Functor

Functor requires only a map method, which for a Parser will have the type signature

Parser[A].map[B](f: A => B): Parser[B]

This means that we can take an existing parser that produces output of one type, and turn it into a parser that produces an output of some other type. Is this useful? It turns out to be very useful. Right now we can only create a Parser[String]. Adding map means that we can turn that String into any other type we care about, like an Int for example. Here's an example.

val one: Parser[Int] = Parser.string("1").map(_.toInt)
one.parse("1")
// res0: Result[Int] = Success(result = 1, input = "1", offset = 1)

Having decided that we want map we'll now implement it in three stages:

  1. creating the map method on Parser;
  2. creating the type class instance in the Parser companion object; and
  3. implementing tests.

Let's go!

Remember we're using the reification pattern for our implementation, for which we need to work out if map is a constructor, combinator, or interpreter. The types tell us: map takes in a Parser and returns a Parser so it is a combinator. This means we reify it.

We create a case class that holds all the input to the map method (remember there is the hidden this parameter!)

final case class ParserMap[A, B](source: Parser[A], f: A => B) extends Parser[B]

and the implementation of map simply creates an instance of this data structure.

def map[B](f: A => B): Parser[B] =
  ParserMap(this, f)

Easy! Now on to parse. We add a case to the pattern match for ParserMap

case ParserMap(source, f) => ???

Our first step is to apply the recursion rule for algebraic data types: when the data is recursive the method is recursive.

case ParserMap(source, f) =>
  loop(source, index)

Our recursive call to loop returns a Result. Result is an algebraic data type so we can use structural recursion to make progress.

case ParserMap(source, f) =>
  loop(source, index) match {
    case Failure(reason, input, start) => ???
    case Success(result, input, offset) => ???
  }

Now we can follow the types to finish up the implementation. In the case of Failure there is nothing we can do, as we have no data of type A to apply to the function f. So we just return the Failure.

case ParserMap(source, f) =>
  loop(source, index) match {
    case Failure(reason, input, start) => Failure(reason, input, start)
    case Success(result, input, offset) => ???
  }

In the Success case we have a result of type A to we apply f to it to create a value of type B and return a success of that.

case ParserMap(source, f) =>
  loop(source, index) match {
    case Failure(reason, input, start) => Failure(reason, input, start)
    case Success(result, input, offset) =>
      Success(f(result), input, offset)
  }

Done, and everything was created using a systematic and repeatable process.

Now it's on to the type class instance. Remember type classes go in the companion object. Beyond that the implementation is straightforward. We only need to implement map and we can do that by calling the map method we've just created.

implicit val parserFunctorInstance: Functor[Parser] =
  new Functor[Parser] {
    def map[A, B](fa: Parser[A])(f: A => B): Parser[B] =
      fa.map(f)
  }

One small note: remember that Applicative extends Functor, and Monad extends Applicative. So if and when you come to implement type class instances for these types make sure you don't have two implementations for the type classes they extend. In other words, if you implement Applicative remove the instance for Functor. If you don't do this there will be ambiguity when the compiler looks for a Functor instance.

Finally we have tests. We need to test two conditions:

  1. map does the expected transform when the underlying parser succeeds; and
  2. map fails if the underlying parser fails.

You might notice that these two conditions are exactly the two cases we implemented in the structural recursion on Result. It would be very hard to incorrectly implement map but we're going add tests anyway.

The tests are straight-forward modifications of the existing tests for Parser.string. I'm not going to include the source code here because it's quite lengthy for very little content. You can find it in the repository.

Now it is over to you. For each type class below think about possible uses in the context of Parser. If you decide it is useful implement it and add tests.

Monoid→