Common Patterns for Delegates

Previous

Delegates provide a mechanism that enables software designs involving minimal coupling between components.

One excellent example for this kind of design is LINQ. The LINQ Query Expression Pattern relies on delegates for all of its features. Consider this simple example:

var smallNumbers = numbers.Where(n => n < 10);

This filters the sequence of numbers to only those less than the value 10. The Where method uses a delegate that determines which elements of a sequence pass the filter. When you create a LINQ query, you supply the implementation of the delegate for this specific purpose.

The prototype for the Where method is:

public static IEnumerable<TSource> Where<in TSource> (IEnumerable<TSource> source, Func<TSource, bool> predicate);

This example is repeated with all the methods that are part of LINQ. They all rely on delegates for the code that manages the specific query. This API design pattern is a very powerful one to learn and understand.

This simple example illustrates how delegates require very little coupling between components. You don't need to create a class that derives from a particular base class. You don't need to implement a specific interface. The only requirement is to provide the implementation of one method that is fundamental to the task at hand.

Building Your Own Components with Delegates

Let's build on that example by creating a component using a design that relies on delegates.

Let's define a component that could be used for log messages in a large system. The library components could be used in many different environments, on multiple different platforms. There are a lot of common features in the component that manages the logs. It will need to accept messages from any component in the system. Those messages will have different priorities, which the core component can manage. The messages should have timestamps in their final archived form. For more advanced scenarios, you could filter messages by the source component.

There is one aspect of the feature that will change often: where messages are written. In some environments, they may be written to the error console. In others, a file. Other possibilities include database storage, OS event logs, or other document storage.

There are also combinations of output that might be used in different scenarios. You may want to write messages to the console and to a file.

A design based on delegates will provide a great deal of flexibility, and make it easy to support storage mechanisms that may be added in the future.

Under this design, the primary log component can be a non-virtual, even sealed class. You can plug in any set of delegates to write the messages to different storage media. The built in support for multicast delegates makes it easy to support scenarios where messages must be written to multiple locations (a file, and a console).

A First Implementation

Let's start small: the initial implementation will accept new messages, and write them using any attached delegate. You can start with one delegate that writes messages to the console.

public static class Logger
{
    public static Action<string> WriteMessage;

    public static void LogMessage(string msg)
    {
        WriteMessage(msg);
    }
}

The static class above is the simplest thing that can work. We need to write the single implementation for the method that writes messages to the console:

public static void LogToConsole(string message)
{
    Console.Error.WriteLine(message);
}

Finally, you need to hook up the delegate by attaching it to the WriteMessage delegate declared in the logger:

Logger.WriteMessage += LogToConsole;

Practices

Our sample so far is fairly simple, but it still demonstrates some of the important guidelines for designs involving delegates.

Using the delegate types defined in the Core Framework makes it easier for users to work with the delegates. You don't need to define new types, and developers using your library do not need to learn new, specialized delegate types.

The interfaces used are as minimal and as flexible as possible: To create a new output logger, you must create one method. That method may be a static method, or an instance method. It may have any access.

Formatting Output

Let's make this first version a bit more robust, and then start creating other logging mechanisms.

Next, let's add a few arguments to the LogMessage() method so that your log class creates more structured messages:

// Logger implementation two
public enum Severity
{
    Verbose,
    Trace,
    Information,
    Warning,
    Error,
    Critical
}

public static class Logger
{
    public static Action<string> WriteMessage;

    public static void LogMessage(Severity s, string component, string msg)
    {
        var outputMsg = $"{DateTime.Now}\t{s}\t{component}\t{msg}";
        WriteMessage(outputMsg);
    }
}

Next, let's make use of that Severity argument to filter the messages that are sent to the log's output.

public static class Logger
{
    public static Action<string> WriteMessage;

    public static Severity LogLevel {get;set;} = Severity.Warning;

    public static void LogMessage(Severity s, string component, string msg)
    {
        if (s < LogLevel)
            return;

        var outputMsg = $"{DateTime.Now}\t{s}\t{component}\t{msg}";
        WriteMessage(outputMsg);
    }
}

Practices

You've added new features to the logging infrastructure. Because the logger component is very loosely coupled to any output mechanism, these new features can be added with no impact on any of the code implementing the logger delegate.

As you keep building this, you'll see more examples of how this loose coupling enables greater flexibility in updating parts of the site without any changes to other locations. In fact, in a larger application, the logger output classes might be in a different assembly, and not even need to be rebuilt.

Building a Second Output Engine

The Log component is coming along well. Let's add one more output engine that logs messages to a file. This will be a slightly more involved output engine. It will be a class that encapsulates the file operations, and ensures that the file is always closed after each write. That ensures that all the data is flushed to disk after each message is generated.

Here is that file based logger:

public class FileLogger
{
    private readonly string logPath;
    public FileLogger(string path)
    {
        logPath = path;
        Logger.WriteMessage += LogMessage;
    }

    public void DetachLog() => Logger.WriteMessage -= LogMessage;

    // make sure this can't throw.
    private void LogMessage(string msg)
    {
        try {
            using (var log = File.AppendText(logPath))
            {
                log.WriteLine(msg);
                log.Flush();
            }
        } catch (Exception e)
        {
            // Hmm. Not sure what to do.
            // Logging is failing...
        }
    }
}

Once you've created this class, you can instantiate it and it attaches its LogMessage method to the Logger component:

var file = new FileLogger("log.txt");

These two are not mutually exclusive. You could attach both log methods and generate messages to the console and a file:

var fileOutput = new FileLogger("log.txt");
Logger.WriteMessage += LogToConsole;

Later, even in the same application, you can remove one of the delegates without any other issues to the system:

Logger.WriteMessage -= LogToConsole;

Practices

Now, you've added a second output handler for the logging sub-system. This one needs a bit more infrastructure to correctly support the file system. The delegate is an instance method. It's also a private method. There's no need for greater accessibility because the delegate infrastructure can connect the delegates.

Second, the delegate-based design enables multiple output methods without any extra code. You don't need to build any additional infrastructure to support multiple output methods. They simply become another method on the invocation list.

Pay special attention to the code in the file logging output method. It is coded to ensure that it does not throw any exceptions. While this isn't always strictly necessary, it's often a good practice. If either of the delegate methods throws an exception, the remaining delegates that are on the invocation won't be invoked.

As a last note, the file logger must manage its resources by opening and closing the file on each log message. You could choose to keep the file open and implement IDisposable to close the file when you are completed. Either method has its advantages and disadvantages. Both do create a bit more coupling between the classes.

None of the code in the Logger class would need to be updated in order to support either scenario.

Handling Null Delegates

Finally, let's update the LogMessage method so that it is robust for those cases when no output mechanism is selected. The current implementation will throw a NullReferenceException when the WriteMessage delegate does not have an invocation list attached. You may prefer a design that silently continues when no methods have been attached. This is easy using the null conditional operator, combined with the Delegate.Invoke() method:

public static void LogMessage(string msg)
{
    WriteMessage?.Invoke(msg);
}

The null conditional operator (?.) short-circuits when the left operand (WriteMessage in this case) is null, which means no attempt is made to log a message.

You won't find the Invoke() method listed in the documentation for System.Delegate or System.MulticastDelegate. The compiler generates a type safe Invoke method for any delegate type declared. In this example, that means Invoke takes a single string argument, and has a void return type.

Summary of Practices

You've seen the beginnings of a log component that could be expanded with other writers, and other features. By using delegates in the design these different components are very loosely coupled. This provides several advantages. It's very easy to create new output mechanisms and attach them to the system. These other mechanisms only need one method: the method that writes the log message. It's a design that is very resilient when new features are added. The contract required for any writer is to implement one method. That method could be a static or instance method. It could be public, private, or any other legal access.

The Logger class can make any number of enhancements or changes without introducing breaking changes. Like any class, you cannot modify the public API without the risk of breaking changes. But, because the coupling between the logger and any output engines is only through the delegate, no other types (like interfaces or base classes) are involved. The coupling is as small as possible.

Next