Building the Right Environment to Support AI, Machine Learning and Deep Learning
During the last year, I have written some small DirectX games. One of the problem I've encountered was the transition between game states. For example, when the game start, I was usually displaying an Introduction page on which you could see a menu showing my game options. Then, according to the user choice, I had to start the game, show the high scores or show something else. I've solved my problem by considering all those choices as states.
Below is a diagram showing my game logic built on the State pattern.
Click here for larger image
In the application class (CStatePattern_GameApp), we have a function called Run(), which provides a default message loop. Basically, this function updates the game and then processes all other messages. The Update function simply ask the view to render the current frame.
To avoid overhead, the DoFrame function will check if it needs to redraw something, according to the frame rate, or not. I won't give too much details on the frame rate stuff since this is not the purpose of my article. But if you are interested, all the code related to the frame rate stuff is located in the CStatePattern_GameView class.
Let's get back to the State pattern. In the next section, I will explain what happens when the DoFrame function is called.
There is two main parts in this function. The first thing that gets done is the game update. Usually, this is where you would update all game objects in the current state( ship position, collision detection, etc. )
The second part is the game rendering.
m_GameManager.Draw(pDC /* device context */ );
The game manager holds a pointer to a CGameState object. Therefore, when we call the Update function of the CGameManager object, this one will simply redirect the call to the current state object:
This call represents a big part of the State pattern. As you can see, we are calling the Update function of the current state object, but we are also giving a pointer to ourself in parameter. The reason is that after performing state-specific work, the Update function of the current state object will be able to change the current state of the game.
The declaration above is really important. For example, let's say that in our game, when the user hits the space bar, the ship fires. Basically, the CStatePattern_GameView will receive a key event. The view will simply redirect the event to the game manager. The game manager does not know anything about the game logic, so it redirects this event to the current state object. If PlayState is the current state, the KeyEvent function of that state will check which key was struck. In that case, it's going to fire. If the current state would have been IntroductionState, this event could have had a totally different effect.
The state transitions is another important part of the pattern.
For example, the Update function of the CPlayState could check if the player has any ship left. If it would not be the case, this function would call the ChangeState function of the CGameManager like so:
// singleton pattern pGameManager->ChangeState(CGameOverState::Instance());
The states are not changed by the game manager, but by the state itself.
The declaration above might look strange, but when you think about it, it's just logic. Let's say that the player needs to press F1 to start a game while he is in the introduction page. The introduction page knows what to do when the user presses F1. In that case, it will tell the game manager to change the current state to PlayState.
In my demo project, I simply demonstrate some state transitions. When the game starts, we are in Introduction state. This state will wait for the user to press a key. According to the key pressed, the Introduction state will change the CGameManager's state. F1 will switch to the CPlayState, and F2 will switch to the CHighScoreState.
In my demo project, I have made some TRACE calls to show you the details of each transition.