JavaScript compose functions

Question

I am reading a book which contains the following example:

var composition1 = function(f, g) {
  return function(x) {
    return f(g(x));
  }
};

Then the author writes: "...naive implementation of composition, because it does not take the execution context into account..."

So the preferred function is that one:

var composition2 = function(f, g) {
  return function() {
    return f.call(this, g.apply(this, arguments));
  }
}; 

Followed by an entire example:

var composition2 = function composition2(f, g) {
    return function() {
        return f.call(this, g.apply(this, arguments));
    }
};

var addFour = function addFour(x) {
    return x + 4;
};

var timesSeven = function timesSeven(x) {
    return x * 7;
};

var addFourtimesSeven2 = composition2(timesSeven, addFour);
var result2 = addFourtimesSeven2(2);
console.log(result2);

Could someone please explain to me why the composition2 function is the preferred one (maybe with an example)?

EDIT:

In the meantime i have tried to use methods as arguments as suggested, but it did not work. The result was NaN:

var composition1 = function composition1(f, g) {
    return function(x) {
        return f(g(x));
    };
};

var composition2 = function composition2(f, g) {
    return function() {
        return f.call(this, g.apply(this, arguments));
    }
};

var addFour = {
    myMethod: function addFour(x) {
        return x + this.number;
    },
    number: 4
};

var timesSeven = {
    myMethod: function timesSeven(x) {
        return x * this.number;
    },
    number: 7
};

var addFourtimesSeven1 = composition1(timesSeven.myMethod, addFour.myMethod);
var result1 = addFourtimesSeven1(2);
console.log(result1);

var addFourtimesSeven2 = composition2(timesSeven.myMethod, addFour.myMethod);
var result2 = addFourtimesSeven2(2);
console.log(result2);

Answer 1

This just answers what composition2 actually does:

composition2 is used when you want to keep this as context in the functions itself. The following example shows that the result is 60 by using data.a and data.b :

'use strict';

var multiply = function(value) {
    return value * this.a;
}
var add = function(value) {
    return value + this.b;
}

var data = {
    a: 10,
    b: 4,
    func: composition2(multiply, add)
};

var result = data.func(2);
// uses 'data' as 'this' inside the 'add' and 'multiply' functions
// (2 + 4) * 10 = 60

But yet, it still breaks the following example (unfortunately):

'use strict';

function Foo() {
    this.a = 10;
    this.b = 4;
}
Foo.prototype.multiply = function(value) {
    return value * this.a;
};
Foo.prototype.add = function(value) {
    return value + this.b;
};


var foo = new Foo();

var func = composition2(foo.multiply, foo.add);
var result = func(2); // Uncaught TypeError: Cannot read property 'b' of undefined

Because the context of composition2 ( this ) is undefined (and is not called in any other way, such as .apply , .call or obj.func() ), you'd end up with this being undefined in the functions as well.

On the other hand, we can give it another context by using the following code:

'use strict';
var foo = new Foo();

var data = {
    a: 20,
    b: 8,
    func: composition2(foo.multiply, foo.add)
}

var result = data.func(2); 
// uses 'data' as 'this'
// (2 + 8) * 10 = 200 :)

Or by explicitly setting the context:

'use strict';

var multiply = function(value) {
    return value * this.a;
};
var add = function(value) {
    return value + this.b;
};


var a = 20;
var b = 8;

var func = composition2(multiply, add);

// All the same
var result1 = this.func(2);
var result2 = func.call(this, 2);
var result3 = func.apply(this, [2]);

Answer 2

composition1 would not pass arguments other than the first to g()

If you do:

var composition1 = function(f, g) {
  return function(x1, x2, x3) {
    return f(g(x1, x2, x3));
  }
};

the function will work for the first three arguments. If you however want it to work for an arbitrary number, you need to use Function.prototype.apply .

f.call(...) is used to set this as shown in Caramiriel's answer.

Answer 3

I disagree with the author.

Think of the use-case for function-composition. Most of the time I utilize function-composition for transformer-functions (pure functions; argument(s) in, result out and this is irrelevant).

2nd. Utilizing arguments the way he does it leads into a bad practice/dead end, because it implies that the function g() might depend on multiple arguments.

That means, that the composition I create is not composable anymore, because it might not get all arguments it needs.
composition that prevents composition; fail

(And as a side-effect: passing the arguments-object to any other function is a performance no-go, because the JS-engine can't optimize this anymore)

Take a look at the topic of partial application , usually misreferenced as currying in JS, wich is basically: unless all arguments are passed, the function returns another function that takes the remaining args; until I have all my arguments I need to process them.

Then you should rethink the way you implement argument-order, because this works best when you define them as configs-first, data-last.
Example:

//a transformer: value in, lowercased string out
var toLowerCase = function(str){
    return String(str).toLowerCase();
}

//the original function expects 3 arguments, 
//two configs and the data to process.
var replace = curry(function(needle, heystack, str){
    return String(str).replace(needle, heystack);
});

//now I pass a partially applied function to map() that only 
//needs the data to process; this is really composable
arr.map( replace(/\s[A-Z]/g, toLowerCase) );

//or I create another utility by only applying the first argument
var replaceWhitespaceWith = replace(/\s+/g);
//and pass the remaining configs later
arr.map( replaceWhitespaceWith("-") );

A slightly different approach is to create functions that are, by design, not intended to get all arguments passed in one step, but one by one (or in meaningful groups)

var prepend = a => b => String(a) + String(b);  //one by one
var substr = (from, to) => value => String(str).substr(from, to);  //or grouped

arr.map( compose( prepend("foo"), substr(0, 5) ) );
arr.map( compose( prepend("bar"), substr(5) ) );
//and the `to`-argument is undefined; by intent

I don't intend to ever call such functions with all the arguments, all I want to pass them is their configs, and to get a function that does the job on the passed data/value.

Instead of substr(0, 5, someString) , I would always write someString.substr(0, 5) , so why take any efforts to make the last argument (data) applyable in the first call?