Click here to Skip to main content
15,569,234 members
Articles / Programming Languages / F#
Posted 3 Sep 2010

Tagged as


24 bookmarked

Basics of Dataflow Programming in F# and C#

Rate me:
Please Sign up or sign in to vote.
4.63/5 (8 votes)
3 Sep 2010CPOL5 min read
This article introduces dataflow programming and provides basic implementations of dataflow variables in C# and F#

Introduction into Dataflow Programming

What is dataflow programming all about? In classical imperative programming, a program is basically a set of operations working with mutable state thus effectively hiding data paths. Dataflow programming is more like a series of workers/robots on an assembly line, who execute only when input material arrives. Imperative programming style can introduce non-determinism in case of concurrent execution (multithreading) without proper synchronization. In dataflow programming, program execution depends on the actual data, or on the data availability to be precise. Dataflow programming yields completely deterministic programs.
Let’s introduce the concept of dataflow variable which is one of the main concepts of dataflow programming. Dataflow variable can have two possible states: bound (assigned a value) or unbound (no value has been yet assigned). Whenever a thread tries to read the value of unbound dataflow variable, it gets blocked until some other thread bounds the variable. Dataflow variable can be bound only once, successive tries to bind the variable will fail. So, what is dataflow programming? With dataflow variable, one can also build blocking queues and streams. Actor model can be implemented using such blocking queues.
Basically, you can get more information on dataflow programming from this Wikipedia article. Also there is a nice article in Groovy GPars guide.

Overview of the Article

This article presents basic implementations of dataflow variable in both C# and F#. Also, this article demonstrates examples of dataflow programming in C# using futures. The best effect of dataflow programming is achieved in programming languages that follow declarative model principles. In our case, C# is imperative language and programming in a dataflow style requires developers to be self-disciplined. Surprisingly, but F# being considered to be a functional programming language, and therefore following declarative programming paradigm, also enables developers to program in an imperative programming way (via mutable keyword). Adding dataflow variables to C# and F# does not make them automatically dataflow programming languages, because there is still no necessary syntactic sugar and language support.
Clojure is one of the most popular modern languages that enable dataflow programming. Clojure supports dataflow programming through premises. It is also possible to do a dataflow programming in other popular languages like Groovy, Scala, Ruby using open-source libraries like GPars for Groovy, but all those languages provide no syntactic support for dataflow variables. As a genuine dataflow programming language, I would distinguish Oz programming language which treats all variables as dataflow variables: reader trying to read an unbound/uninitialized variable will be blocked until variable is bound/initialized. On one hand, it saves us from famous NullReferenceException exceptions, but on the other hand, it can introduce program hangs.

First, I will present implementations in C# and F# and later I will dig into the thread synchronization details.

Dataflow Variables in C#

Let’s start with the simple example of how to use a dataflow variable in C#.

var variable = new DataflowVariable<int>(); //create variable
variable.Bind(value); //bind variable
int value = 1000 + variable;//read variable

C# is not very extensible when it comes to operator overloading (as you later see in F# implementation) and this is the reason we are using Bind method here. Actually this is a matter of taste – whether to use operators when working with dataflow variables or simply properties/functions, but in my opinion, operators look more natural. What I love about C# is implicit conversion operators. Now the code itself:

public class DataflowVariable<t>
        private readonly object syncLock = new object();
        private volatile bool isInitialized = false;
        private volatile object value;

        private T Value
                return (T)value;
                lock (syncLock)
                    if (isInitialized)
                        throw new System.InvalidOperationException
				("Dataflow variable can be set only once.");
                        this.value = value;
                        isInitialized = true;

        public void Bind(T newValue)
            this.Value = newValue;

        public static implicit operator T(DataflowVariable<t> myVar)
            return myVar.Value;

Dataflow Variables in F#

Let’s start with the simple example of how to use a dataflow variable in F#.

let myVar = new DataflowVariable<int>() // create variable
myVar  <~ value //bind variable
let value = (1000 + !!myVar) //read variable

Here we use operator (<~) to bind the dataflow variable and operator (!!) to read its value. Now the code itself:

type public DataflowVariable<'T> () =
            let mutable value : option<'T> = None

            let syncLock = new System.Object()

            member private this.Value
                with get() : 'T =
                    match value with
                    | Some(initializedVal) -> initializedVal
                    | None -> 
                        lock syncLock (fun () -> 
                                            while (value.IsNone) do
                                                ignore (System.Threading.Monitor.Wait
                and set(newVal : 'T) =
                    lock syncLock (fun () ->
                                    match value with
                                    | Some(_) -> invalidOp 
				"Dataflow variable can be set only once."
                                    | None -> 
                                        value <- Some(newVal)
            static member public (<~) (var:DataflowVariable<'T>, initValue:'T) = 
                var.Value <- initValue

            static member public (!!) (var:DataflowVariable<'T>) : 'T =

You may have noticed [<volatilefield>] attribute. As per pretty stingy documentation, this attribute effectively replaces volatile keyword in C#, but I haven’t performed thorough testing to verify it. What? F# has no keyword for volatile fields? And this is as it has to be. Volatile fields belong to the domain of imperative programming and F#, being first of all functional programming language (which is the implementation of declarative model), tries to avoid shared state (remember mutable keyword?). F# does not support overloading of implicit conversion operators, that’s why we need some kind of dereferencing prefix operator (!!). F# implementation is more elegant, because we expose Option type here and thus do not have to deal with isInitialized field as in the case of C# implementation.

Implementation Details and Some Thoughts on Thread Synchronization

For synchronization in both implementations, I have used volatile fields in conjunction with a simple pattern for Monitor.Wait/Monitor.Pulse. You can get more information regarding Monitor.Pulse/Monitor.Wait in this very nice article by Joe Albahari. Volatile fields here are used to prevent instruction reordering and ensure CPU cache synchronization. Also as an option, instead of using volatile field, we could use here Thread.VolatileRead method (we do not need to use also Thread.VolatileWrite because actual write is done in within the lock statement which effectively prevents reordering and flushes and invalidates CPU cache, and anyway Thread.VolatileWrite only flushes the CPU cache but does not invalidate it). Basically, the static VolatileRead and VolatileWrite methods in the Thread class reads/writes a variable while enforcing (technically, a superset of) the guarantees made by the volatile keyword.

Dataflow Programming Examples in C# and F#

In C#, I will demonstrate a simple example of dataflow programming with Parallel extensions library (futures and continuations). Basically using Task.Factory.ContinueWhenAll, one can achieve similar results as with dataflow variables, but dataflow variables provide developers with much more flexibility.

var input1 = new DataflowVariable<int>();
var input2 = new DataflowVariable<int>();
var output1 = new DataflowVariable<int>();
var output2 = new DataflowVariable<int>();

Task<int> task1 = Task.Factory.StartNew<int>(
                () =>
                        output1.Bind(input1 + input2);
                        return output1*10;
Task<int> task = Task.Factory.StartNew<int>(() =>
                                                     output2.Bind(input1 + output1);
                                                     return input1;


Console.WriteLine(10 + output1 + output2);


This article described the basic implementation of dataflow variables in C# and F# programming languages and basic examples of dataflow programming using continuations/futures. Please consider this article as a starting point in a journey into the world of dataflow programming.


  • 3rd September, 2010: Initial post


This article, along with any associated source code and files, is licensed under The Code Project Open License (CPOL)

Written By
Technical Lead bwin Interactive Entertainment AG
Austria Austria
The views expressed in my articles are mine and do not necessarily reflect the views of my employer.

SendMessage(string.Format("{0}@{1}.com", "liptchinski_vit", "yahoo"));

More info in my LinkedIn profile:

Comments and Discussions

Generalnd here's an gui-based approach... Pin
Alexander Golde11-Oct-13 1:36
Alexander Golde11-Oct-13 1:36 
GeneralMy vote of 5 Pin
tec-goblin7-Sep-10 0:52
tec-goblin7-Sep-10 0:52 
Very interesting, I am interested in dataflow programming since I'm working every day with ETL tools like Sql Server Integration Services which do use this kind of programming to get high levels of parallelization.
GeneralMy vote of 4 Pin
Eric Xue (brokensnow)6-Sep-10 12:13
Eric Xue (brokensnow)6-Sep-10 12:13 
GeneralThis reminds me... Pin
Marc Clifton4-Sep-10 10:20
mvaMarc Clifton4-Sep-10 10:20 
GeneralRe: This reminds me... Pin
Vitaliy Liptchinsky4-Sep-10 11:57
Vitaliy Liptchinsky4-Sep-10 11:57 
GeneralRe: This reminds me... Pin
Marc Clifton5-Sep-10 7:41
mvaMarc Clifton5-Sep-10 7:41 
GeneralRe: This reminds me... Pin
Vitaliy Liptchinsky6-Sep-10 2:34
Vitaliy Liptchinsky6-Sep-10 2:34 
GeneralRe: This reminds me... Pin
Marc Clifton6-Sep-10 2:55
mvaMarc Clifton6-Sep-10 2:55 

General General    News News    Suggestion Suggestion    Question Question    Bug Bug    Answer Answer    Joke Joke    Praise Praise    Rant Rant    Admin Admin   

Use Ctrl+Left/Right to switch messages, Ctrl+Up/Down to switch threads, Ctrl+Shift+Left/Right to switch pages.