# Distinguishing Delegates and Events

Previous

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. This is a difficult concept, because the two language features are very 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 runtime. 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. 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.

## Event Listeners Often Have Longer Lifetimes

This 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 this 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.