Loosely defined, a curried function takes a single argument and returns another function that takes the next argument until it can evaluate to a result.

When used in code, that looks like this:

// Let f be a curried function that requires 3 parameters in order to
// evaluate to a result
f(1) // returns a function that takes the 2nd argument
f(1)(2) // returns a function that takes the 3rd argument
f(1)(2)(3) // returns the result since all 3 parameters are satisfied

This has many useful properties.

One advantage is that if want to pass a function to another function with some of its arguments already specified, you do not need to wrap that function in an additional closure.

// When trying to use a first argument of 1 for some function f:
// Instead of
foo((a) => f(1, a));
// You can use
foo(f(1));

The wrapping is done for you.

Implementation in ES6

Turns out implementing this is quite succinct in ES6. With barely any code, you can take advantage of all the benefits of currying.

Most of the heavy lifting required to curry a function can be done by the ES6/Javascript bind() function.

From the bind() documentation:

The bind() method creates a new function that, when called, has its this keyword set to the provided value, with a given sequence of arguments preceding any provided when the new function is called.

func.bind(thisArg[, arg1[, arg2[, …]]])

The part of currying missing from bind() is automatically evaluating the result of the function when enough arguments are provided.

Here we take bind() and add in the additional logic need to make any JavaScript function curried:

function curry(f, ...args) {
  if (args.length >= f.length) {
    return f(...args);
  }
  else {
    return (...next) => curry(f.bind(f, ...args), ...next);
  }
}

This function is so small that (if you dare) it can even be reduced to a single-line of JavaScript:

const curry = (f, ...args) => (args.length >= f.length) ? f(...args) : (...next) => curry(f.bind(f, ...args), ...next);

I suggest using the equivalent and slightly more verbose version above in any real code.

Usage of curry()

This curry() function gives us a lot of flexibility. Not only does it support passing a single argument into the curried function, it also allows you to pass multiple arguments into any stage along the way.

// maximum flexibility
// f = curry((a, b, c) => ...);
f(1)(2)(3);
f(1, 2)(3);
f(1)(2, 3);
f(1, 2, 3);

This makes it possible to do things like this:

const numbers = [1, 2, 3, 4, 5, 6];

const fancy = curry((a, b, c) => a + b / c);
numbers.reduce(fancy(1), 0);
numbers.reduce(fancy(10), 0);
numbers.reduce(fancy(100), 0);

For the sake of comparison, notice how noisy the uncurried version is:

const fancy = (a, b, c) => a + b / c;
numbers.reduce((b, c) => fancy(1, b, c), 0);
numbers.reduce((b, c) => fancy(10, b, c), 0);
numbers.reduce((b, c) => fancy(100, b, c), 0);

Supporting the passing of multiple arguments makes these curried functions compatible with regular JavaScript paradigms and be extremely convenient. Notice that the standard reduce() method does not need to explicitly support curried functions.

Curried functions are infinitely composable:

const f = curry(curry((a, b, c) => a + b * c));
console.log(f(1)(2)(3));

The second argument of curry() allows you to provide initial arguments to the function. This is a bit of syntatic sugar for the following:

// Instead of
foo(curry(f)(1, 2))
// You can use
foo(curry(f, 1, 2))

Implementation without bind()

An alternate, but equivalent way to define the curry() function without using bind():

function curry(f, ...args) {
  if (args.length >= f.length) {
    return f(...args);
  }
  else {
    return (...next) => curry(f, ...args, ...next);
  }
}

This is still succinct and reducible to a single line if that is what you desire.

The difference between using bind() and using this method is that with bind(), your functions will be special ES6 bound functions.