Event Signature in .NET -- Using a Strong Typed 'Sender'?

Question

I fully realize that what I am proposing does not follow the .NET guidelines, and, therefore, is probably a poor idea for this reason alone. However, I would like to consider this from two possible perspectives:

(1) Should I consider using this for my own development work, which is 100% for internal purposes.

(2) Is this a concept that the framework designers could consider changing or updating?

I am thinking about using an event signature that utilizes a strong typed 'sender', instead of typing it as 'object', which is the current .NET design pattern. That is, instead of using a standard event signature that looks like this:

class Publisher
{
    public event EventHandler<PublisherEventArgs> SomeEvent;
}

I am considering using an event signature that utilizes a strong-typed 'sender' parameter, as follows:

First, define a "StrongTypedEventHandler":

[SerializableAttribute]
public delegate void StrongTypedEventHandler<TSender, TEventArgs>(
    TSender sender,
    TEventArgs e
)
where TEventArgs : EventArgs;

This is not all that different from an Action<TSender, TEventArgs>, but by making use of the StrongTypedEventHandler, we enforce that the TEventArgs derives from System.EventArgs.

Next, as an example, we can make use of the StrongTypedEventHandler in a publishing class as follows:

class Publisher
{
    public event StrongTypedEventHandler<Publisher, PublisherEventArgs> SomeEvent;

    protected void OnSomeEvent()
    {
        if (SomeEvent != null)
        {
            SomeEvent(this, new PublisherEventArgs(...));
        }
    }
}

The above arrangement would enable subscribers to utilize a strong-typed event handler that did not require casting:

class Subscriber
{
    void SomeEventHandler(Publisher sender, PublisherEventArgs e)
    {     
        if (sender.Name == "John Smith")
        {
            // ...
        }
    }
}

I do fully realize that this breaks with the standard .NET event-handling pattern; however, keep in mind that contravariance would enable a subscriber to use a traditional event handling signature if desired:

class Subscriber
{
    void SomeEventHandler(object sender, PublisherEventArgs e)
    {     
        if (((Publisher)sender).Name == "John Smith")
        {
            // ...
        }
    }
}

That is, if an event handler needed to subscribe to events from disparate (or perhaps unknown) object types, the handler could type the 'sender' parameter as 'object' in order to handle the full breadth of potential sender objects.

Other than breaking convention (which is something that I do not take lightly, believe me) I cannot think of any downsides to this.

There may be some CLS compliance issues here. This does run in Visual Basic .NET 2008 100% fine (I've tested), but I believe that the older versions of Visual Basic .NET through 2005 do not have delegate covariance and contravariance. [Edit: I have since tested this, and it is confirmed: VB.NET 2005 and below cannot handle this, but VB.NET 2008 is 100% fine. See "Edit #2", below.] There may be other .NET languages that also have a problem with this, I can't be sure.

But I do not see myself developing for any language other than C# or Visual Basic .NET, and I do not mind restricting it to C# and VB.NET for .NET Framework 3.0 and above. (I could not imagine going back to 2.0 at this point, to be honest.)

Can anyone else think of a problem with this? Or does this simply break with convention so much that it makes people's stomachs turn?

Here are some related links that I've found:

(1) Event Design Guidelines [MSDN 3.5]

(2) C# simple Event Raising - using “sender” vs. custom EventArgs [StackOverflow 2009]

(3) Event signature pattern in .net [StackOverflow 2008]

I am interested in anyone's and everyone's opinion on this...

Thanks in advance,

Mike

Edit #1: This is in response to Tommy Carlier's post :

Here's a full working example that shows that both strong-typed event handlers and the current standard event handlers that use a 'object sender' parameter can co-exist with this approach. You can copy-paste in the code and give it a run:

namespace csScrap.GenericEventHandling
{
    class PublisherEventArgs : EventArgs
    {
        // ...
    }

    [SerializableAttribute]
    public delegate void StrongTypedEventHandler<TSender, TEventArgs>(
        TSender sender,
        TEventArgs e
    )
    where TEventArgs : EventArgs;

    class Publisher
    {
        public event StrongTypedEventHandler<Publisher, PublisherEventArgs> SomeEvent;

        public void OnSomeEvent()
        {
            if (SomeEvent != null)
            {
                SomeEvent(this, new PublisherEventArgs());
            }
        }
    }

    class StrongTypedSubscriber
    {
        public void SomeEventHandler(Publisher sender, PublisherEventArgs e)
        {
            MessageBox.Show("StrongTypedSubscriber.SomeEventHandler called.");
        }
    }

    class TraditionalSubscriber
    {
        public void SomeEventHandler(object sender, PublisherEventArgs e)
        {
            MessageBox.Show("TraditionalSubscriber.SomeEventHandler called.");
        }
    }

    class Tester
    {
        public static void Main()
        {
            Publisher publisher = new Publisher();

            StrongTypedSubscriber strongTypedSubscriber = new StrongTypedSubscriber();
            TraditionalSubscriber traditionalSubscriber = new TraditionalSubscriber();

            publisher.SomeEvent += strongTypedSubscriber.SomeEventHandler;
            publisher.SomeEvent += traditionalSubscriber.SomeEventHandler;

            publisher.OnSomeEvent();
        }
    }
}

Edit #2: This is in response to Andrew Hare's statement regarding covariance and contravariance and how it applies here. Delegates in the C# language have had covariance and contravariance for so long that it just feels "intrinsic", but it's not. It might even be something that is enabled in the CLR, I don't know, but Visual Basic .NET did not get covariance and contravariance capability for its delegates until the .NET Framework 3.0 (VB.NET 2008). And as a result, Visual Basic.NET for .NET 2.0 and below would not be able to utilize this approach.

For example, the above example can be translated into VB.NET as follows:

Namespace GenericEventHandling
    Class PublisherEventArgs
        Inherits EventArgs
        ' ...
        ' ...
    End Class

    <SerializableAttribute()> _
    Public Delegate Sub StrongTypedEventHandler(Of TSender, TEventArgs As EventArgs) _
        (ByVal sender As TSender, ByVal e As TEventArgs)

    Class Publisher
        Public Event SomeEvent As StrongTypedEventHandler(Of Publisher, PublisherEventArgs)

        Public Sub OnSomeEvent()
            RaiseEvent SomeEvent(Me, New PublisherEventArgs)
        End Sub
    End Class

    Class StrongTypedSubscriber
        Public Sub SomeEventHandler(ByVal sender As Publisher, ByVal e As PublisherEventArgs)
            MessageBox.Show("StrongTypedSubscriber.SomeEventHandler called.")
        End Sub
    End Class

    Class TraditionalSubscriber
        Public Sub SomeEventHandler(ByVal sender As Object, ByVal e As PublisherEventArgs)
            MessageBox.Show("TraditionalSubscriber.SomeEventHandler called.")
        End Sub
    End Class

    Class Tester
        Public Shared Sub Main()
            Dim publisher As Publisher = New Publisher

            Dim strongTypedSubscriber As StrongTypedSubscriber = New StrongTypedSubscriber
            Dim traditionalSubscriber As TraditionalSubscriber = New TraditionalSubscriber

            AddHandler publisher.SomeEvent, AddressOf strongTypedSubscriber.SomeEventHandler
            AddHandler publisher.SomeEvent, AddressOf traditionalSubscriber.SomeEventHandler

            publisher.OnSomeEvent()
        End Sub
    End Class
End Namespace

VB.NET 2008 can run it 100% fine. But I've now tested it on VB.NET 2005, just to be sure, and it does not compile, stating:

Method 'Public Sub SomeEventHandler(sender As Object, e As vbGenericEventHandling.GenericEventHandling.PublisherEventArgs)' does not have the same signature as delegate 'Delegate Sub StrongTypedEventHandler(Of TSender, TEventArgs As System.EventArgs)(sender As Publisher, e As PublisherEventArgs)'

Basically, delegates are invariant in VB.NET versions 2005 and below. I actually thought of this idea a couple of years ago, but VB.NET's inability to deal with this bothered me... But I've now moved solidly to C#, and VB.NET can now handle it, so, well, hence this post.

Edit: Update #3

Ok, I have been using this quite successfully for a while now. It really is a nice system. I decided to name my "StrongTypedEventHandler" as "GenericEventHandler", defined as follows:

[SerializableAttribute]
public delegate void GenericEventHandler<TSender, TEventArgs>(
    TSender sender,
    TEventArgs e
)
where TEventArgs : EventArgs;

Other than this renaming, I implemented it exactly as discussed above.

It does trip over FxCop rule CA1009, which states:

"By convention, .NET events have two parameters that specify the event sender and event data. Event handler signatures should follow this form: void MyEventHandler( object sender, EventArgs e). The 'sender' parameter is always of type System.Object, even if it is possible to employ a more specific type. The 'e' parameter is always of type System.EventArgs. Events that do not provide event data should use the System.EventHandler delegate type. Event handlers return void so that they can send each event to multiple target methods. Any value returned by a target would be lost after the first call."

Of course, we know all this, and are breaking the rules anyway. (All event handlers can use the standard 'object Sender' in their signature if preferred in any case -- this is a non-breaking change.)

So the use of a SuppressMessageAttribute does the trick:

[SuppressMessage("Microsoft.Design", "CA1009:DeclareEventHandlersCorrectly",
    Justification = "Using strong-typed GenericEventHandler<TSender, TEventArgs> event handler pattern.")]

I hope that this approach becomes the standard at some point in the future. It really works very nicely.

Thanks for all your opinions guys, I really appreciate it...

Mike

Answer 1

I take issue with the following statements:

• I believe that the older versions of Visual Basic .NET through 2005 do not have delegate covariance and contravariance.
• I do fully realize that this verges on blasphemy.

First of all, nothing you have done here has anything to do with covariance or contravariance. (Edit: The previous statement is wrong, for more information please see Covariance and Contravariance in Delegates) This solution will work just fine in all CLR versions 2.0 and up (obviously this will not work in a CLR 1.0 application as it uses generics).

Secondly, I strongly disagree that your idea verges on "blasphemy" as this is a wonderful idea.

Answer 2

What you're proposing does make alot of sense actually, and I just wonder if this is one of those things that's simply the way it is because it was originally designed before generics, or if there's a real reason for this.

Answer 3

I don't think there's anything wrong with what you want to do. For the most part, I suspect that the object sender parameter remains in order to continue to support pre 2.0 code.

If you really want to make this change for a public API, you might want to consider creating your own base EvenArgs class. Something like this:

public class DataEventArgs<TSender, TData> : EventArgs
{
    private readonly TSender sender, TData data;

    public DataEventArgs(TSender sender, TData data)
    {
        this.sender = sender;
        this.data = data;
    }

    public TSender Sender { get { return sender; } }
    public TData Data { get { return data; } }
}

Then you can declare your events like this

public event EventHandler<DataEventArgs<MyClass, int>> SomeIndexSelected;

And methods like this:

private void HandleSomething(object sender, EventArgs e)

will still be able to subscribe.

EDIT

That last line made me think a bit... You should actually be able to implement what you propose without breaking any outside functionality since the runtime has no problem downcasting parameters. I would still lean toward the DataEventArgs solution (personally). I would do so, however knowing that it is redundant, since the sender is stored in the first parameter and as a property of the event args.

One benefit of sticking with the DataEventArgs is that you can chain events, changing the sender (to represent the last sender) while the EventArgs retains the original sender.

Answer 4

I think it is a great idea and MS might simply not have the time or interest to invest in making this better as for example when they moved from ArrayList to generic based lists.

Answer 5

With the current situation (sender is object), you can easily attach a method to multiple events:

button.Click += ClickHandler;
label.Click += ClickHandler;

void ClickHandler(object sender, EventArgs e) { ... }

If sender would be generic, the target of the click-event would not be of type Button or Label, but of type Control (because the event is defined on Control). So some events on the Button-class would have a target of type Control, others would have other target types.

Answer 6

And how would you handle events on interfaces? Target would be of the interface-type?

Answer 7

Go for it. For non component based code, I often simplify Event signatures to be simply

public event Action<MyEventType> EventName

where MyEventType does not inherit from EventArgs. Why bother, if I never intend to use any of the members of EventArgs.

Answer 8

From what I understand, the "Sender" field is always supposed to refer to the object which holds the event subscription. If I had my druthers, there would also be a field holding information sufficient to unsubscribe an event should it become necessary(*) (consider, for example, a change-logger which subscribes to 'collection-changed' events; it contains two parts, one of which does the actual work and holds the actual data, and the other of which provides a public interface wrapper, the main part could hold a weak reference to the wrapper part. If the wrapper part gets garbage-collected, that would mean there was no longer anybody interested in the data that was being collected, and the change-logger should thus unsubscribe from any event it receives).

Since it's possible that an object may send events on behalf of another object, I can see some potential usefulness for having a "sender" field which is of Object type, and for having the EventArgs-derived field contain a reference to the object which should be acted upon. The usefuless of the "sender" field, however, is probably limited by the fact that there's no clean way for an object to unsubscribe from an unknown sender.

(*) Actually, a cleaner way of handling unsubscriptions would be to have a multicast delegate type for functions which return Boolean; if a function called by such a delegate returns True, the delegate would be patched to remove that object. This would mean that delegates would no longer be truly immutable, but it should be possible to effect such change in thread-safe manner (e.g. by nulling out the object reference and having the multicast delegate code ignore any embedded null object references). Under this scenario, an attempt to publish and event to a disposed object could be handled very cleanly, no matter where the event came from.