Thursday, September 19, 2013

C# Delegate: A closer look

C# Delegate and Multicast

In the previous post, we looked at the C# callback mechanism using delegates, and the syntactic sugar provided by Func and Action.

Can we dig a little into the C# delegate implementation and learn more?

As we know, the delegate is really a function (method) pointer. But it does more - we can assign multiple methods to a delegate. When the delegate is invoked, it calls the methods assigned to it in order.

Consider 2 logging classes which implement their respective log methods:

public class ConsoleClass
  public void LogToConsole( string mesg )
    Console.WriteLine( mesg );

public class FileClass
  public void LogToFile( string mesg )
    string file = "syslog.txt";
    using ( TextWriter f = File.CreateText( file ) ) {
      f.WriteLine( mesg );
The log methods have the same signature. So we can assign them to one delegate and invoke it.

//Test client code
public delegate void SyslogDelegate( string mesg );

private static void TestDelegates()
  ConsoleClass console = new ConsoleClass();
  SyslogDelegate d = console.LogToConsole;
  FileClass file = new FileClass();
  //Add second pointer to the delegate (multicast).
  d += file.LogToFile;

  d.Invoke( LogMesg ); //Shorthand: d( mesg );
As expected, invoking the delegate calls both the console and file log methods in order.

What is the C# delegate?

Looking at some methods/properties the delegate class itself exposes:

int i = 1;
Console.WriteLine( "Methods which were referred to by " +
                    d.GetType().Name + ": " );
foreach ( var delegated in d.GetInvocationList() ) {
    Console.Write( i.ToString() + ". " );
    Console.WriteLine( delegated.Target.GetType().Name + 
                       "->" + delegated.Method.Name );
This prints:

Methods which were referred to by SyslogDelegate:
1. ConsoleClass->LogToConsole
2. FileClass->LogToFile
We can infer from the public method GetInvocationList() of the delegate, that the delegate simply holds a list of classes which enclose the referred methods. If we look at the IL for our SyslogDelegate in the Ildasm disassembler tool:

.class public auto ansi sealed Delegates.SyslogDelegate
       extends [mscorlib]System.MulticastDelegate
} // end of class Delegates.SyslogDelegate
we can see the delegate keyword generates to a class which derives from System.MulticastDelegate.

Looking at the System.MulticastDelegate class in a disassembler and reading MSDN, we can gather a couple of things:
1. The System.MulticastDelegate is something the compiler uses, we cannot normally inherit from it.
2. Since the delegate keeps a reference to the method, at least a part of the class which contains the method is alive as long as the delegate is alive.

The second point implies that as long as our delegate lives, the GC will keep the referred class around. At least if the method the delegate refers to, uses other variables inside the scope the class**. We can test this easily with a test program and a memory profiler.

This has implications wherever delegates are used, especially when dealing with higher level abstractions. For example, we can forget removing long lived events which could cause the referred objects to live for a long time, leading to a memory leak.

Simulating delegate

To understand the C# delegate mechanism better, let us try to write a naive and limited mechanism using reflection.

public class MyDelegate
  public MyDelegate(object target, string method) {
    this.Target = target;
    this.Method = method;
  public object Target { get; set; }
  public string Method { get; set; }
  public void Invoke(Object[] paramArray) {
    MethodInfo m = Target.GetType().GetMethod( Method );
    m.Invoke( Target, paramArray );

public class MyMulticastDelegate
  public List< MyDelegate > InvocationList { get; set; }
  public MyMulticastDelegate()
    InvocationList = new List();
  public void Add( object target, string method )
    InvocationList.Add( new MyDelegate( target, method ) );
  public void Remove()
    if ( InvocationList.Count > 0 ) {
      InvocationList.RemoveAt( InvocationList.Count - 1 );
  public void Invoke( object[] paramArray )
    for ( int i = 0; i < InvocationList.Count(); i++ ) {
      InvocationList[i].Invoke( paramArray );
In the test client code:

MyMulticastDelegate m = new MyMulticastDelegate();
//Add method 'references' to our delegate
m.Add( new ConsoleClass(), "LogToConsole" );
m.Add( new FileClass(), "LogToFile" );   

//Call delegate
m.Invoke( new object[] { LogMesg } );
This would call both the methods like the implementation with real delegates.

Of course, this is a naive implementation. What the C# delegate mechanism provides is:
- The ability to invoke a method at runtime based on its signature. Note that our implementation does not handle return types in any way.
- A type safe way of invoking multicast function pointers. So we get compile time checks (and with Visual Studio as-you-type checking). This is useful, otherwise we can easily cause runtime errors like we could in a reflection based invoke mechanism.

Let us explore closures and delegates in the next post.
Source Code

Side note: Could unsafe pointers and pinning be used to implement a naive delegate like mechanism 'from scratch'?

** If it were a method with no side-effects it could in theory be generated as a static which does not really require the class instance. BTW if the method actually uses variables outside its local scope but in the enclosing class's scope, we have something similar to a closure. More on this in the next post.

No comments:

Post a Comment

Boston, MA, United States