Command pattern implementation

Hi dear reader! As my PhD defense day had passed, I have a lot more time than some weeks ago, so I will be posting more often. Maybe not all the posts will be much interesting, or interesting for many people, but still they will be useful for some. This time I'm going to write about Command pattern implementation, the most simple, indeed.

As a little foreword, I want to say that most of the design patterns are very acceptable in terms of logic, many people are using them every day, but they are not aware that this small techniques or ideas are called patterns and have their specific names. The basic idea behind many patterns is decoupling the components of the system, or just the interacting/communicating objects. So that it is easy to replace them separately, to add new features, to make available dynamic (run-time) decision-making, to avoid compiling the whole code after a small change, etc.

    So command pattern is meant for decoupling the command as a separate executable from the object it is bound to, i.e. treating the command as an object per se, not as an action. The most common examples of this are menus and toolbars, where each item executes a command, and this command is just a separate object bound to item. Besides the decoupling there is one more big plus in this approach - one can bind the command to as many items as wanted (this reminds about flyweight pattern but its another topic). So let's go ahead and start looking into details.

What we should have to look into details is:

• a command - the action to be executed
• an invoker - this can be anything starting from a button, a menu or toolbar item up to some callback manager, or a thread pool.
• a receiver - this should be the object the command acts upon

It's obvious that we need three different objects, thus, three different classes for the above list to exist. As far we have an only requirement from the command : it should be executable. Unless we know how it should be executed, or what it should do we can provide just an interface with a single pure virtual function:

class Command

{

public:

       virtual void Execute() = 0;

};

The receiver should be able to do something, so that the command can act upon it. So let's take a simple class with single method doing something.

class Receiver

{

public:

       void Action()

       {

              std::cout << "Command received" << std::endl;

              // don't forget to include <iostream> header

       }

};

Now we need an invoker to invoke the command. It should know which command to invoke, so should somehow keep a reference to the command. The key point here is that the invoker should not care about how the command will execute, what it should act upon, and what it should do at all. So it uses the provided interface above and just calls the Execute() function as seen below.

class Invoker

{

       Command *m_command;

public:

       Invoker(Command *cmd = 0) : m_command(cmd)

       {

       }

       void SetCommad(Command *cmd)

       {

              m_command = cmd;

       }

       void Invoke()

       {

              if (0 != m_command)

              {

                     m_command->Execute();

              }

       }

};

Note the SetCommand() function which lets to bind another command to the Invoker at run-time, so we are free to decide which command should be invoked by the current invoker.

What else do we need to test the system? A command. Actually we have provided an interface, but we don't have a concrete type to instantiate. So we need a concrete command with Execute() function body defined. Here we go:

class MyCommand : public Command

{

       Receiver *m_receiver;

public:

       MyCommand(Receiver *rcv = 0) : m_receiver(rcv)

       {

       }

       void SetReceiver(Receiver *rcv)

       {

              m_receiver = rcv;

       }

       virtual void Execute()

       {

              if (0 != m_receiver)

              {

                     m_receiver->Action();

              }

       }

};

Let's look what we have in MyCommand. The implementation of Execute() method, which is a simple call to Receiver's Action(). Thus we decoupled the command from the real action it does under hood, binding a receiver to a command. The SetReceiver() function makes it easy to bind another receiver to the command. Actually this would be more meaningful if we had an abstract receiver class (interface with single Action() method) and then derived concrete receivers, which would act to the same command in an intrinsic way, but this does not change the essence.

So now we can test our command.

int main()

{

       Receiver r;

       MyCommand cmd(&r);

       Invoker caller(&cmd);

       caller.Invoke();

       return 0;

}

As we expected the command has been invoked. We've reached our goal. So what we get ? We get a freedom to change/remove the command of the invoker, we can change the receiver, all this at run-time. Now about compilation, if we have our interfaces in one module and we create some new commands in another, we don't need to rebuild the first one to use the new commands. Feel free to improvise on this scheme. This is all for this time, thank you for your time and interest.