Functional programming is a style of programming which models computations as the evaluation of expressions. This article is meant to describe it briefly; however, the best way to understand functional programming is to learn the basics of one of the functional programming languages (learn Haskell).
In functional programming, programs are executed by evaluatingexpressions, in contrast with imperative programming where programs are composed ofstatements which change globalstate when executed. Functional programming typically avoids using mutable state.
Functional programming requires that functions arefirst-class, which means that they are treated like any other values and can be passed as arguments to other functions or be returned as a result of a function. Being first-class also means that it is possible to define and manipulate functions from within other functions. Special attention needs to be given to functions that reference local variables from their scope. If such a function escapes their block after being returned from it, the local variables must be retained in memory, as they might be needed later when the function is called. Often it is difficult to determine statically when those resources can be released, so it is necessary to use automaticmemory management.
Many programming languages support programming in both functional and imperative style but the syntax and facilities of a language are typically optimised for only one of these styles, and social factors like coding conventions and libraries often force the programmer towards one of the styles. Therefore, programming languages may be categorized into functional and imperative ones.
The following table shows which languages support functional programming (by supporting first-class functions) and for which the functional style is the dominant one.
| Language | Closures | Functional |
|---|---|---|
| C | No | No |
| Pascal | No | No |
| C++ | Yes | No |
| Java | Yes | No |
| Modula-3 | Yes | No |
| Python | Yes | No |
| Ruby | Yes | No |
| D (2.0) | Yes | No |
| Ocaml | Yes | Yes |
| Erlang | Yes | Yes |
| Haskell | Yes | Yes |
Higher-order functions (HOFs) are functions that take other functions as their arguments. A basic example of a HOF ismap which takes a function and a list as its arguments, applies the function to all elements of the list, and returns the list of its results. For instance, we can write a function that subtracts 2 from all elements of a list without using loops or recursion:
subtractTwoFromList l = map (\x -> x - 2) l
We can generalize this function to subtract any given number:
subtractFromList l y = map (\x -> x - y) l
The function given tomap then becomes aclosure because\x -> x - y references a local variabley from outside its body.
Higher-order functions are very useful for refactoring code and reduce the amount of repetition. For example, typically mostfor loops can be expressed using maps orfolds. Custom iteration schemes, such as parallel loops, can be easily expressed using HOFs.
Higher-order functions are often used to implementdomain-specific languages embedded in Haskell as combinator libraries.
Higher-order functions can be usually simulated in object-oriented languages by functions that takefunction-objects, also calledfunctors (note thatfunctor in Haskell is an entirely different concept). Variables from the scope of the call can be bound inside the function-object which acts as if it were a closure. This way of simulating HOFs is, however, very verbose and requires declaring a new class each time we want to use a HOF.
Some functional languages allow expressions to yield actions in addition to return values. These actions are calledside effects to emphasize that the return value is the most important outcome of a function (as opposed to the case in imperative programming). Languages that prohibit side effects are calledpure. Even though some functional languages are impure they often contain a pure subset that is also useful as a programming language. It is usually beneficial to write a significant part of a functional program in a purely functional fashion and keep the code involving state and I/O to the minimum as impure code is more prone to errors.
Purely functional programs typically operate onimmutable data. Instead of altering existing values, altered copies are created and the original is preserved. Since the unchanged parts of the structure cannot be modified, they can often be shared between the old and new copies, which saves memory.
Pure computations yield the same value each time they are invoked. This property is calledreferential transparency and makes possible to conductequational reasoning on the code. For instance ify = f x andg = h y y then we should be able to replace the definition ofg withg = h (f x) (f x) and get the same result; only the efficiency might change.
Since pure computations are referentially transparent they can be performed at any time and still yield the same result. This makes it possible to defer the computation of values until they are needed, that is, to compute themlazily.Lazy evaluation avoids unnecessary computations and allows, for example, infinite data structures to be defined and used.
If a pure computation is referentially transparent then so are its sub-computations. This makes it much easier to run those pure sub-computationsin parallel to obtain the result of the entire computation.
Even though purely functional programming is very beneficial, the programmer might want to use features that are not available in pure programs, like efficient mutable arrays or convenient I/O. There are two approaches to this problem.
Side effects in the languageSome functional languages extend their purely functional core with side effects. The programmer must thenalways be careful not to use impure functions in places where only pure functions are expected. Side effects through monadsAnother way of introducing side effects to a pure language is to simulate them usingmonads. While the language remains pure and referentially transparent, monads can provide implicit state by threading it inside them. The compiler does not even have to 'know' about the imperative features because the language itself remains pure, however usually the implementations do 'know' about them due to the efficiency reasons, for instance to provide mutable arrays. Allowing side effects only through monads and keeping the language pure makes it possible to have lazy evaluation that does not conflict with the effects of impure code. Even though lazy expressions can be evaluated in any order, the monad structure forces the effects to occur in the correct order. |
Recursion is heavily used in functional programming as it is the canonical and often the only way toiterate. Functional language implementations will often includetail call optimisation to ensure that heavy recursion does not consume excessive memory.
Functional programming is known to provide better support forstructured programming than imperative programming. To make a program structured it is necessary to developabstractions and split it intocomponents which interface each other with those abstractions. Functional languages aid this by making it easy to create clean and simple abstractions. It is easy, for instance, to abstract out a recurring piece of code by creating a higher-order function, which will make the resulting code more declarative and comprehensible.
Functional programs are often shorter and easier to understand than their imperative counterparts. Since various studies have shown that the average programmer's productivity in terms of lines of code is more or less the same for any programming language, this translates also to higher productivity.