Functional JavaScript - Composition

How to compose functions in JavaScript using Ramda

Composition is a nice way of reusing existing functions to create new ones. Each of those functions will act on some data that is passed by as an argument, transforming it and then returning it so that the next function in the composition can do the same. This may sound complicated at first, but it’s simpler than it reads.

By the way, this is the second post on the subject of functional programming using JavaScript, first I wrote about curried functions which is an essential technique when doing composition. So I suggest you read it and understand curried functions before reading this post, it will make things a lot easier. If you do know what curried functions are, be my guest and have a good reading.

I think that a good way to understand composition is by comparing it to the “traditional” way of doing things. So let’s start with an example using the imperative way?

Say we have two functions, a function that takes an array and returns the fist element, and a second function that takes a string and returns it all caps.

Like this:

function head(arr) {
  return arr[0];
}

function toUpper(str) {
  return str.toUpperCase();
}

Using those functions in an imperative way it would look like this:

var movies = ['Star Wars', 'The Matrix', 'Forrest Gump'];
var firstMovie = head(movies);
var upperMovie = toUpper(firstMovie);
console.log(upperMovie); // STAR WARS

Or another way we could do to achieve the same result:

var movies = ['Star Wars', 'The Matrix', 'Forrest Gump'];
console.log(toUpper(head(movies)));

Let’s keep in mind that this example is very simple. But in both examples, we have some issues.

In the first example, one problem is that we have a couple of unnecessary variables firstMovie and upperMovie, and also, we have to tell the program what to do every step of the way. Step one, find the first item in the array, step two, transform the text to uppercase, step three, print it.

In the second example, we removed the issue of the unnecessary variables but now, we have a code a little harder to read, because we have to read it from the inside out.

And how would that work with a composition?

First, we will use the two functions already created head and toUpper and we will compose them into a third one. And then we will call that function.

var movies = ['Star Wars', 'The Matrix', 'Forrest Gump'];
var uppercasedFirstMovie = compose(toUpper, head);
console.log(uppercasedFirstMovie(movies));

So, what’s happening here? First, we have a function uppercasedFirstMovie that describes exactly what it’s doing without having any line of code written specifically for it. It is just the result of the combination of two other functions. Don’t worry about the compose function; we’ll talk about that in a moment.

And then we’re logging the result. At this point, one could say that we still have an inside-out way of reading the code, so let’s fix that.

var movies = ['Star Wars', 'The Matrix', 'Forrest Gump'];
var uppercasedFirstMovie = compose(toUpper, head);
var logMovie = compose(console.log, uppercasedFirstMovie);
logMovie(movies);

See? With another composition, a logMovie function was created combination the console.log and the previously created function uppercasedFirstMovie. Now, one just have to call logMovie and pass in the array.

The beauty of this code is that we have small functions that are describing what the code is supposed to do. We’re not telling every step of the way how to do it. That’s the declarative way of programming.

That’s great but what is that compose function? I haven’t shown it yet, so let’s see an implementation of it.

var compose = function(f, g) {
  return function(data) {
    return f(g(data));
  }
};

Compose is a function that takes two other functions as arguments, and return a new function. That new function takes in the data that will be transformed and will call both functions from right to left passing the data to first one and its result to the second.

Why from right to left? Well, because Math. Functional programming is all about applying Math principles to programming, and that’s how it works within Math.

And talking about Math, the idea of composition is all about Math. The idea is: if you a function f that takes A as an argument, and the result is B. Then, you have a function g takes B as an argument, and the result is C, you can say that the result of the composition of g and f when having A as an argument will always result in C.

But don’t take my word for it, check this image:

OK, that’s the Math of it and the theory. Now let’s get back to the code.

The above implementation of compose is a very simple one, that does not take many things into consideration, for instance, what if I’d like to compose more than two functions, like so:

var logMovie = compose(console.log, toUpper, head);

Well with that implementation we can’t. But some libraries can do it. In my previous post on curried functions I talked about Ramda and I’ll be using it in the next examples as well.

So, using Ramda, it’s possible to compose multiple functions:

var movies = ['Star Wars', 'The Matrix', 'Forrest Gump'];
var logMovie = R.compose(console.log, toUpper, head);
logMovie(movies);

Ramda even has the toUpper and head functions, so we wouldn’t have to write those.

var movies = ['Star Wars', 'The Matrix', 'Forrest Gump'];
var logMovie = R.compose(console.log, R.toUpper, R.head);
logMovie(movies);

Now let’s get a little deeper and say we have an array of movie objects:

var movies = [
    {
        title: 'Star Wars',
        director: 'George Lucas'
    },
    {
        title: 'The Matrix',
        director: 'Lana Wachowski'
    },
    {
        title: 'Forrest Gump',
        director: 'Robert Zemeckis'
    }
];

And with that array, I want to return the first name of director from the first movie.

One could write it this way:

var directorsFirstName = R.compose(console.log, R.head, R.split(' '), R.prop('director'), R.head);
directorsFirstName(movies); // George

OK. There’s a lot in this piece of code, so let’s review it step by step.

From right to left:

R.head: will take the first element in the array, in this case: { title: ‘Star Wars’, director: ‘George Lucas’ }.

R.prop(‘director’): this is a Ramda function that will look for the value of a property in a object, in this case, we’re looking for the director property, so the result is: ‘George Lucas.’

R.split: this is also a Ramda function, that will split a string by the text passed in as a parameter, so the result is: [‘George’, ‘Lucas’].

R.head: well, you know that one, the result is: ‘George’.

And finally the result will be logged to the console.

The great thing is, all of this was achieved without even writing a new line of logic. A new function was created by combining all those other ones that already existed.

Another important thing to mention is that those two functions split and prop are curried functions that take two arguments. split takes a separator and a string, and prop takes a property name and an object.

Because of the nature of composition, where the return value of a function will be passed as an argument to the next one, each function in the composition must expect only one argument. So when building the composition we have to partially apply those functions that expect more than one argument. That’s the importance of currying when composing functions.

Another way to write the same composition and maybe making it a little bit more clear to read is this way:

var splitSpace = R.split(' ');
var findDirector = R.prop('director');
var directorsFirstName = R.compose(console.log, R.head, splitSpace, findDirector, R.head);
directorsFirstName(movies); // George

split and prop were partially applied before the composition and assigned to splitSpace and findDirector respectively. Those functions now take only one argument and can be used in the composition. Both examples are equivalent.

To finish, one last example. Let’s find the first name of all directors in all the movies.

Now, instead of retrieving the first item of the array, we’ll be iterating over it. Sounds like a job for map.

Every array has a map function, but the problem with that function is that it will act on an instance of an array. And what we want to do is compose the map function to operate on any instance of a movies array. So, we’ll be using the map function provided by Ramda which is also a curried function.

Here’s how it looks:

var directorsFirstNames = R.compose(console.log, R.map(R.compose(R.head, R.split(' '), R.prop('director'))));

directorsFirstNames(movies);

In the above example, we have two compositions. One that will log the results of the mapping, and one that will be applied to every item in the array.

The idea is the same as before, but let’s split that a bit to make it more readable.

var directorsFirstName = R.compose(R.head, R.split(' '), R.prop('director'));

var logDirectors =  R.compose(console.log, R.map(directorsFirstName));

logDirectors(movies);

Now we have two functions; directorsFirstName will return the first name of a director from a movie object, just as before. This function is passed to the map function which is curried and is now waiting for an array of movies to be executed.

logDirectors is a function that will take the result of the mapping and log it to the console.

Now, when we call logDirectors and pass in the movies array, the map functions has all the arguments it needs to execute so that it will call directorsFirstName for each of its elements.

That’s neat! And again just using the power of combining functions.

And that’s it for today.