Object Oriented Programming

Object Types

Object oriented constructs in F# are compatible with the OO support in .NET CLR, which implies that F# supports single implementation inheritance (a class can have one base class), multiple interface inheritance (a class can implement several interfaces and an interface can inherit from multiple interfaces), subtyping (an inherited class can be casted to the base class type) and dynamic type tests (it is possible to test whether a value is a value of an inherited class casted to a base type). Finally, all object types share a common base class which is called obj in F# and is an alias to the CLR type System.Object.
F# object types can have fields, constructors, methods and properties (a property is just a syntactic sugar for getter and setter methods). The following example introduces the F# syntax for object types:

type MyCell(n:int) =
let mutable data = n + 1
do printf "Creating MyCell(%d)" n

member x.Data
with get() = data
and set(v) = data <- v

member x.Print() =
printf "Data: %d" data
override x.ToString() =
sprintf "(Data: %d)" data

static member FromInt(n) =
MyCell(n)


Type MyCell has a mutable field called data, a property called Data, an instance method Print, a static method FromInt and the type also contains one overridden method called ToString, which is inherited from the obj type and returns a string representation of the object. Finally, the type has an implicit constructor. Implicit constructors are syntactical feature which allows us to place the constructor code directly inside the type declaration and to write the constructor arguments as part of the type construct. In our example, the constructor initializes the mutable field and prints a string as a side effect. F# also supports explicit constructors that have similar syntax as other members, but these are needed rarely.
In the previous example we implemented a concrete object type (a class), which means that it is possible to create an instance of the type and call its methods in the code. In the next example we will look at declaration of an interface (called abstract object type in F#). As you can see, it is similar to the declaration of a class:

type AnyCell =
abstract Value : int with get, set
abstract Print : unit -> unit

The interesting concept in the F# object oriented support is that it is not needed to explicitly specify whether the object type is abstract (interface), concrete (class) or partially implemented (class with abstract methods), because the F# complier infers this automatically depending on the members of the type. Abstract object types (interfaces) can be implemented by a concrete object type (class) or by an object expression, which will be discussed shortly. When implementing an interface in an object type we use the interface .. with construct and define the members required by the interface. Note that the indentation is significant in the lightweight F# syntax, meaning that the members implementing the interface type have to be indented further:

type ImplementCell(n:int) =
let mutable data = n + 1
interface AnyCell with
member x.Print() = printf "Data: %d" data
member x.Value
with get() = data
and set(v) = data <- v

The type casts supported by F# are upcast, used for casting an object to a base type or to an implemented interface type (written as o :> TargetType), downcast, used for casting back from a base type (written as o :?> TargetType), which throws an exception when the value isn’t a value of the specified type and finally, a dynamic type test (written as o :? TargetType), which tests whether a value can be casted to a specified type.


Object expressions

As already mentioned, abstract types can be also implemented by an object expression. This allows us to implement an abstract type without creating a concrete type and it is particularly useful when you need to return an implementation of a certain interface from afunction or build an implementation on the fly using functions already defined somewhere else in your program. The following example implements the AnyCell type:

> let newCell n =
let data = ref n
{ new AnyCell with
member x.Print() = printf "Data: %d" (!data)
member x.Value
with get() = !data
and set(v) = data:=v };;
val newCell : int -> AnyCell

In this code we created a function that takes an initial value as an argument and returns a cell holding this value. In this example we use one more type of mutable values available in F#, called reference cell, which are similar to a mutable values, but more flexible (the F# compiler doesn’t allow using an ordinary mutable value in this example). A mutable cell is created by a ref function taking an initial value. The value is accessed using a prefix ! operator and can be modified using := operator. When implementing the abstract type, we use a new ... with construct with members implementing the functionality required by the abstract type (an object expression can’t add any members). In this example we need a reference cell to hold the value, so the cell is declared in a function and captured in a closure, which means that it will exist until the returned object will be garbage collected.

Continue

.NET interoperability

The .NET BCL is built in an object oriented way, so the ability to work with existing classes is essential for the interoperability. Many (in fact almost all) of the classes are also mutable, so the eager evaluation and the support for side-effects are two key features when working with any .NET library. The following example demonstrates working with the mutable generic ResizeArray type from the BCL (ResizeArray is an alias for a type System.Collections. Generic.List to avoid a confusion with the F# list type):

> let list = new ResizeArray<_>()
list.Add("hello")
list.Add("world")
Seq.to_list list;;
val it : string list = ["hello"; "world"]

As you can see, we used the underscore symbol when creating an instance of the generic type, because the type inference algorithm in F# can deduce the type argument from the code (in this example it infers that the type argument is string, because the Add method is called with a string as an argument). After creating an instance we used Add method to modify the list and add two new items. Finally, we used a Seq.to_list function to convert the collection to the F# list. As a fully compatible .NET language, F# also provides a way for declaring its own classes (called object types in F#), which are compiled into CLR classes or interfaces and therefore the types can be accessed from any other .NET language as well as used to extend classes written in other .NET languages. This is an important feature that allows accessing complex .NET libraries like Windows Forms or ASP.NET from F#.

Continue

Imperative Programming and Mutable Values

Similarly as ML and OCaml, F# adopts an eager evaluation mechanism, which means that a code written using sequencing operator is executed in the same order in which it is written and expressions given as an arguments to a function are evaluated before calling the function (this mechanism is used in most imperative languages including C#, Java or Python). This makes it semantically reasonable to support imperative programming features in a functional language. As already mentioned, the F# value bindings are by default immutable, so to make a variable mutable the mutable keyword has to be used. Additionally F# supports a few imperative language constructs (like for and while), which are expressions of type unit:


> // Imperative factorial calculation
let n = 10
let mutable res = 1
for n = 2 to n do
res <- res * n
// Return the result
res;;
val it : int = 3628800


The use of the eager evaluation and the ability to use mutable values makes it very easy to interoperate with other .NET languages (that rely on the use of mutable state), which is an important aspect of the F# language. In addition it is also possible to use the mutable keyword for creating a record type with a mutable field.

Arrays

As mentioned earlier, another type of value that can be mutated is .NET array. Arrays can be either created using [| .. |] expressions (in the following example we use it together with a range expression, which initializes an array with a range) or using functions from the Array module, for example Array.create. Similarly to the mutable variables introduced in the previous section, the value of an array element can be modified using the <- operator:

> let arr = [| 1 .. 10 |]
val arr : array
> for i = 0 to 9 do
arr.[i] <- 11 - arr.[i]
(...)
> arr;;
val it : array = [| 10; 9; 8; 7; 6; 5; 4; 3; 2; 1 |]

Continue

Expressions and Variable Scoping

The F# language doesn’t have a different notion of a statement and an expression, which means that every language construct is an expression with a known return type. If the construct performs only a side effect (for example printing to a screen or modifying a global mutable variable or a state of .NET object) and doesn’t return any value then the type of the construct is unit, which is a type with only one possible value (written as “()”). The semicolon symbol (;) is used for sequencing multiple expressions, but the first expression in the sequence should have a unit as a result type. The following example demonstrates how the if construct can be used as an expression in F# (though in the optional F# lightweight syntax, which makes whitespace significant and which we used in the rest of this overview, the semicolon symbol can be omitted):


> let n = 1
let res =
if n = 1 then
printfn "..n is one..";
"one"
else
"something else";;
..n is one..
val res : string = "one"

When this code executes it calls the true branch of the if expression, which first calls a side-effecting function, which prints a string and then returns a string ("one") as the result. The result is then assigned to the res value.
Unlike some languages that allow one variable name to appear only once in the entire function body (e.g. C#) or even treat all variables declared inside the body of a function as a variable with scope of the whole function (e.g. Visual Basic or JavaScript), the scope of F# values is determined by the let binding and it is allowed to hide a value by declaring a value with the same name. The following (slightly esoteric) example demonstrates this:


> let n = 21
let f =
if n < n =" n"> print_int n)
else
let n = n / 2
(fun () -> print_int n)
let n = 0
f ();;
42
val it : unit


In this example, the value n declared inside a branch of the if expression is captured by a function created using the fun keyword, which is returned from the if expression and bound to the value named f. When the f is invoked it indeed uses the value from the scope where it was created, which is 42. In languages, where the variable named n would refer to a value stored globally, it would be rather problematic to write a code like this. Of course, writing a code similar to what I demonstrated in this example isn't a good idea, because it makes the code very difficult to read. There are however situations where hiding a value that is no longer needed in the code is practical.

Continue

Function Composition

One of the most interesting aspects of working with functions in functional programming languages is the possibility to use function composition operator. This means that you can very simply build a function that takes an argument, invokes a first function with this argument and passes the result to a second function. For example, you can compose a function fst, which takes a tuple (containing two elements) and returns the first element in the tuple with a function uppercase, which takes a string and returns it in an uppercase:


> (fst >> String.uppercase) ("Hello world", 123);;
val it : string = "HELLO WORLD"
> let data = [ ("Jim", 1); ("John", 2); ("Jane", 3) ];;
val data : (string * int) list
> data |> List.map (fst >> String.uppercase);;
val it : string list = ["JIM"; "JOHN"; "JANE"]


In the first command, we just compose the functions and call the returned function with a tuple as an argument, however the real advantage of this trick becomes more obvious in the third command, where we use the function composition operator (>>) to build a function that is given as an argument to a map function that we used earlier. The function composition allows us to build a function without explicitly using a lambda function (written using the fun keyword) and when this features are used reasonably it makes the code more compact and keeps it very readable.

Continue

Functions

Finally, the type that gives name to the whole functional programming is a function. In F#, similarly to other functional languages, functions are first-class values, meaning that they can be used in a same way as any other types. They can be given as an argument to other functions or returned from a function as a result (a function that takes function as an argument or returns function as a result is called high-order function) and the function type can be used as a type argument to generic types - you can for example create a list of functions. The important aspect of working with functions in functional languages is the ability to create closures – creating a function that captures some values available in the current stack frame. The following example demonstrates a function that creates and returns a function for adding specified number to an initial integer:

> let createAdder n = (fun arg -> n + arg);;
val createAdder : int -> int -> int
> let add10 = createAdder 10;;
val add10 : int -> int
> add10 32;;
val it : int = 42


In the body of the createAdder function we use a fun keyword to create a new unnamed function (a function constructed in this way is called a lambda function). The type of createAdder (int -> int -> int) denotes that when the function is called with int as an argument, it produces a value of type function (which takes an integer as a parameter and produces an integer as a result). In fact, the previous example could be simplified, because any function taking more arguments is treated as a function that produces a function value when it is given the first argument, which means that the following code snippet has the same behavior. Also note that the types of the function createAdder declared earlier and the type of the function add are the same):

> let add a b = a + b;;
val add : int -> int -> int
> let add10 = add 10;;
val add10 : int -> int

When declaring the function value add10 in this example, we used a function that expects two arguments with just one argument. The result is a function with a fixed value of the first argument which now expects only one (the second) argument. This aspect of working with functions is known as currying. Many functions in the F# library are implemented as high-order functions and functions as an arguments are often used when writing a generic code, that is a code that can work with generic types (like list<'a>, which we discussed earlier). For example standard set of functions for manipulating with list values is demonstrated in the following example:

> let odds = List.filter (fun n -> n%2 <> 0) [1; 2; 3; 4; 5];;
val odds : list = [1; 3; 5]
> let squares = List.map (fun n -> n * n) odds;;
val squares : list = [1; 9; 25]

It is interesting to note that the functions that we used for manipulating with lists are generic (otherwise they wouldn’t be very useful!). The signature of the filter function is ('a -> bool) -> list<'a> -> list<'a>, which means that the function takes list of some type as a second argument and a function that returns a true or false for any value of that type, finally the result type is same as the type of the second argument. In our example we instantiate the generic function with a type argument int, because we’re filtering a list of integers. The signatures of generic functions often tell a lot about the function behavior. When we look at the signature of the map function (('a -> 'b) -> list<'a> -> list<'b>) we can deduce that map calls the function given as a first argument on all the items in the list (given as a second argument) and returns a list containing the results. In the last example we will look at the pipelining operator (|>) and we will also look at one example that demonstrates how currying makes writing the code easier - we will use the add function declared earlier:

> let nums = [1; 2; 3; 4; 5];;
val nums : list
> let odds_plus_ten =
nums
|> List.filter (fun n-> n%2 <> 0)
|> List.map (add 10)
val odds_plus_ten : list = [11; 13; 15];;

Sequences of filter and map function calls are very common and writing it as a single expression would be quite difficult and not very readable. Luckily, the sequencing operator allows us to write the code as a single expression in a more readable order - as you can see in the previous example, the value on the left side of the |> operator is given as a last argument to the function call on the right side, which allows us to write the expression as sequence of ordinary calls, where the state (current list) is passed automatically to all functions. The line with List.map also demonstrates a very common use of currying. We want to add 10 to all numbers in the list, so we call the add function with a single argument, which produces a result of the type we needed - a function that takes an integer as an argument and returns an integer (produced by adding 10) as the result.

Continue

Lists

The types used for storing collections of values are list and array. F# list is a typical linked-list type known from many functional languages – it can be either an empty list (written as []) or a cell containing a value and a reference to the tail, which is itself a list (written as value::tail). It is also possible to write a list using a simplified syntax, which means that you can write [1; 2; 3] instead of 1::2::3::[] (which is exactly the same list written just using the two basic list constructors). Array is a .NET compatible mutable array type, which is stored in a continuous memory location and is therefore very efficient – being a mutable type, array is often used in imperative programming style, which will be discussed later. The following example shows declaration of a list value and an implementation of a recursive function that adds together all elements in the list:


> let nums = [1; 2; 3; 4; 5];;
val nums : list
> let rec sum list =
match list with
| h::tail -> (sum tail) + h
| [] -> 0
val sum : list -> int


Similarly as earlier we declared a recursive function using let rec and inside the body we used pattern matching to test whether the list is an empty list or a list cell. Note that list is a generic type, which means that it can store values of any F# type. The type in our example is list, which means that the declared instance of list contains integers. Functions working with generic types can be restricted to some specific type - for example the sum function above requires a list of integers that can be added (this is inferred by the type inference, because the default type used with the + operator is int). Alternatively, the function can be generic as well, which means that it works with any lists - for example a function that returns the last element in the list doesn’t depend on the type and so it can be generic. The signature of a generic function to return the last element would be last : list<'a> -> 'a. An important feature when writing recursive functions in F# is the support for tail-calls. This means that when the last operation performed by the function is a call to a function (including a recursive call to itself), the runtime drops the current stack frame, because it isn’t needed anymore - the value returned by the called function is a result of the caller. This minimizes a chance for getting a stack overflow exception. The sum function from the previous example can be written using an auxiliary function that uses a tail recursion as following:

> // 'acc' is usually called an 'accumulator' variable
let rec sumAux acc list =
match list with
| h::tail -> sumAux (acc + h) tail
| [] -> acc
val sum : int -> list -> int
> let sum list = sumAux 0 list
val sum : list -> int

Continue

Records

The next data type that we will look at is a record type. It can be viewed as a tuple with named members (in case of record these are called labels), which can be accessed using a dot-notation and as mentioned earlier it is good to use this type when it would be difficult to understand what the members in a tuple represent. One more difference between a record type and a tuple is that records have to be declared in advance using a type construct:


> // Declaration of a record type
    type Product = { Name:string; Price:int };;


> // Constructing a value of the 'Product' type
    let p = { Name="Test"; Price=42; };;
val p : Product


> p.Name;; val it : string = "Test"


> // Creating a copy with different 'Name'
    let p2 = { p with Name="Test2" };;
val p2 : Product



The last command uses an interesting construct - the with keyword. The record types are by default immutable, meaning that the value of the member can’t be modified. Since the members are immutable you will often need to create a copy of the record value with one (or more) modified members. Doing this explicitly by listing all the members would be impractical, because it would make adding a new members very difficult, so F# supports the with keyword to do this.
F# records are in many ways similar to classes and they can be, indeed, viewed as simplified classes. Record types are by default immutable, which also means that F# use a structural comparison when comparing two values of a record type (instead of the default reference comparison used when working with classes) and if you need this behavior (e.g. for storing records as a keys in a dictionary) it is very practical to use them. Also, using a record instead of a class is a good idea in a functional code where you can use the with construct. Exposing a record type in a public interface of the module requires additional care and it is often useful to make the labels available as members, which makes it easier to modify implementation of the type later. This topic will be further discussed in the third part of this article series.

Continue

Discriminated Union

In the next sample we demonstrate working with the discriminated union type. This type is used for representing a data type that store one of several possible options (where the options are well known when writing the code). One common example of data type that can be represented using discriminated unions is an abstract syntax tree (i.e. an expression in some programming language):


> // Declaration of the 'Expr' type
type Expr =
| Binary of string * Expr * Expr
| Variable of string
| Constant of int;;
(...)
> // Create a value 'v' representing 'x + 10'
let v = Binary("+", Variable "x", Constant 10);;
val v : Expr

To work with the values of a discriminated union type, we can again use pattern matching. In this case we use the match language construct, which can be used for testing a value against several possible patterns – in case of the Expr type, the possible options are determined by all identifiers used when declaring the type (these are called constructors), namely Binary, Variable and Constant. The following example declares a function eval, which evaluates the given expression (assuming that getVariableValue is a function that returns a value of variable):

> let rec eval x =
match x with
| Binary(op, l, r) ->
let (lv, rv) = (eval l, eval r)
if (op = "+") then lv + rv
elif (op = "-") then lv - rv
else failwith "Unknonw operator!"
| Variable(var) ->
getVariableValue var
| Constant(n) ->
n;;
val eval : Expr -> int


When declaring a function we can use the let keyword that is used for binding a value to a name. I don’t use a term variable known from other programming languages for a reason that will be explained shortly. When writing a recursive function, we have to explicitly state this using the rec keyword as in the previous example.
Discriminated unions form a perfect complement to the typical object-oriented inheritance structure. In an OO hierarchy the base class declares all methods that are overridden in derived classes, meaning that it is easy to add new type of value (by adding a new inherited class), but adding a new operation requires adding method to all the classes. On the other side, a discriminated union defines all types of values in advance, which means that adding a new function to work with the type is easy, but adding a new type of value (new constructor to the discriminated union) requires modification of all existing functions. This suggests that discriminated unions are usually a better way for implementing a Visitor design pattern in F#.

Continue

F# Data Types Overview

Tuples

The first example demonstrates constructing and deconstructing a tuple type. Tuple is simple type that groups together two or more values of any (possibly different) types, for example int and string:

>let tuple = (42, "Hello world!");; val tuple : int * string > let (num, str) = tuple;; val num : int val str : string

As you can see, the compiler deduced a type of the expression that is present on the right side of the equals sign and the F# interactive printed the type, so we can review it. In this example the type of a first element in a tuple is int and the type of the second element is string. The asterisk denotes that the type is a tuple. Similarly, you can define a tuple with more than three elements, but the type changes with the number of elements in a tuple, which means that tuples can't be used for storing an unknown number of values. This can be done using lists or arrays, which will be discussed later.

The syntax used for deconstructing the value into variables num and str is in general called pattern matching and it is used very often in the F# language – the aim of pattern matching is to allow matching a value against a pattern that specifies different view of the data type – in case of tuple, one view is a single value (of type tuple) and the second view is a pair of two values (of different types). Pattern matching can be used with all standard F# types, most notably with tuples, discriminated unions and record types. In addition, F# also supports generalized pattern matching constructs called active patterns, which are discussed later in this overview.

Tuple types are very handy for returning multiple values from functions, because this removes the need to declare a new class or use references when writing a function that performs some simple operation resulting in more returned values (especially in places where C# uses ref and out parameters). In general, I would recommend using tuples when the function is either simple (like division with remainder), local (meaning that it will not be accessed from a different module or file) or it is likely to be used with pattern matching. For returning more complicated structures it is better to use record types which will be discussed shortly.

Continue

Functional Programming

As already mentioned, F# is a typed functional language, by which I mean that types of all values are determined during the compile-time. However, thanks to the use of a type inference, the types are explicitly specified in the code very rarely as we will see in the following examples. The type inference means that the compiler deduces the type from the code, so for example when calling a function that takes int as an argument and returns string as a result, the compiler can infer the type of the variable where the result is assigned (it has to be string) as well as the type of the variable that is given as an argument (it has to be int). Basic data types (aside from a standard set of primitive numeric and textual types that are present in any .NET language) available in F# are tuple, discriminated union, record, array, list, function and object. In the following quick overview, we will use the F# interactive, which is a tool that compiles and executes the entered text on the fly.
The F# interactive can be either used from Visual Studio or by running the fsi.exe from the F# installation directory. In the whole article series we will also use the F# lightweight syntax, which makes the code white-space sensitive, but simplifies many of the syntactical rules. To enable the lightweight syntax enter the following command in the FSI:

> #light;;

The double semicolon is used as the end of the FSI input and sends the entered text to the compiler. This allows us to enter multiple lines of code in a single command. With a few exceptions (mostly when showing a declaration of a complex type) all the code samples in this article are written as commands for FSI including the double semicolon and the result printed by the FSI. Longer code samples can be entered to FSI as well - just add the double semicolon to terminate the input.

Continue

Introducing F#

In one of the papers about F#, the F# designers gave the following description: "F# is a multi-paradigm .NET language explicitly designed to be an ML suited to the .NET environment. It is rooted in the Core ML design and in particular has a core language largely compatible with OCaml". In other words this means that the syntax of the F# language is similar to ML or OCaml (don’t worry if you don’t know these languages, we’ll look at some examples shortly), but the F# language targets .NET Framework, which means that it can natively work with other .NET components and also that it contains several language extensions to allow smooth integration with the .NET object system.

Another important aspect mentioned in this description is that F# is multi-paradigm language. This means that it tries to take the best from many programming languages from very different worlds. The first paradigm is functional programming (the languages that largely influenced the design of F# in this area are ML, OCaml and other), which has a very long tradition and is becoming more important lately for some very appealing properties, including
the fact that functional code tends to be easier to test and parallelize and is also extensible in a ways where object oriented code makes extending difficult.

The second paradigm is widely adopted object oriented programming, which enables interoperability with other .NET languages. In F# it is often used for implementing elementary data types (meaning that the operations on the type are well known and change very rarely), for grouping a set of elementary functions that are together used to perform some complicated operation (i.e. implementing an interface) and also when working with object oriented user
interface frameworks.

Finally, the third paradigm supported by F# is language oriented programming (the design of F# in this area is largely influenced by ML, Haskell and also by LINQ). In general, language oriented programming is focused on developing executors for some code which has a structure of a language (be it a declarative language like XML, or a fully powerful language like some subset of F#). In this overview, I will focus on two techniques provided by F# that allow you to give a different meaning to blocks of F# code. In a programming language theory, this is often called internal domain specific languages, because the code is written in the host language, but is specifically designed as a way for solving problems from some specific domain. An example of such language (and an associated executor) is a block of code that is written as a linear code, but is executed asynchronously (in F# this can be implemented using computation expressions), or a query that is written in F#, but is executed as a SQL code by some database server (this can be implemented using F# quotations).

Continue

Introduction

This text is based on a short overview of the F# language that was included in my bachelor thesis, but I extended it to cover all the important F# aspects and topics. The goal of this article is to introduce all the features in a single (relatively short) text, which means that understanding of a few advanced topics discussed later in the text may require some additional knowledge or previous experience with functional programming.

Anyway, the article still tries to introduce all the interesting views on programming that the F# language provides and the goal is to show that these views are interesting, even though not all of them are fully explained as they would deserve. Of course, this text won’t teach you everything about F#, but it tries to cover the main F# design goals and (hopefully) presents all
the features that make F# interesting and worth learning. In this first part I will shortly introduce F# and the supported paradigms that will be discussed further in the text.

Continue