Dude, where is my Kaizen?

Björn Rochel's weblog

I’ve spend some time with the Pub / Sub topic on my own in the past. Although I still like my own implementation, even a year after I’ve originally written it (yes, rare but sometimes that happens), I really like how Jeremy implemented it. The EventAggregator in StoryTeller is one of those examples of how much you can achieve with only a few lines of code. So let’s go diving again

A type who is interested in recieving messages from the EventAggregator in StoryTeller has to implement the IListener<T> interface where T specifies the concrete message / event. A container extension is used for automatic registration of instances at the EventAggregator after instances have been created in the container (more on this later in the post).

public interface IListener<T>
{
      void Handle(T message);
}

The interface to the EventAggregator looks like this.

public interface IEventAggregator
{
      void SendMessage<T>(Action<T> action) where T : class;
      void SendMessage<T>(T message);
      void AddListener(object listener);
      void RemoveListener(object listener);
}

Most of the interface looks familiar to me, except the SendMessage<T>(Action<T> action) method. The EventAggregator implementation in StoryTeller adds a interesting feature to the topic: Using delegates instead of explicit event classes. One of the things Jeremy mentioned in his NDC “Presentation Patterns” talk is that sometimes creating event classes for events felt a bit tedious to him, especially when those events only have signal character and don’t carry any data with them around. IIRIC the varation with delegates instead of event classes implemented in StoryTeller came up in discussion with Glenn Block and the Prism team. However it didn’t make it into Prism in the end. See the difference in usage for yourself:

aggregator.SendMessage(new ScreenClosingMessage(screen));
aggregator.SendMessage<IWantToKnowWhenAScreenClosed>(x => x.ScreenHasClosed(screen));

Below is the code for the EventAggregator. There are some interesting things to notice.

• First of all, the EventBroker doesn’t know about subscriptions for a particular type. It only knows listener objects. Compatible listeners are found on the fly when the event / message is published by iterating over all known listener objects and calling the CallOn extension method. I’ll spare the code for this extension method because all it does is executing an Action<T> only when an object can be cast to the type specified by T.
• Automatic thread-synchronization to the main-thread is applied by using the SynchronizationContext class. Imho, one of the gems in the .NET 2.0 release that more people should be aware of. I think this class is a really great feature for freeing client code from dealing with callback synchronization issues. Just let the framework handle the InvokeRequired stuff for you. No one likes to write that stuff anyway.
• Jeremy likes it functional . I think this is an interesting example how C# 3.0 code can actually differ from an 1.0 implementation. The shown code is really, really dense and focussed by using a lot of the C# 2.0 and 3.0 features like lamda expressions, extension methods and of course generics. Personally I really like this coding style, however a lot of my colleagues don’t. The debugging story differs a lot from the one using classic if and for loops, which isn’t such a problem for the test-first or test-parallel guys, but I can see where this might feel a bit awkward when you’ve relied on the debugger for most of your developing efforts in the past. In my opinion it’s just a matter of personal taste. Just give it a try and make your own opinion . . .

    public class EventAggregator : IEventAggregator
    {
        private readonly SynchronizationContext _context;
        private readonly List<object> _listeners = new List<object>();
        private readonly object _locker = new object();

        public EventAggregator(SynchronizationContext context)
        {
            _context = context;
        }

        #region IEventAggregator Members

        public void SendMessage<T>(Action<T> action) where T : class
        {
            sendAction(() => all().Each(x => x.CallOn(action)));
        }

        public void SendMessage<T>(T message)
        {
            sendAction(() => all().CallOnEach<IListener<T>>(x => { x.Handle(message); }));
        }

        public void AddListener(object listener)
        {
            withinLock(() =>
            {
                if (_listeners.Contains(listener)) return;
                _listeners.Add(listener);
            });
        }

        public void RemoveListener(object listener)
        {
            withinLock(() => _listeners.Remove(listener));
        }

        #endregion

        private object[] all()
        {
            lock (_locker)
            {
                return _listeners.ToArray();
            }
        }

        private void withinLock(Action action)
        {
            lock (_locker)
            {
                action();
            }
        }

        protected virtual void sendAction(Action action)
        {
            _context.Send(state => { action(); }, null);
        }

        public void AddListeners(params object[] listeners)
        {
            foreach (object listener in listeners)
            {
                AddListener(listener);
            }
        }

        public bool HasListener(object listener)
        {
            return _listeners.Contains(listener);
        }

        public void RemoveAllListeners()
        {
            _listeners.Clear();
        }
    }

As I said earlier in this post, instances are registered after they’ve been created by the container. The functionality for this is provided by a TypeInterceptor class. This class is part of the StructureMap API. Once it has been registered every created instances is passed through this interceptor after it has been resolved. If the instance is identified as relevant for the EventAggregator it is automatically registered.

    public class EventAggregatorInterceptor : TypeInterceptor
    {
        #region TypeInterceptor Members

        public object Process(object target, IContext context)
        {
            context.GetInstance<IEventAggregator>().AddListener(target);
            return target;
        }

        public bool MatchesType(Type type)
        {
            return
                type.ImplementsInterfaceTemplate(typeof (IListener<>)) ||
                type.CanBeCastTo(typeof (ITestListener)) ||
                type.CanBeCastTo(typeof (ICloseable));
        }

        #endregion
    }

Conclusion

EventAggregation is one of the standard patterns in modern enterprise applications. I like the elegant way how Jeremy implemented this. However having spend some time with this in the past, I miss a functional part in his implementation: Type based subscriptions. Type based subscriptions can be really handy when you’re implementing PageFlows or state-ful MessageHandlers. Once a message is recieved the type is resolved by the container and the related handler method is called on the newly created type. I had them in my last application and liked them a lot. Maybe StoryTeller goes alternative routes to achive a similar effect. We’ll see as we proceed in the code . . .