Functions

  1. Anonymous functions
  2. Named functions
  3. Bindings, scopes, and closures
  4. Functions of multiple arguments and currying
  5. Partial application
  6. The syntactic sugar for function definitions
  7. Formal parameter is a pattern
  8. Operators are functions
  9. Named and optional arguments
  10. Exercises

In previous chapters we’ve learnt how to use variables and arithmetic functions. Now it’s time to learn how to make our own functions.

→ Anonymous functions

We will start with anonymous functions. The reason is that in OCaml, named functions are simply anonymous functions bound to names, and the special syntax for creating named functions is merely a syntactic sugar.

This is the syntax for anonymous functions: fun <formal parameter> -> <expression>.

Let’s write a simple program using an anonymous function: 

let a = (fun x -> x * x) 3

let () = print_int a

The fun x -> x * x function takes a single argument and squares it. Since we already know that the * operator works with integers, we can infer that this function has type int -> int. In the program it is applied to 3 and the result is bound to a, which is later printed.

→ Named functions

Now, before we learn the syntactic sugar for named functions, let’s create a named function the hard way: 

let square = fun x -> x * x

let a = square 3 (* 9 *)

As you can see, the syntax is exactly the same for variable and function bindings. In OCaml, they are both expressions. Like constants, functions (if not applied to any arguments), evaluate to themselves.

→ Bindings, scopes, and closures

Now let’s examine how functions interact with other bindings. Unlike other expressions, functions contain free variables. A free variable is a variable whose name is not bound to any value.

In our square example, x is a free variable, and the expression in the right-hand side contains no other variables. When that function is applied to an argument, x will be replaced with that argument in the x * x expression and that expression will be evaluated. We can visualize that process like this: 

let square = fun x -> x * x

square 3 =
(fun x -> x * x) 3 =
(fun 3 -> 3 * 3) =
3 * 3 =
9

Of course, if we have previously created any bindings, we can refer to them in our functions. 

let two = 2

let plus_two = fun x -> x + two

let number = plus_two 10 (* number = 12 *)

let () = print_int number

However, what happens if the name two is redefined after the definition of plus_two? Let’s try and see what happens: 

let two = 2

let plus_two = fun x -> x + two

let two = 3

let () = print_int (plus_two 10)

This program will print 12 rather than 13 (i.e. 10 + 2, not 10 + 3).

The reason is that functions capture bound variables from the scope where they were created—forever. This is called lexical scoping. If the name two was bound to 2 at the time we defined plus_two, then plus_two will be fun x -> x + 2, even if we redefine the variable two later.

Some languages use a different approach called dynamic scoping where variable names are resolved when the function is evaluated, so redefining a variable retroactively changes the behaviour of all functions that use it. Dynamic scoping makes reusing variable names a risky thing to do. Since OCaml is lexically scoped, you can reuse variable names safely.

We can rewrite the plus_two function using a let ... in binding instead: 

let plus_two =
  let two = 2 in
  fun x -> x + two

let () = print_int (plus_two 3)

Here, the variable two doesn’t even exist for the rest of the program because it was a local binding only visible to the fun x -> x + two expression, but it continues to exist for the plus_two function.

A function stored together with its environment is called a closure. Since you cannot prevent functions from capturing their environment in OCaml, there is no distinction between functions and closures, and for simplicity they are all called just functions. You should always remember about the effect though, and it has important practical uses.

For example, it can be used to create functions of multiple arguments using only functions of one argument. In fact, this is how functions of multiple arguments work in OCaml. Any function “of two arguments” really takes one argument and produces another function where the first argument is fixed but the second one is free.

→ Functions of multiple arguments and currying

Let’s write a function for calculating the average of two values. 

let average = fun x -> (fun y -> (x +. y) /. 2.0)

Its type is float -> (float -> float). It means that when applied to one argument (like average 2.0), this function creates another function, where x is bound, but y is free—a closure. That new function can be applied to another argument to complete the computation.

The expression let f = average 3.0 will be equivalent to let f = fun y -> (3.0 +. y) /. 2.0), because functions capture bound variables from the scope where they are created. Then you can apply it to something else, like (average 3.0) 4.0.

We used parentheses in float -> (float -> float) and (average 3.0) 4.0 for clarity, but in fact they are not needed.

OCaml and most other functional languages use a convention where arrows associate to the right. The type of the average function can be written float -> float -> float, and it’s assumed to mean float -> (float -> float). Likewise, you can apply that function without any parentheses: let a = average 3.0 4.0.

The process of creating a function “of multiple arguments” from functions of one argument is called currying, after Haskell Curry who was already mentioned in the history section, even though he wasn’t the first to invent it, as it often happens with named laws.

→ Partial application

A big advantage of curried functions is that they make partial application especially easy. In many languages, partial application either needs a special syntax or isn’t possible at all. In languages that use closures and currying, it is very easy: just use only the first argument(s) and omit the rest, and you get a function with first argument(s) fixed that can be applied to different remaining arguments as needed.

To see some real examples, let’s introduce the Printf.printf function that is used for formatted output. The reason we used print_int, print_string and similar in the first chapter is that we pretended that we know nothing about functions with more than one argument until we had a chance to learn about them properly. In practice, people almost always use Printf.printf instead because it’s much more powerful and convenient. Its name means that it belongs to the Printf module that comes with the standard library, but we will discuss modules later, for now let’s consider it just an unusual name.

The type of that function is rather complicated and we will not discuss it right now. Let’s just say that its first argument is a format string. Format string syntax is very similar to that of C and all languages inspired by its printf. The Hello World program could be written: 

let () = Printf.printf "%s\n" "hello world"

When applied to a format string, that function will produce functions of one or more arguments depending on the format string. For example, in let f = Printf.printf "%s %d", f will be a function of type string -> int -> unit.

Now let’s write a simple program using Printf.printf and partial application of it: 

let greet = Printf.printf "Hello %s!\n"

let () = greet "world"

The "Hello %s!\n" string is stored in a closure together with the function that Printf.printf produced from it. A similar idiom in an object oriented language might have been greeter = new Formatter("Hello %s!\n"); greeter.format("world"). As Peter Norvig put it in his Design Patterns for Dynamic Languages talk, objects are data with attached behaviour, while closures are behaviours with attached state data.

One danger of curried functions, however, is that failing to supply enough arguments is not a syntax error, but a valid expression, just not of the type that might be expected. Luckily, since OCaml is statically typed, this kind of errors rarely goes unnoticed and programs fail to compile.

Consider this program: 

let add = fun x -> (fun y -> x + y)

let x = add 3 (* forgot second argument *)

let () = Printf.printf "%d\n" x

It will fail to compile because add 3 expression has type int -> int, while Printf.printf "%d\n" is int -> unit.

→ The syntactic sugar for function definitions

Of course, creating functions by binding anonymous functions to names can quickly get cumbersome, especially when multiple arguments are involved, so OCaml provides syntactic sugar for it.

Let’s rewrite the functions we’ve already written in a simpler way: 

let plus_two x = x + 2

let average x y = (x +. y) /. 2.0

OCaml also supports multiple arguments to the fun keyword too, so anonymous functions of multiple arguments can be easily created: let f = fun x y -> x + y.

→ Formal parameter is a pattern

We have already seen that the left hand side of a let-expression can be not only a name, but any valid pattern, including the wildcard or any valid constant. This also applied to the left hand side of function definitions.

For example, we can create a function that ignores its argument using the wildcard pattern: 

let always_zero _ = 0

let always_one = fun _ -> 1

We can also use the () constant, which is a constant of type unit, as a function argument. Since functions with no arguments cannot exist in OCaml, this is the standard for functions used solely for their side effects: 

let print_hello () = print_endline "hello world"

let () = print_hello ()

The print_hello function in this example has type unit -> unit.

→ Operators are functions

It’s not really true that we haven’t seen functions of multiple arguments in the first chapter, we just pretended that operators are not functions. While in many languages they are indeed special constructs, in most functional languages operators are just functions that can be used in an infix form.

In OCaml, every infix operator can also be used in a prefix form if enclosed in parentheses: 

let a = (+) 2 3
let b = (/.) 5. 2.
let c = (^) "hello " "world"

You can also define your own operators like any other functions using the same parentheses syntax: 

let (^.^) x y = x ^ x ^ y ^ y

let s = "foo" ^.^ "bar" (* foofoobarbar *)

You can also define a prefix operator if you use a name that starts with a tilde: 

let (~*) x = x * x

let a = ~* 2 (* 4 *)

The first character of the operator name determines its associativity and precedence 1.

→ Named and optional arguments

OCaml supports named and optional arguments. Named arguments are preceded with the tilde symbol: 

let greet ~greeting ~name = Printf.printf "%s %s!\n" greeting name

let () = greet ~name:"world" ~greeting:"hello"

If you look at the inferred type of the greet function, you will see that argument labels are embedded in its type: greeting:string -> name:string -> unit. Named arguments can be used in any order as long as you specify the labels, but if arguments come in the same order as they are defined in the function, you can omit the labels: 

let greet ~greeting ~name = Printf.printf "%s %s!\n" greeting name

let () = greet "hello" "world"

The syntax of optional arguments is a bit more complicated. Suppose we want to write a function that takes a string and prints hello <string> by default, but allows us to use a different greeting, for example, “hi” This is how we can do it: 

let greet ?(greeting="hello") name = Printf.printf "%s %s!\n" greeting name

let () = greet "world" ~greeting:"hi"

The inferred type of this greet function will look like this: ?greeting:string -> string -> unit. It is recommended to put optional arguments first, because otherwise you will not be able to omit them.

→ Exercises

Using the Printf.printf function, make a join_strings function that takes two strings and an optional separator argument that defaults to space and joins them. What is the type of that function?

Write a function that has type int -> float -> string using any functions we already encountered.

1This system is rigid, but predictable, which is especially important when you import operators from a module. You can find the precedence table in the manual.