Skip to content

Chapter 5: Strictness and laziness

Piyush Maheshwari edited this page Apr 14, 2016 · 14 revisions

Contents

  • Notes: Chapter notes and links to further reading related to the content in this chapter
  • FAQ: Questions related to the chapter content. Feel free to add questions and/or answers here related to the chapter.

Notes

Non-strictness vs laziness

The Haskell website has a good explanation of the difference between non-strictness and laziness.

In short, "non-strict" just means "not strict". There are many possible evaluation strategies one could employ when evaluating a program, and strict evaluation is one of them. Non-strict evaluation is a class of evaluation strategies, and lazy evaluation is one non-strict strategy (also known as "call-by-need").

In Scala, non-strict arguments are sometimes called "by name" arguments, in reference to the fact that the evaluation strategy Scala employs for those arguments is call-by-name. We can turn an argument into call-by-need by caching it in a lazy val inside the function:

def pairIf[A](b: Boolean, x: => A) = {
  lazy val y = x
  if (b) (y, y)
}

This function will evaluate x only once, or never if the boolean b is false. If we said (x, x) instead of (y, y), it would evaluate x twice.

The chapter explains that when an expression does not terminate, it is said to evaluate to bottom. At first glance this is counterintuitive because there is a natural tendency to think of infinity as having no bottom. But the bottom to which the chapter refers is actually the bottom type. See Haskell's definition of bottom for a more thorough description. Scala refers to it as Nothing, which is at the bottom of the inheritance hierarchy.

Corecursion and codata

The Wikipedia article on corecursion is a good starting point for understanding the concept.

The article on Coinduction has some further links. Dan Piponi's article "Data and Codata" talks about corecursion as "guarded" recursion.

Ralf Hinze's paper "Reasoning about Codata" brings equational reasoning to corecursive programs by employing applicative functors. Hinze's paper will be more comprehensible to readers who have finished part 3 of our book.

Tying the knot

Non-strictness allows us to create cyclic streams such as:

val cyclic: Stream[Int] = 0 #:: 1 #:: cyclic

This may seem like it shouldn't work. The stream is referencing itself in its own tail! But the trick is that the #:: constructor is non-strict in its second argument. The evaluation of cyclic will stop without expanding the expression 1 #:: cyclic. It's not until somebody takes the tail of the tail of cyclic that the recursive reference is expanded, and again it expands only one element at a time, allowing for an infinite, cyclic stream.

Note that the cyclic stream is reusing its own structure. cyclic.tail.tail is not a new stream that looks like cyclic. It really is the same object as cyclic in every sense:

scala> cyclic.tail.tail eq cyclic
res0: Boolean = true

This technique is sometimes called "Tying the Knot". For more information see the Haskell.org article.

However, be careful of creating such structures using the Scala standard library's Stream. They can be quite fragile. The reason is that scala.Stream is strict in the head element and it will also memoize the computed contents, which can lead to memory leaks.

Stream.apply

Be careful using Stream.apply, both in the standard library and in the exercises: they are constructed using repeated parameters, which are always strict. This means that e.g.

Stream({println("One"); 1}, {println("Two"); 2}, {println("Three"); 3})

will immediately print One, Two, Three. Although the Stream will be constructed lazily, the contents have already been evaluated.

For truly lazily constructed Streams you can always resort to #:: (which still evaluates the head value strictly!) or nested cons(..,cons(..,..)) operators in the exercises.

FAQ

You can’t perform that action at this time.