[c#] event Action<> vs event EventHandler<>

Is there any different between declaring event Action<> and event EventHandler<>.

Assuming it doesn't matter what object actually raised an event.

for example:

public event Action<bool, int, Blah> DiagnosticsEvent;

vs

public event EventHandler<DiagnosticsArgs> DiagnosticsEvent;

class DiagnosticsArgs : EventArgs
{
    public DiagnosticsArgs(bool b, int i, Blah bl)
    {...}
    ...
}

usage would be almost the same in both cases:

obj.DiagnosticsEvent += HandleDiagnosticsEvent;

There are several things that I don’t like about event EventHandler<> pattern:

• Extra type declaration derived from EventArgs
• Compulsory passing of object source – often no one cares

More code means more code to maintain without any clear advantage.

As a result, I prefer event Action<>

However, only if there are too many type arguments in Action<>, then an extra class would be required.

Looking at Standard .NET event patterns we find

The standard signature for a .NET event delegate is:

void OnEventRaised(object sender, EventArgs args);

[...]

The argument list contains two arguments: the sender, and the event arguments. The compile time type of sender is System.Object, even though you likely know a more derived type that would always be correct. By convention, use object.

Below on same page we find an example of the typical event definition which is something like

public event EventHandler<EventArgs> EventName;

Had we defined

class MyClass
{
  public event Action<MyClass, EventArgs> EventName;
}

the handler could have been

void OnEventRaised(MyClass sender, EventArgs args);

where sender has the correct (more derived) type.

The advantage of a wordier approach comes when your code is inside a 300,000 line project.

Using the action, as you have, there is no way to tell me what bool, int, and Blah are. If your action passed an object that defined the parameters then ok.

Using an EventHandler that wanted an EventArgs and if you would complete your DiagnosticsArgs example with getters for the properties that commented their purpose then you application would be more understandable. Also, please comment or fully name the arguments in the DiagnosticsArgs constructor.

On the most part, I'd say follow the pattern. I have deviated from it, but very rarely, and for specific reasons. In the case in point, the biggest issue I'd have is that I'd probably still use an Action<SomeObjectType>, allowing me to add extra properties later, and to use the occasional 2-way property (think Handled, or other feedback-events where the subscriber needs to to set a property on the event object). And once you've started down that line, you might as well use EventHandler<T> for some T.

If you follow the standard event pattern, then you can add an extension method to make the checking of event firing safer/easier. (i.e. the following code adds an extension method called SafeFire() which does the null check, as well as (obviously) copying the event into a separate variable to be safe from the usual null race-condition that can affect events.)

(Although I am in kind of two minds whether you should be using extension methods on null objects...)

public static class EventFirer
{
    public static void SafeFire<TEventArgs>(this EventHandler<TEventArgs> theEvent, object obj, TEventArgs theEventArgs)
        where TEventArgs : EventArgs
    {
        if (theEvent != null)
            theEvent(obj, theEventArgs);
    }
}

class MyEventArgs : EventArgs
{
    // Blah, blah, blah...
}

class UseSafeEventFirer
{
    event EventHandler<MyEventArgs> MyEvent;

    void DemoSafeFire()
    {
        MyEvent.SafeFire(this, new MyEventArgs());
    }

    static void Main(string[] args)
    {
        var x = new UseSafeEventFirer();

        Console.WriteLine("Null:");
        x.DemoSafeFire();

        Console.WriteLine();

        x.MyEvent += delegate { Console.WriteLine("Hello, World!"); };
        Console.WriteLine("Not null:");
        x.DemoSafeFire();
    }
}

Based on some of the previous answers, I'm going to break my answer down into three areas.

First, physical limitations of using Action<T1, T2, T2... > vs using a derived class of EventArgs. There are three: First, if you change the number or types of parameters, every method that subscribes to will have to be changed to conform to the new pattern. If this is a public facing event that 3rd party assemblies will be using, and there is any possiblity that the event args would change, this would be a reason to use a custom class derived from event args for consistencies sake (remember, you COULD still use an Action<MyCustomClass>) Second, using Action<T1, T2, T2... > will prevent you from passing feedback BACK to the calling method unless you have a some kind of object (with a Handled property for instance) that is passed along with the Action. Third, you don't get named parameters, so if you're passing 3 bool's an int, two string's, and a DateTime, you have no idea what the meaning of those values are. As a side note, you can still have a "Fire this event safely method while still using Action<T1, T2, T2... >".

Secondly, consistency implications. If you have a large system you're already working with, it's nearly always better to follow the way the rest of the system is designed unless you have an very good reason not too. If you have publicly facing events that need to be maintained, the ability to substitute derived classes can be important. Keep that in mind.

Thirdly, real life practice, I personally find that I tend to create a lot of one off events for things like property changes that I need to interact with (Particularly when doing MVVM with view models that interact with each other) or where the event has a single parameter. Most of the time these events take on the form of public event Action<[classtype], bool> [PropertyName]Changed; or public event Action SomethingHappened;. In these cases, there are two benefits. First, I get a type for the issuing class. If MyClass declares and is the only class firing the event, I get an explicit instance of MyClass to work with in the event handler. Secondly, for simple events such as property change events, the meaning of the parameters is obvious and stated in the name of the event handler and I don't have to create a myriad of classes for these kinds of events.

I realize that this question is over 10 years old, but it appears to me that not only has the most obvious answer not been addressed, but that maybe its not really clear from the question a good understanding of what goes on under the covers. In addition, there are other questions about late binding and what that means with regards to delegates and lambdas (more on that later).

First to address the 800 lb elephant/gorilla in the room, when to choose event vs Action<T>/Func<T>:

• Use a lambda to execute one statement or method. Use event when you want more of a pub/sub model with multiple statements/lambdas/functions that will execute (this is a major difference right off the bat).
• Use a lambda when you want to compile statements/functions to expression trees. Use delegates/events when you want to participate in more traditional late binding such as used in reflection and COM interop.

As an example of an event, lets wire up a simple and 'standard' set of events using a small console application as follows:

public delegate void FireEvent(int num);

public delegate void FireNiceEvent(object sender, SomeStandardArgs args);

public class SomeStandardArgs : EventArgs
{
    public SomeStandardArgs(string id)
    {
        ID = id;
    }

    public string ID { get; set; }
}

class Program
{
    public static event FireEvent OnFireEvent;

    public static event FireNiceEvent OnFireNiceEvent;


    static void Main(string[] args)
    {
        OnFireEvent += SomeSimpleEvent1;
        OnFireEvent += SomeSimpleEvent2;

        OnFireNiceEvent += SomeStandardEvent1;
        OnFireNiceEvent += SomeStandardEvent2;


        Console.WriteLine("Firing events.....");
        OnFireEvent?.Invoke(3);
        OnFireNiceEvent?.Invoke(null, new SomeStandardArgs("Fred"));

        //Console.WriteLine($"{HeightSensorTypes.Keyence_IL030}:{(int)HeightSensorTypes.Keyence_IL030}");
        Console.ReadLine();
    }

    private static void SomeSimpleEvent1(int num)
    {
        Console.WriteLine($"{nameof(SomeSimpleEvent1)}:{num}");
    }
    private static void SomeSimpleEvent2(int num)
    {
        Console.WriteLine($"{nameof(SomeSimpleEvent2)}:{num}");
    }

    private static void SomeStandardEvent1(object sender, SomeStandardArgs args)
    {

        Console.WriteLine($"{nameof(SomeStandardEvent1)}:{args.ID}");
    }
    private static void SomeStandardEvent2(object sender, SomeStandardArgs args)
    {
        Console.WriteLine($"{nameof(SomeStandardEvent2)}:{args.ID}");
    }
}

The output will look as follows:

If you did the same with Action<int> or Action<object, SomeStandardArgs>, you would only see SomeSimpleEvent2 and SomeStandardEvent2.

So whats going on inside of event?

If we expand out FireNiceEvent, the compiler is actually generating the following (I have omitted some details with respect to thread synchronization that isn't relevant to this discussion):

   private EventHandler<SomeStandardArgs> _OnFireNiceEvent;

    public void add_OnFireNiceEvent(EventHandler<SomeStandardArgs> handler)
    {
        Delegate.Combine(_OnFireNiceEvent, handler);
    }

    public void remove_OnFireNiceEvent(EventHandler<SomeStandardArgs> handler)
    {
        Delegate.Remove(_OnFireNiceEvent, handler);
    }

    public event EventHandler<SomeStandardArgs> OnFireNiceEvent
    {
        add
        {
            add_OnFireNiceEvent(value)
        }
        remove
        {
            remove_OnFireNiceEvent(value)

        }
    }

The compiler generates a private delegate variable which is not visible to the class namespace in which it is generated. That delegate is what is used for subscription management and late binding participation, and the public facing interface is the familiar += and -= operators we have all come to know and love : )

You can customize the code for the add/remove handlers by changing the scope of the FireNiceEvent delegate to protected. This now allows developers to add custom hooks to the hooks, such as logging or security hooks. This really makes for some very powerful features that now allows for customized accessibility to subscription based on user roles, etc. Can you do that with lambdas? (Actually you can by custom compiling expression trees, but that's beyond the scope of this response).

To address a couple of points from some of the responses here:

• There really is no difference in the 'brittleness' between changing the args list in Action<T> and changing the properties in a class derived from EventArgs. Either will not only require a compile change, they will both change a public interface and will require versioning. No difference.

• With respect to which is an industry standard, that depends on where this is being used and why. Action<T> and such is often used in IoC and DI, and event is often used in message routing such as GUI and MQ type frameworks. Note that I said often, not always.

• Delegates have different lifetimes than lambdas. One also has to be aware of capture... not just with closure, but also with the notion of 'look what the cat dragged in'. This does affect memory footprint/lifetime as well as management a.k.a. leaks.

One more thing, something I referenced earlier... the notion of late binding. You will often see this when using framework like LINQ, regarding when a lambda becomes 'live'. That is very different than late binding of a delegate, which can happen more than once (i.e. the lambda is always there, but binding occurs on demand as often as is needed), as opposed to a lambda, which once it occurs, its done -- the magic is gone, and the method(s)/property(ies) will always bind. Something to keep in mind.