Awesome Open Source
Awesome Open Source

You don't (may not) need loops ➿

Join the community on Spectrum

Prerequisites:

Loops are one of the first constructs that junior programmers learn, but they can pose many potential issues in the software development process, and could be avoided in many cases.

"Isn't this another loops vs recursions?" No, neither is particularly good in fact. reduce is pretty low level too and functional developers don't use it much either. But it's very important to know where those expressive higher-order functions are coming from.

Loops include for, forEach, while, do, for...of and for...in. You might argue that built in array methods such as map or reduce also uses loops. Well that's true, so we are going to define our own. In real life, you’d use a library or built in array methods, but it's good to start from scratch and understand the principles. The performance won't be great, you ask. Yes I heard you, and please read on for now.

JavaScript is about trade-offs. There’s a tension between writing code that is performant, code that is maintainable and easy to understand, and code that is correct by construction. It's probably very hard to balance them and that's the source of debates in your pull requests.

Correctness by construction

Simple English: No bugs

Loops have four main problems: Off-by-one error, Infinite loop, Statefulness and Hidden intent. You might argue loops like for...in won't have Off-by-one error, yes but it's still stateful and can hide intent. Recursions have some of the problems too.

Ergonomics and maintainability

Simple English: No refactoring

Many developers hate it when there's change of requirements, because they have spent so much time on writing performant and bug-free code. When there's new requirements, you'll have to restructure your code and update your unit tests. Can you move your loops freely in your codebase? probably not, because there must be side effects or mutations. Big loops and nested loops are inevitable sometimes for performance reasons. You could do anything in a loop including uncontrolled side effects and therefore, it often breaks rule of least power. Languages such as Haskll uses fusion to "merge" iterations. Wholemeal programming is a nice pattern to make code modular and reusable.

Runtime performance

You can write the most performant code with loops and everything. But is it still performant when there's change of requirements? Is your performant code understandable by other people? Is your code still performant once you've refactored your code? At a larger scale, Manual optimization reduces code reusability, modularity and makes components more complex. Code becomes harder to understand, and harder to test for correctness.

Keep in mind that your code will CHANGE and will be read by your colleagues. If you write throw away code, don't bother worrying about code quality at all.

So, it's all about balancing the triangle. In modern engineering teams, 95% of the time you'd sacrifice performance for correctness and ergonomics since computers are fast enough and premature optimization is usually considered bad. But to replace loops, there will be huge performance hit and even stack overflow. While all three point are equally important, this article focuses more on correctness and ergonomics, and in real-world projects, you'll need to use your best knowledge to do the trade-offs. If you are interested in striving not to sacrifice any of these three, have a look at Haskell. It is designed from the ground up to be highly performant in FP.

We do expect you to know some very basic stuff about functional programming and there are many other articles online (EG: why ternary is used instead of ifs, why you shouldn't mutate variables and Complexity/TCO, etc).

You are welcome to contribute with more items provided below.

  • Please send a PR if you want to add or modify the code. No need to open an issue unless it's something big and you want to discuss.

Voice of Developers

Early imperative languages didn't support recursion at all and even modern ones support it poorly, forcing them to use something else to iterate—loops. A combination of recursion and higher-order functions does the same thing but better and more naturally.

Tikhon Jelvis, lead data scientist at Target working on supply chain optimization and simulation

We use side effects to change the state of a program over time in the loop - the program state is mutable. But state, especially non-local state, makes programs hard to write, debug, and maintain.

Mark Sheldon, Lecturer in Computing at Tufts University

Avoid The One-off Problem, Infinite Loops, Statefulness and Hidden intent.

JOAB JACKSON, Managing Editor at The New Stack University

Nested loops, continue, break and goto are clever tricks to trap you. They are confusing and unmaintainable.

—Well, this one is kinda common sense :)

If you are still writing loops, you’re not a bad person. Just think about whether you need to write loops or if there’s a better alternative. Loops are best executed at the CPU level, well-beneath the concerns of most developers.

Marco Emrich, Software crafter, web dev, code coach, code retreat facilitator, author, consultant

ESLint Plugin

There's a rule in eslint-plugin-fp. There are also many other useful rules in the plugin so please do check them out!

Potential problems 👿

Name Off-by-one error Infinite loop Statefulness Hidden intent
Loops Yes 😱 Yes 😱 Yes 😱 Yes 😱
Iterables NO 💚 NO 💚 Yes 😱 Yes 😱
Recursion (Without higher-order functions) NO 💚 Yes 😱 NO 💚 Yes 😱
Recursion (With higher-order functions) NO 💚 NO 💚 NO 💚 NO 💚
Corecursion NO 💚 NO 💚 NO 💚 NO 💚
Transducers NO 💚 NO 💚 NO 💚 NO 💚
Monoids NO 💚 NO 💚 NO 💚 NO 💚
F-Algebras NO 💚 NO 💚 NO 💚 NO 💚

Limitations

Name Iteration Transformation Accumulation
Loops
Recursion
Corecursion
Transducers
Monoids
F-Algebras

Quick Links

Recursion

  1. Sum
  2. Reverse
  3. Tail recursive sum
  4. Reduce

With higher-order functions

  1. Sum
  2. Reverse
  3. Map
  4. Filter
  5. All
  6. Any
  7. Size
  8. Max
  9. Min
  10. SortBy
  11. Find
  12. GroupBy
  13. First
  14. Last
  15. Take
  16. Drop
  17. Paramorphism

Corecursion

  1. Unfold
  2. Range
  3. Linked list
  4. Tree

Transducers

  1. Map
  2. Filter
  3. Filter and Map

Monoids

  1. Sum
  2. Product
  3. Max
  4. All
  5. Any

F-Algebras

  1. Catamorphism
  2. Sum
  3. Map
  4. Anamorphism
  5. arrToList
  6. makeAlphabet
  7. range
  8. Real-world examples

Recursion

You can immediately avoid off-by-one error and state by using recursions.

Let's define some helper functions:

const first = xs => xs[0]
const rest = xs => xs.slice(1)

NOTE: functions like this could be defined with reduce too, but you can easily hit stack overflow. For all intensions and purposes let's use existing array methods.

Sum

const sum = xs =>
  xs.length === 0
    ? 0
    : first(xs) + sum(rest(xs));

⬆ back to top

Reverse

const reverse = xs => 
  xs.length === 0
    ? []
    : reverse(rest(xs)).concat(first(xs));

⬆ back to top

Tail recursive sum

const sum = list => {
  const go = (acc, xs) =>
    xs.length === 0
      ? acc
      : go(acc + first(xs), rest(xs));
  return go(0, list) 

⬆ back to top

Reduce

const reduce = (f, acc, xs) =>
  xs.length === 0
    ? acc
    : reduce(f, f(acc, first(xs)), rest(xs));

NOTE: Since tail call optimization is currently only supported by Safari, tail recursion may cause stack overflow in most other JavaScript environments. While others, such as the Chrome devs, appear to be discussing the subject on-and-off, you may wish to, in this case, use a loop here to compromise (and this is an example of balancing the triangle):

const reduce = function(reduceFn, accumulator, iterable){
  for (let i of iterable){
    accumulator = reduceFn(accumulator, i)
  }
  return accumulator
}

⬆ back to top

Higher-order functions

Recursion is too low-level. Not low-level in the sense of direct access to the machine but low-level in the sense of language design and abstraction. Both loops and recursions do a poor job of signalling intent. This is where higher-order functions come in. Map, filter, fold and friends package up common recursive patterns into library functions that are easier to use than direct recursion and signal intent.

Sum

const sum = xs => 
  reduce((acc, x) => x + acc, 0, xs)
sum([1,2,3])
// => 6

⬆ back to top

Reverse

const reverse = xs =>
  reduce((acc, x) => [x].concat(acc), [], xs)

⬆ back to top

Map

const map = (f, xs) =>
  reduce((acc, x) => acc.concat(f(x)), [], xs)

⬆ back to top

Filter

const filter = (f, xs) =>
  reduce((acc, x) => f(x) ? acc.concat(x) : acc, [], xs)

⬆ back to top

All

const all = xs =>
  reduce((acc, x) => acc && x, true, xs)

⬆ back to top

Any

const any = xs =>
  reduce((acc, x) => acc || x, false, xs)

⬆ back to top

NOTE: The following sections are considered somewhat advanced. You don't have to understand all the details of the jargons, but rather get an overall intuition on how you could abstract things so that they can compose well. You can start learning it here. This course is widely recommended by Haskell learners.

Paramorphism

const para = (f, acc, xs) =>
  xs.length === 0 
    ? acc
    : para(f, f(acc, first(xs), xs), rest(xs));

⬆ back to top

Corecursion

Unfold

const unfold = (f, seed) => {
  const go = (f, seed, acc) => {
    const res = f(seed);
    return res ? go(f, res[1], acc.concat([res[0]])) : acc; 
  }
  return go(f, seed, [])
}
unfold(x => 
  x < 26
    ? [String.fromCharCode(x + 65), x + 1]
    : null
, 0);
//=> [A,B,C,D,E,F,G,H,I,J,K,L,M,N,O,P,Q,R,S,T,U,V,W,X,Y,Z]

⬆ back to top

Range

const range = (i, count) =>
  unfold(x => (x <= count) 
    ? [x, x+1]
    : null
, i);
range(5, 10)
//=> [ 5, 6, 7, 8, 9, 10 ]

⬆ back to top

Linked list

const Nil = {}
const _Cons = function(h, tl) {
  this.head = h;
  this.tail = tl;
};
const Cons = (h, tl) => 
  new _Cons(h, tl)
const fold = (f, acc, xs) => 
  xs.head 
    ? fold(f, f(acc, xs.head), xs.tail)
    : acc
const lst = Cons(3, Cons(4, Cons(5, Nil)));
fold((acc, x) => acc + x, 0, lst)
//=> 12

⬆ back to top

Tree

const Empty = {}
const _Leaf = function(x) { this.x = x; }
const Leaf = x => new _Leaf(x)
const _Node = function(l, x, r) {
  this.left = l;
  this.x = x;
  this.right = r;
}
const Node = (l, x, r) => new _Node(l, x, r)
const tree = Node(Node(Leaf(2), 1, Leaf(3)), 0, Leaf(4))
fold((acc, x) => acc + x, 0, tree) // Try to implement `fold` yourself
//=> 10

⬆ back to top

Transducers

Adding stateful transducers and grouping operations.

Helper functions:

const concat = (a, b) => a.concat(b)

Map

const mapper = (f, cnct) => (acc, x) => 
  cnct(acc, f(x))
reduce(mapper(x => x + 1, concat), [], [1,2,3]) 
//=> [2,3,4]

⬆ back to top

Filter

const filterer = (f, cnct) => (acc, x) => 
  f(x) ? cnct(acc, x) : acc
reduce(filterer(x => x > 1, concat), [], [1,2,3]) 
//=> [2,3]

⬆ back to top

Filter and Map

reduce(
  filterer(x => x > 1,
  mapper(x => x + 1, concat)),
  [], [1,2,3]
)
//=> [3,4]
// Try to implement append yourself 
reduce(filterer(x => x > 1,
       mapper(x => x + 1, append)),
       Nil, Cons(1, Cons(2, Cons(3, Nil))))
//=> [3,4]
// Try to implement insert yourself 
reduce(filterer(x => x > 1,
       mapper(x => x + 1, insert)),
       Empty, Node(Node(Leaf(2), 1, Leaf(3)), 0, Leaf(4)))
//=> [3,4]

Iteration ✔ | Transformation ✔ | Accumulation ✖

⬆ back to top

Monoids

Helper functions:

const fold = xs =>
  xs.length
      ? first(xs).concat(fold(rest(xs)))
      : empty

Sum

const _Sum = function(x) { this.val = x }
const Sum = x => new _Sum(x)

_Sum.prototype.concat = y =>
  Sum(this.val + y.val)
_Sum.prototype.empty = () => Sum(0)
const empty = _Sum.prototype.empty()
fold([Sum(1), Sum(2), Sum(3), Sum(4)])
//=> Sum(10)

⬆ back to top

Product

const _Product = function(x) { this.val = x }
const Product = x => new _Product(x)
_Product.prototype.concat = y => Product(this.val * y.val)
_Product.prototype.empty = () => Product(1)
const empty = _Product.prototype.empty()
fold([Product(1), Product(2), Product(3), Product(4)])
//=> Product(24)

⬆ back to top

Max

const _Max = function(x) { this.val = x }
const Max = x => new _Max(x)
_Max.prototype.concat = function(y){
  return Max(this.val > y.val ? this.val : y.val)
}
_Max.prototype.empty = () => Max(-Infinity)
const empty = _Max.prototype.empty()
fold([Max(11), Max(16), Max(3), Max(9)])
//=> Max(16)

⬆ back to top

All

const _All = function(x) { this.val = x }
const All = x => new _All(x)
_All.prototype.concat = function(y){
  return All(this.val && y.val)
}
_All.prototype.empty = () => All(true)
const empty = _All.prototype.empty()
fold([All(false), All(false), All(true), All(false)])
//=> All(false)

⬆ back to top

Any

const _Any = function(x) { this.val = x }
const Any = x => new _Any(x)
_Any.prototype.concat = function(y){
  return Any(this.val || y.val)
}
_Any.prototype.empty = () => Any(false)
const empty = _Any.prototype.empty()
fold([Any(false), Any(false), Any(true), Any(false)]) 
//=> Any(true)

Iteration ✔ | Transformation ✖ | Accumulation ✔

⬆ back to top

F-Algebras

Catamorphism

const cata = (f, xs) =>
  f(xs.map(ys => cata(f,ys)))

⬆ back to top

Sum

Nil.map = f => Nil
_Cons.prototype.map = function(f) {
   return Cons(this.head, f(this.tail))
}
const sum = (x) =>
  (x === Nil) ? 0 : x.head + x.tail
const lst = Cons(2, Cons(3, Cons(4, Nil)));
cata(sum, lst);
//=> 9

⬆ back to top

Map

const map = (f, xs) =>
  cata(x => (x == Nil) ? Nil : Cons(f(x.head), x.tail), xs)
map(x => x + 1, Cons(2, Cons(3, Cons(4, Nil))))
//=> Cons(3, Cons(4, Cons(5, Nil)))
Empty.map = f => Empty
_Leaf.prototype.map = function(f) {
  return Leaf(this.x)
}
_Node.prototype.map = function(f) {
  return Node(f(this.left), this.x, f(this.right))
}
const tr = Node(Node(Leaf(2), 1, Leaf(3)), 0, Leaf(4))
cata(t =>
  t.constructor === _Node
    ? t.left + t.x + t.right
    : t.constructor === _Leaf
      ? t.x
      : 0
, tr)
//=> 10

⬆ back to top

Anamorphism

const ana = (g, a) => g(a).map(x => ana(g, x))

⬆ back to top

arrToList

const arrToList = xs =>
  xs.length === 0 ? Nil : Cons(first(xs), rest(xs))
ana(arrToList, [1, 2, 3, 4, 5])
//=> Cons(1, Cons(2, Cons(3, Cons(4, Cons(5, Nil)))))

⬆ back to top

makeAlphabet

const makeAlphabet = x =>
  x > 25
    ? Nil
    : Cons(String.fromCharCode(x + 65), x + 1)
ana(makeAlphabet, 0)
//=> Cons(A, Cons(B, Cons(C, Cons(D, Cons(E, Cons(F, Cons(G, Cons(H...

⬆ back to top

range

const range = (acc, count) =>
  ana(x => (x >= count) ? Nil : Cons(x, x + 1), acc)
range(2, 10)
//=> Cons(2, Cons(3, Cons(4, Cons(5, Cons(6, Cons(7, Cons(8, Cons(9,  Nil))))))))

⬆ back to top

Real-world examples

const _Const = function(val) { this.val = val }
const Const = x => new _Const(x)
const _Add = function(x, y) {
  this.x = x;
  this.y = y;
}
const Add = (x, y) => new _Add(x, y)
const _Mul = function(x, y) { 
  this.x = x
  this.y = y
}
const Mul = (x, y) => new _Mul(x, y)

_Const.prototype.map = function(f) { return this }
_Add.prototype.map = function(f) {
  return Add(f(this.x), f(this.y))
}
_Mul.prototype.map = function(f) {
  return Mul(f(this.x), f(this.y))
}

const interpret = a =>
  a.constructor === _Mul
    ? a.x * a.y
    : a.constructor === _Add
      ? a.x + a.y
      : /* a.constructor === _Const */ a.val
const program = Mul(Add(Const(2), Const(3)), Const(4))
cata(interpret, program);
//=> 20
const _Concat = function(v, next) {
  this.val = v;
  this.next = next;
}
const Concat = (v, x) => new _Concat(v, x)
const _Replace = function(v, x, next) { 
  this.val = v;
  this.x = x;
  this.next = next;
}
const Replace = (v, x, nt) => new _Replace(v, x, nt)
const _Input = function(v) { this.val = v }
const Input = v => new _Input(v)

_Concat.prototype.map = function(f) {
  return Concat(this.val, f(this.next))
}
_Replace.prototype.map = function(f) {
  return Replace(this.val, this.x, f(this.next))
}
_Input.prototype.map = function(f) {
  return Input(this.val)
}

const interpret = t =>
  t.constructor === _Concat
    ? t.next.concat(t.val)
    : t.constructor === _Replace
      ? t.next.replace(t.val, t.x)
      : /* t.constructor === _Input */ t.val
const prog = Concat("world", Replace("h", "m", Input("hello")))
cata(interpret, prog)
//=> melloworld

const interpret1 = t =>
  t.constructor === _Concat
    ? "concatting "+t.val+" after "+t.next
    : t.constructor === _Replace
      ? "replacing "+t.val+" with "+t.x+" on "+t.next
      : /* t.constructor === _Input */ t.val
const prog = Concat("world", Replace("h", "m", Input("hello")))
cata(interpret1, prog)
//=> concatting world after replacing h with m on hello

⬆ back to top

Iteration ✖ | Transformation ✔ | Accumulation ✔

Inspired by:

License

MIT


Get A Weekly Email With Trending Projects For These Topics
No Spam. Unsubscribe easily at any time.
fp (54
recursion (29

Find Open Source By Browsing 7,000 Topics Across 59 Categories