C# 事件在幕后如何工作？

Question

I'm using C#, .NET 3.5. I understand how to utilize events, how to declare them in my class, how to hook them from somewhere else, etc. A contrived example:

public class MyList
{
    private List<string> m_Strings = new List<string>();
    public EventHandler<EventArgs> ElementAddedEvent;

    public void Add(string value)
    {
        m_Strings.Add(value);
        if (ElementAddedEvent != null)
            ElementAddedEvent(value, EventArgs.Empty);
    }
}

[TestClass]
public class TestMyList
{
    private bool m_Fired = false;

    [TestMethod]
    public void TestEvents()
    {
        MyList tmp = new MyList();
        tmp.ElementAddedEvent += new EventHandler<EventArgs>(Fired);
        tmp.Add("test");
        Assert.IsTrue(m_Fired);
    }

    private void Fired(object sender, EventArgs args)
    {
        m_Fired = true;
    }
}

However, what I do not understand, is when one declares an event handler

public EventHandler<EventArgs> ElementAddedEvent;

It's never initialized - so what, exactly, is ElementAddedEvent? What does it point to? The following won't work, because the EventHandler is never initialized:

[TestClass]
public class TestMyList
{
    private bool m_Fired = false;

    [TestMethod]
    public void TestEvents()
    {
        EventHandler<EventArgs> somethingHappend;
        somethingHappend += new EventHandler<EventArgs>(Fired);
        somethingHappend(this, EventArgs.Empty);
        Assert.IsTrue(m_Fired);
    }

    private void Fired(object sender, EventArgs args)
    {
        m_Fired = true;
    }
}

I notice that there is an EventHandler.CreateDelegate(...), but all the method signatures suggest this is only used for attaching Delegates to an already existing EventHandler through the typical ElementAddedEvent += new EventHandler(MyMethod).

I'm not sure if what I am trying to do will help... but ultimately I'd like to come up with an abstract parent DataContext in LINQ whose children can register which table Types they want "observed" so I can have events such as BeforeUpdate and AfterUpdate, but specific to types. Something like this:

public class BaseDataContext : DataContext
{
    private static Dictionary<Type, Dictionary<ChangeAction, EventHandler>> m_ObservedTypes = new Dictionary<Type, Dictionary<ChangeAction, EventHandler>>();

    public static void Observe(Type type)
    {
        if (m_ObservedTypes.ContainsKey(type) == false)
        {
            m_ObservedTypes.Add(type, new Dictionary<ChangeAction, EventHandler>());

            EventHandler eventHandler = EventHandler.CreateDelegate(typeof(EventHandler), null, null) as EventHandler;
            m_ObservedTypes[type].Add(ChangeAction.Insert, eventHandler);

            eventHandler = EventHandler.CreateDelegate(typeof(EventHandler), null, null) as EventHandler;
            m_ObservedTypes[type].Add(ChangeAction.Update, eventHandler);

            eventHandler = EventHandler.CreateDelegate(typeof(EventHandler), null, null) as EventHandler;
            m_ObservedTypes[type].Add(ChangeAction.Delete, eventHandler);
        }
    }

    public static Dictionary<Type, Dictionary<ChangeAction, EventHandler>> Events
    {
        get { return m_ObservedTypes; }
    }
}


public class MyClass
{
    public MyClass()
    {
        BaseDataContext.Events[typeof(User)][ChangeAction.Update] += new EventHandler(OnUserUpdate);
    }

    public void OnUserUpdated(object sender, EventArgs args)
    {
        // do something
    }
}

Thinking about this made me realize I don't really understand what's happening under the hod with events - and I would like to understand :)

Answer 1

I've written this up in a fair amount of detail in an article , but here's the summary, assuming you're reasonably happy with delegates themselves:

• An event is just an "add" method and a "remove" method, in the same way that a property is really just a "get" method and a "set" method. (In fact, the CLI allows a "raise/fire" method as well, but C# never generates this.) Metadata describes the event with references to the methods.
• When you declare a field-like event (like your ElementAddedEvent) the compiler generates the methods and a private field (of the same type as the delegate). Within the class, when you refer to ElementAddedEvent you're referring to the field. Outside the class, you're referring to the field.
• When anyone subscribes to an event (with the += operator) that calls the add method. When they unsubscribe (with the -= operator) that calls the remove.

• For field-like events, there's some synchronization but otherwise the add/remove just call Delegate. Combine / Remove to change the value of the auto-generated field. Both of these operations assign to the backing field - remember that delegates are immutable. In other words, the autogenerated code is very much like this:

   // Backing field // The underscores just make it simpler to see what's going on here. // In the rest of your source code for this class, if you refer to // ElementAddedEvent, you're really referring to this field. private EventHandler<EventArgs> __ElementAddedEvent; // Actual event public EventHandler<EventArgs> ElementAddedEvent { add { lock(this) { // Equivalent to __ElementAddedEvent += value; __ElementAddedEvent = Delegate.Combine(__ElementAddedEvent, value); } } remove { lock(this) { // Equivalent to __ElementAddedEvent -= value; __ElementAddedEvent = Delegate.Remove(__ElementAddedEvent, value); } } }

• The initial value of the generated field in your case is null - and it will always become null again if all subscribers are removed, as that is the behaviour of Delegate.Remove.

• If you want a "no-op" handler to subscribe to your event, so as to avoid the nullity check, you can do:

   public EventHandler<EventArgs> ElementAddedEvent = delegate {};

  The delegate {} is just an anonymous method which doesn't care about its parameters and does nothing.

If there's anything that's still unclear, please ask and I'll try to help!

Answer 2

Under the hood, events are just delegates with special calling conventions. (For example, you don't have to check for nullity before raising an event.)

In pseudocode, Event.Invoke() breaks down like this:

If Event Has Listeners Call each listener synchronously on this thread in arbitrary order.

Since events are multicast, they will have zero or more listeners, held in a collection. The CLR will loop through them, calling each in an arbitrary order.

One big caveat to remember is that event handlers execute in the same thread as the event is raised in. It's a common mental error to think of them as spawning a new thread. They do not.