Home

Awesome

Quark: Common combinators for Elixir

Build Status Inline docs Deps Status hex.pm version API Docs license

Table of Contents

Quick Start


def deps do
  [{:quark, "~> 2.3"}]
end

defmodule MyModule do
  use Quark

  # ...
end

Summary

Elixir is a functional programming language, but it lacks some of the common built-in constructs that many other functional languages provide. This is not all-together surprising, as Elixir has a strong focus on handling the complexities of concurrency and fault-tolerance, rather than deeper functional composition of functions for reuse.

Includes

Functional Overview

Curry

Functions

curry creates a 0-arity function that curries an existing function. uncurry applies arguments to curried functions, or if passed a function creates a function on pairs.

Macros: defcurry and defcurryp

Why define the function before currying it? defcurry and defcurryp return fully-curried 0-arity functions.


defmodule Foo do
  import Quark.Curry

  defcurry div(a, b), do: a / b
  defcurryp minus(a, b), do: a - b
end

# Regular
div(10, 2)
# => 5

# Curried
div.(10).(5)
# => 2

# Partially applied
div_ten = div.(10)
div_ten.(2)
# => 5

Partial

:crown: We think that this is really the crowning jewel of Quark. defpartial and defpartialp create all arities possible for the defined function, bare, partially applied, and fully curried. This does use up the full arity-space for that function name, however.

Macros: defpartial and defpartialp


defmodule Foo do
  import Quark.Partial

  defpartial one(), do: 1
  defpartial minus(a, b, c), do: a - b - c
  defpartialp plus(a, b, c), do: a + b + c
end

# Normal zero-arity
one
# => 1

# Normal n-arity
minus(4, 2, 1)
# => 1

# Partially-applied first two arguments
minus(100, 5).(10)
# => 85

# Partially-applied first argument
minus(100).(10).(50)
# => 40

# Fully-curried
minus.(10).(2).(1)
# => 7

Pointfree

Allows defining functions as straight function composition (ie: no need to state the argument). Provides a clean, composable named functions. Also doubles as an aliasing device.

defmodule Contrived do
  import Quark.Pointfree
  defx sum_plus_one, do: Enum.sum() |> fn x -> x + 1 end.()
end

Contrived.sum_plus_one([1,2,3])
#=> 7

Compose

Compose functions to do convenient partial applications. Versions for composing left-to-right and right-to-left are provided

The operator <|> is done "the math way" (right-to-left). The operator <~> is done "the flow way" (left-to-right).

Versions on lists also available.

import Quark.Compose

# Regular Composition
sum_plus_one = fn x -> x + 1 end <|> &Enum.sum/1
sum_plus_one.([1,2,3])
#=> 7

add_one = &(&1 + 1)
piped = fn x -> x |> Enum.sum |> add_one.() end
composed = add_one <|> &Enum.sum/1
piped.([1,2,3]) == composed.([1,2,3])
#=> true

sum_plus_one = (&Enum.sum/1) <~> fn x -> x + 1 end
sum_plus_one.([1,2,3])
#=> 7

# Reverse Composition (same direction as pipe)
x200 = (&(&1 * 2)) <~> (&(&1 * 10)) <~> (&(&1 * 10))
x200.(5)
#=> 1000

add_one = &(&1 + 1)
piped = fn x -> x |> Enum.sum() |> add_one.() end
composed = (&Enum.sum/1) <~> add_one
piped.([1,2,3]) == composed.([1,2,3])
#=> true

Common Combinators

A number of basic, general functions, including id, flip, const, pred, succ, fix, and self_apply.

Classics

SKI System

The SKI system combinators. s and k alone can be combined to express any algorithm, but not usually with much efficiency.

We've aliased the names at the top-level (Quark), so you can use const rather than having to remember what k means.

 1 |> i()
#=> 1

"identity combinator" |> i()
#=> "identity combinator"

Enum.reduce([1,2,3], [42], &k/2)
#=> 3

BCKW System

The classic b, c, k, and w combinators. A similar "full system" as SKI, but with some some different functionality out of the box.

As usual, we've aliased the names at the top-level (Quark).

c(&div/2).(1, 2)
#=> 2

reverse_concat = c(&Enum.concat/2)
reverse_concat.([1,2,3], [4,5,6])
#=> [4,5,6,1,2,3]

repeat = w(&Enum.concat/2)
repeat.([1,2])
#=> [1,2,1,2]

Fixed Point

Several fixed point combinators, for helping with recursion. Several formulations are provided, but if in doubt, use fix. Fix is going to be kept as an alias to the most efficient formulation at any given time, and thus reasonably future-proof.

fac = fn fac ->
  fn
    0 -> 0
    1 -> 1
    n -> n * fac.(n - 1)
  end
end

factorial = y(fac)
factorial.(9)
#=> 362880

Sequence

Really here for pred and succ on integers, by why stop there? This works with any ordered collection via the Quark.Sequence protocol.

succ 10
#=> 11

42 |> origin() |> pred() |> pred()
#=> -2