Distinguishing Delegates and Events
Developers that are new to the .NET Core platform often struggle
when deciding between a design based on delegates and a design
based on events. The choice of delegates or events is often difficult, because the two
language features are similar. Events are even built using
the language support for delegates.
They both offer a late binding scenario: they enable scenarios
where a component communicates by calling a method that is only
known at run time. They both support single and multiple subscriber
methods. You may find this referred to as singlecast and multicast
support. They both support similar syntax for adding and removing
handlers. Finally, raising an event and calling a delegate use exactly the same method call syntax. They even both support the same Invoke()
method syntax for use with the ?. operator.
With all those similarities, it is easy to have trouble determining when to use which.
Listening to Events is Optional
The most important consideration in determining which language feature to use is whether or not there must be an attached subscriber. If your code must call the code supplied by the subscriber, you should use a design based on delegates when you need to implement callback. If your code can complete all its work without calling any subscribers, you should use a design based on events.
Consider the examples built during this section. The code you built
using List.Sort() must be given a comparer function in order to
properly sort the elements. LINQ queries must be supplied with delegates
in order to determine what elements to return. Both used a design built
with delegates.
Consider the Progress event. It reports progress on a task.
The task continues to proceed whether or not there are any listeners.
The FileSearcher is another example. It would still search and find
all the files that were sought, even with no event subscribers attached.
UX controls still work correctly, even when there are no subscribers
listening to the events. They both use designs based on events.
Return Values Require Delegates
Another consideration is the method prototype you would want for your delegate method. As you've seen, the delegates used for events all have a void return type. You've also seen that there are idioms to create event handlers that do pass information back to event sources through modifying properties of the event argument object. While these idioms do work, they are not as natural as returning a value from a method.
Notice that these two heuristics may often both be present: If your delegate method returns a value, it will likely impact the algorithm in some way.
Events Have Private Invocation
Classes other than the one in which an event is contained can only add and remove event listeners; only the class containing the event can invoke the event. Events are typically public class members. By comparison, delegates are often passed as parameters and stored as private class members, if they are stored at all.
Event Listeners Often Have Longer Lifetimes
That event listeners have longer lifetimes is a slightly weaker justification. However, you may find that event-based designs are more natural when the event source will be raising events over a long period of time. You can see examples of event-based design for UX controls on many systems. Once you subscribe to an event, the event source may raise events throughout the lifetime of the program. (You can unsubscribe from events when you no longer need them.)
Contrast that with many delegate-based designs, where a delegate is used as an argument to a method, and the delegate is not used after that method returns.
Evaluate Carefully
The above considerations are not hard and fast rules. Instead, they represent guidance that can help you decide which choice is best for your particular usage. Because they are similar, you can even prototype both, and consider which would be more natural to work with. They both handle late binding scenarios well. Use the one that communicates your design the best.
