {
  case s: String = Toast(this, s, Toast.LENGTH_LONG).makeText()
  case _ => ()

//somewhere inside UI thread
val that = this
  val msg = doExpensiveWork()
  that ! msg

Not the most concise way to write it, but I believe the most clear one. Method doExpensiveWork is done in the background so it doesn't block UI and it posts the result back as a message.

Async composition - finally

What I want to do now is use function composition to do something like

(input --> expensiveOne --> expensiveTwo) -!> displayResults

In other words, do “waterfall composition” in background using some input I have now and post the result back into the UI thread to method displayResults. That should be the magic of the -!> operator. Do the left side in bg and post it to the right side async. I need a new trait for that

trait MessageHandlerExec extends MessageHandler { outer =>
  override protected val handler = new Handler {
    override def handleMessage(msg: Message) = msg.obj match {
      case r: Runnable =>
      case other: AnyRef => react(other)

  implicit def any2asyncComposable[A](a: => A) = new AnyRef{
    def -!>[B](f: A=>B) = outer ! new Runnable{
      def run() = f(a)

The trick here is using by-name parameters in the implicit conversion. This delays the execution of a(which in example above would be a waterfall and moves it into a worker thread.

Last modified on 2012-09-17