Scala Shapeless Code for Project Euler #2

Question

I've started on my Shapeless solution to Project Euler Problem #2.

I can sum together all the even fibs up to the Nth even one with this code:

import shapeless._, nat._, ops.nat.Sum

trait Fibonacci[N <: Nat] { type Out <: Nat }

object Fibonacci {
  type Aux[N <: Nat, Out0 <: Nat] = Fibonacci[N] { type Out = Out0 }

  def apply[N <: Nat](i: Nat)(implicit fib: Aux[i.N, N], n: Witness.Aux[N]):N = n.value

  implicit val fib0 = new Fibonacci[_0] { type Out = _2 }
  implicit val fib1 = new Fibonacci[_1] { type Out = _3 }

  implicit def fibN[I <: Nat, L <: Nat, M <: Nat](implicit l: Aux[I, L],
                                                  m: Aux[Succ[I], M],
                                                  sum: Sum[L, M]) =
    new Fibonacci[Succ[Succ[I]]] { type Out = sum.Out }
}

trait Fibs[N <: Nat] { type Out <: Nat }

object Fibs {
  type Aux[N <: Nat, Out0 <: Nat] = Fibs[N] { type Out = Out0 }

  def apply[N <: Nat](i: Nat)(implicit fibs: Aux[i.N, N], n: Witness.Aux[N]):N = n.value

  implicit def fibs0(implicit f: Fibonacci[_0]) = new Fibs[_0] { type Out = f.Out }

  implicit def fibsN[N <: Nat, R <: Nat, Z <: Nat](implicit fib: Fibonacci.Aux[Succ[Succ[Succ[N]]], R],
                                                   fibs: Aux[N, Z],
                                                   sum: Sum[R, Z]) =
    new Fibs[Succ[N]] {
      type Out = sum.Out
    }
}

Now I can do:

val (evenFibs0, evenFibs1) = (Fibs(0), Fibs(1))
typed[_2](evenFibs0)
typed[_10](evenFibs1)

This is how I get all even fibs: I start with the sequence 2, 3, ... , and I sum up every third Fibonacci number.

Now, I am stuck. I'd like to have functionality similar to takeWhile, so I can write a function that accepts a limit and returns the sum of my even fibs whose terms don't exceed that limit. Any ideas?

Here's my effort for what I've tried so far:

trait EvenFibsWithLimit[N <: Nat, M <: Nat] { type Out <: Nat }

trait LowPriorityFibs3 {
  type Aux[N <: Nat, M <: Nat, Out0 <: Nat] = EvenFibsWithLimit[N, M] { type Out = Out0 }

  implicit def fibs0[M <: Nat] = new EvenFibsWithLimit[_0, M] { type Out = _0 }

  implicit def fibsGT[N <: Nat, M <: Nat, O <: Nat](implicit f: EvenFibsWithLimit[N, M],
                                                    fib: Fibs.Aux[N, O],
                                                    l: ops.nat.LT[M, O]) = f
}

object EvenFibsWithLimit extends LowPriorityFibs3 {
  def apply[N <: Nat, O <: Nat](limit: Nat)(implicit fibs: Aux[N, limit.N, O],
                                            o: Witness.Aux[O]): O = o.value

  implicit def fibsN[N <: Nat, M <: Nat, O <: Nat](implicit f: EvenFibsWithLimit[N, M],
                                                   f2: Fibs.Aux[Succ[N], O],
                                                   d: ops.nat.Diff[M, O]) =
    new EvenFibsWithLimit[Succ[N], d.Out] {
      type Out = O
    }
}

The idea was to recursively subtract the limit by the output until the output is less than the limit. I can definitely smell something is off. I don't think I need Diff at all.. I've tried some other variations too, but I keep getting stuck. When I compile, I get the error diverging implicit expansion for fibsN.

EDIT:

I was thinking maybe I can construct an HList of my Fibs, and use Selector with a predicate typeclass to simulate a takeWhile. Thoughts?

Answer 1

I find that the best way to approach this kind of problem is to step through the natural numbers while thinking about the information I need to keep track of.

On paper I'd use something like this:

N  1  2  3  4  5  6  7  8  9 10 11 12 13 14 15
C  1  2  3  3  5  5  5  8  8  8  8  8 13 13 13
P  1  1  2  2  3  3  3  5  5  5  5  5  8  8  8
L  0  2  2  2  2  2  2 10 10 10 10 10 10 10 10

Here C tracks the current number in the Fibonacci sequence—i.e. the largest one that's less than or equal to N. P is the Fibonacci number before that, and L is the current sum of the even ones we've seen so far.

We can translate this into a type class:

import shapeless._, ops.nat.{ Mod, Sum }, nat.{ _0, _1, _2 }

trait EvenFibs[N <: Nat] {
  type L <: Nat
  type C <: Nat
  type P <: Nat
}

Now there are four cases we need to handle. First I'll give the one that needs to have the lowest priority in order to get the implicit resolution right:

trait LowPriorityEvenFibs {
  type Aux[N <: Nat, L0 <: Nat, C0 <: Nat, P0 <: Nat] = EvenFibs[N] {
    type L = L0
    type C = C0
    type P = P0
  }

  implicit def ef3[N <: Nat](implicit
    ef: EvenFibs[N]
  ): Aux[Succ[N], ef.L, ef.C, ef.P] = new EvenFibs[Succ[N]] {
    type L = ef.L
    type C = ef.C
    type P = ef.P
  }
}

This is the case where Succ[N] is not a Fibonacci number. It doesn't require us to update any of the values we're keeping track of. The next case is a little more complex:

trait MidPriorityEvenFibs extends LowPriorityEvenFibs {
  implicit def ef2[N <: Nat, L0 <: Nat, PP <: Nat, P0 <: Nat](implicit
    ef: Aux[N, L0, P0, PP],
    sum: Sum.Aux[PP, P0, Succ[N]]
  ): Aux[Succ[N], L0, Succ[N], P0] = new EvenFibs[Succ[N]] {
    type L = L0
    type C = Succ[N]
    type P = P0
  }
}

This is the case where Succ[N] is an odd Fibonacci number. In this case we need to update C and P, but not L.

Our last Succ[N] case is the one where Succ[N] is an even Fibonacci number. I'll give it here with the base case:

object EvenFibs extends MidPriorityEvenFibs {
  implicit def ef0: Aux[_0, _0, _1, _1] = new EvenFibs[_0] {
    type L = _0
    type C = _1
    type P = _1
  }

  implicit def ef1[N <: Nat, L <: Nat, PP <: Nat, P0 <: Nat](implicit
    ef: Aux[N, L, P0, PP],
    sum: Sum.Aux[PP, P0, Succ[N]],
    mod: Mod.Aux[Succ[N], _2, _0],
    current: Sum[Succ[N], L]
  ): Aux[Succ[N], current.Out, Succ[N], P0] = new EvenFibs[Succ[N]] {
    type L = current.Out
    type C = Succ[N]
    type P = P0
  }

  def apply[N <: Nat](implicit ef: EvenFibs[N]): Aux[N, ef.L, ef.C, ef.P] = ef
}

Lastly we can define a helper class that will make it easier to check our work:

class ComputeHelper[N <: Nat] {
  def value[L <: Nat, C <: Nat, P <: Nat](implicit
    ef: EvenFibs.Aux[N, L, C, P],
    wit: Witness.Aux[L]
  ): L = wit.value
}

def compute[N <: Nat]: ComputeHelper[N] = new ComputeHelper[N]

And then:

test.typed[_0](compute[_0].value)
test.typed[_0](compute[_1].value)
test.typed[_2](compute[_2].value)
test.typed[_2](compute[nat._3].value)
test.typed[_2](compute[nat._4].value)
test.typed[_2](compute[nat._5].value)
test.typed[_2](compute[nat._6].value)
test.typed[_2](compute[nat._7].value)
test.typed[nat._10](compute[nat._8].value)
test.typed[nat._10](compute[nat._9].value)

The last line takes about twenty seconds to compile on my machine, but it works.