Scala for Haskellers

    •    

It’s only half an hour till ICFPC 2015 and your team suddenly decides to switch to Scala for this year? Don’t panic — you’ve got just enough time to bring yourself up to speed.

Let’s kick off with…

Random facts about Scala

Basic types

There’s a whole set of integer types (just like in Java):

Haskell:   Int8  Int16  Int32  Int64   -- from Data.Int
Scala:     Byte  Short  Int    Long

Floating-point types are the same two, Float and Double; no surprise there.

Character types, Char and String, are spelled the same, but Scala’s Char is actually a 16-bit unsigned Unicode character, and thus String is in UTF-16.

Boolean type is spelled a bit differently and, most importantly, its values are all lowercase:

Haskell:   Bool      -- True and False
Scala:     Boolean   // true and false

() (the type) is spelled Unit (but the value is still ()).

Save for String, all of the above are value types. They have a nice property: they can’t be null. Reference types, on the other hand, can. This distinction is mirrored in the class hierarchy of Scala: value types are subtypes of AnyVal, while reference types are subtypes of AnyRef. Finally, all the types are subtypes of Any.

Nothing is a subtype of every other type.

Null is a type inhabited only by null.

Algebraic data types

Being an object-oriented language, Scala uses class hierarchies to build ADTs.

data Tree a = Node { left  :: Tree a
                   , elem  :: a
                   , right :: Tree a }
            | Leaf
sealed abstract class Tree[A]
case class Node[A](left: Tree[A], elem: A, right: Tree[A])
    extends Tree
object Leaf extends Tree

Note the sealed keyword — that makes Tree a closed ADT, i.e. one that can’t be extended with new value constructors (just like in Haskell). You can omit the keyword if you need an open ADT, of course.

case keyword before the class definition enables the following:

Leaf could’ve been defined as another class, but since it has no constructor parameters, all the instances will be the same. Thus an object definition which makes Leaf a singleton.

Pattern matching

smth match {
    case 1 => 0;
    case x => x-1;
    case _ => 42;
}

Anonymous functions

(x, y) => x + y

Function definitions

Return type annotations are optional, argument type annotations aren’t:

def fib(n: Int): Int = …
def fib2(n: Int) = …

Unlike Haskell, Scala functions aren’t curried by default — it’s your responsibility:

def greet(greeting: String)(name: String): Unit =
    println(greeting + ", " + name + "!")

greet("Hello")("dear reader")

Partial application

You can put single _ after the function or method to get partially applied function:

val welcome = greet("Welcome") _
val anotherNameForGreet = greet _

But you can also use _ as a placeholder for some of the arguments:

def f(a: Double, b: Double, c: Double): Double = a + b / c
val g = f(1, _, 3)

Polymorphism and type constraints

Scala doesn’t rely on hierarchy of typeclasses to overload common functions and operators — here, you can just implement method + and be done with it. Thus, type constraints are used only to limit subtyping.

For example, if we have class C2 that extends class C1 which, in turn, extends AnyVal, the following code will work on both C1 and C2, but not on AnyVal:

def f[A <: C1](x: A) = …

Conversely, the following will work on C1 and AnyVal (and Any, which is a superclass of AnyVal), but not on C2:

def f2[A >: C1](x: A) = …

(These margins are too narrow to contain meaningful examples, sorry.)

Traits

Traits are like Haskell’s typeclasses — they define an interface that a type should implement.

trait PrettyPrintable {
    def PrettyPrint(): Unit
}

Implicit parameters

Okay, so if traits and type constraints are not used to write contracts, what is? Meet implicit parameters.

def adder[A](x: A, y: A)(implicit a: Numeric[A]): A = {
    a.plus(x, y)
    /* Alternatively:
    
    import a._
    x + y
    */
}

Variance annotations

Another thing that you never had to deal with in Haskell is co- and contravariance (because Haskell doesn’t have subtyping). If you want your trees to be covariant, just add plus before the type variable:

class Tree[+A](…)

To make it contravariant, use - instead.

Lists

Lists are eager. All the usual functions on lists — head, tail, map etc. — are methods in Scala.

colours :: [String]
colours = [ "red", "green", "blue" ]
bw = "black" : "white" : []
more_colours = colours ++ bw
h, t = (head bw, tail bw)

empty :: Bool
empty = null bw
val colours: List[String] = List("red", "green", "blue")
val bw = "black" :: "white" :: Nil
val more_colours = colours ::: bw
val (h, t) = (bw.head, bw.tail)
val empty: Boolean = bw.isEmpty

forM_ is spelled foreach, all and any become forall and exists.

foldLeft in Scala (also spelled /:) is the same as flip $ foldl' in Haskell, i.e. it takes the base value as the first argument and the function as the second. There’s also reduceLeft which is like foldl1, only strict; foldRight a.k.a. :\ (arguments are flipped, too); and reduceRight.

Streams

Streams are effectively lazy lists.

a = "one" : "two" : []
b = a ++ ["three", "four"]
val a: Stream[String] = Stream.cons("one",
                        Stream.cons("two",
                        Stream.empty))
val b = a.append(Stream("three", "four"))

Iterators

…are imperative counterparts of streams. Iterator[+A] is a trait specifying two methods, hasNext: Boolean and next: A. There’s also BufferedIterator[+A] specifying method head: A, which is just like next, but doesn’t advance the iterator.

Iterators implement all of the usual HOFs.

You can turn your own class into iterator by extending the trait, or you can create it ad-hoc:

val fib = new Iterator[Int] {
    private var current = 1;
    private var previous = 1;
    def hasNext = true
    def next = {
        val result = current + previous
        previous = current
        current = result
        result
    }
}

For-comprehensions

[ x * y | x <- [1..10], y <- [1..x], y `mod` 2 == 0 ]

import Control.Monad (forM_)
forM_ [1..4] $ \x ->
    print $ x * x
for { x <- List.range(1, 11)
      y <- List.range(1, x+1)
      if y % 2 == 0 }
yield x * y

for (x <- List.range(1, 5)) {
    println(x * x)
}

Mutability

To create a mutable variable instead of immutable value, use var instead of val:

var a:Int = _
a = 42

_ is a wildcard that will initialize the variable with the default value (0 for numbers, false for Booleans, null for reference types).

Laziness

You can defer initialization of any val by prefixing its declaration with lazy keyword. This can also be used to avoid initialization order headaches.

To make functions accept their arguments by name (default is by value), prefix them with =>:

def myWhile(condition: => Boolean)(action: => Unit): Unit =
    if(condition) {
        action
        myWhile(condition)(action)
    }

Imperative programming

Scala has all the traditional imperative control structures: while and do-while loops (for loops are present as for-comprehensions), ifs with optional else branches, return and so on. Note, though, that there’s no break and continue for loops.

Minimal working program

You’re done with theory, now on to writing some real code! Oh wait, you don’t know how to compile stuff yet. Well, just head over to tpolecat’s blog and read “Setting up to write some Scala” — it’ll get you all set up in no time. He provides example program, too, so my job here is done.

Note of thanks

I’d like to thank kind people on scala@conference.jabber.ru for answering my confused questions and explaining how things should be done. The same goes for attendees of #scala and #scalaz on Freenode — you people rock!

Update 08.08.2015:

 

Till next time! Minoru out.

Your thoughts are welcome by email
(here’s why my blog doesn’t have a comments form)