by Pete Shinners <email@example.com>
Revision 1.1, April 12th, 2002
Pygame version 1.3 comes with a new module, pygame.sprite. This module is written in python and includes some higher level classes to manage your game objects. By using this module to its full potential, you can easily manage and draw your game objects. The sprite classes are very optimized, so it’s likely your game will run faster with the sprite module than without.
The sprite module is also meant to be very generic. It turns out you can use it with nearly any type of gameplay. All this flexibility comes with a slight penalty, it needs a little understanding to properly use it. The reference documentation for the sprite module can keep you running, but you’ll probably need a bit more explanation of how to use :mod:pygame.sprite in your own game.
Several of the pygame examples (like “chimp” and “aliens”) have been updated to use the sprite module. You may want to look into those first to see what this sprite module is all about. The chimp module even has it’s own line-by-line tutorial, which may help get more an understanding of programming with python and pygame.
Note that this introduction will assume you have a bit of experience programming with python, and are somewhat framiliar with the different parts of creating a simple game. In this tutorial the word “reference” is occasionally used. This represents a python variable. Variables in python are references, so you can have several variables all pointing to the same object.
The term “sprite” is a holdover from older computer and game machines. These older boxes were unable to draw and erase normal graphics fast enough for them to work as games. These machines had special hardware to handle game like objects that needed to animate very quickly. These objects were called “sprites” and had special limitations, but could be drawn and updated very fast. They usually existed in special overlay buffers in the video. These days computers have become generally fast enough to handle sprite like objects without dedicated hardware. The term sprite is still used to represent just about anything in a 2D game that is animated.
The sprite module comes with two main classes. The first is Sprite, which should be used as a base class for all your game objects. This class doesn’t really do anything on its own, it just includes several functions to help manage the game object. The other type of class is Group. The Group class is a container for different Sprite objects. There are actually several different types of group classes. Some of the Groups can draw all the elements they contain, for example.
This is all there really is to it. We’ll start with a description of what each type of class does, and then discuss the proper ways to use these two classes.
As mentioned before, the Sprite class is designed to be a base class for all your game objects. You cannot really use it on its own, as it only has several methods to help it work with the different Group classes. The sprite keeps track of which groups it belongs to. The class constructor (__init__ method) takes an argument of a Group (or list of Groups) the Sprite instance should belong to. You can also change the Group membership for the Sprite with the add() and remove() methods. There is also a groups() method, which returns a list of the current groups containing the sprite.
When using the your Sprite classes it’s best to think of them as “valid” or “alive” when they are belonging to one or more Groups. When you remove the instance from all groups pygame will clean up the object. (Unless you have your own references to the instance somewhere else.) The kill() method removes the sprite from all groups it belongs to. This will cleanly delete the sprite object. If you’ve put some little games together, you’ll know sometimes cleanly deleting a game object can be tricky. The sprite also comes with an alive() method, which returns true if it is still a member of any groups.
The Group class is just a simple container. Similar to the sprite, it has an add() and remove() method which can change which sprites belong to the group. You also can pass a sprite or list of sprites to the constructor (__init__() method) to create a Group instance that contains some initial sprites.
The Group has a few other methods like empty() to remove all sprites from the group and copy() which will return a copy of the group with all the same members. Also the has() method will quickly check if the Group contains a sprite or list of sprites.
The other function you will use frequently is the sprites() method. This returns an object that can be looped on to access every sprite the group contains. Currently this is just a list of the sprites, but in later version of python this will likely use iterators for better performance.
As a shortcut, the Group also has an update() method, which will call an update() method on every sprite in the group. Passing the same arguments to each one. Usually in a game you need some function that updates the state of a game object. It’s very easy to call your own methods using the Group.sprites() method, but this is a shortcut that’s used enough to be included. Also note that the base Sprite class has a “dummy” update() method that takes any sort of arguments and does nothing.
Lastly, the Group has a couple other methods that allow you to use it with the builtin len() function, getting the number of sprites it contains, and the “truth” operator, which allows you to do “if mygroup:” to check if the group has any sprites.
At this point the two classes seem pretty basic. Not doing a lot more than you can do with a simple list and your own class of game objects. But there are some big advantages to using the Sprite and Group together. A sprite can belong to as many groups as you want. Remember as soon as it belongs to no groups, it will usually be cleared up (unless you have other “non-group” references to that object).
The first big thing is a fast simple way to categorize sprites. For example, say we had a Pacman-like game. We could make separate groups for the different types of objects in the game. Ghosts, Pac, and Pellets. When Pac eats a power pellet, we can change the state for all ghost objects by effecting everything in the Ghost group. This is quicker and simpler than looping through a list of all the game objects and checking which ones are ghosts.
Adding and removing groups and sprites from each other is a very fast operation, quicker than using lists to store everything. Therefore you can very efficiently change group memberships. Groups can be used to work like simple attributes for each game object. Instead of tracking some attribute like “close_to_player” for a bunch of enemy objects, you could add them to a separate group. Then when you need to access all the enemies that are near the player, you already have a list of them, instead of going through a list of all the enemies, checking for the “close_to_player” flag. Later on your game could add multiple players, and instead of adding more “close_to_player2”, “close_to_player3” attributes, you can easily add them to different groups or each player.
Another important benefit of using the Sprites and Groups, the groups cleanly handle the deleting (or killing) of game objects. In a game where many objects are referencing other objects, sometimes deleting an object can be the hardest part, since it can’t go away until it is not referenced by anyone. Say we have an object that is “chasing” another object. The chaser can keep a simple Group that references the object (or objects) it is chasing. If the object being chased happens to be destroyed, we don’t need to worry about notifying the chaser to stop chasing. The chaser can see for itself that its group is now empty, and perhaps find a new target.
Again, the thing to remember is that adding and removing sprites from groups is a very cheap/fast operation. You may be best off by adding many groups to contain and organize your game objects. Some could even be empty for large portions of the game, there isn’t any penalties for managing your game like this.
The above examples and reasons to use Sprites and Groups are only a tip of the iceberg. Another advantage is that the sprite module comes with several different types of Groups. These groups all work just like a regular old Group, but they also have added functionality (or slightly different functionality). Here’s a list of the Group classes included with the sprite module.
This is the standard “no frills” group mainly explained above. Most of the other Groups are derived from this one, but not all.
This works exactly like the regular Group class, but it only contains the most recently added sprite. Therefore when you add a sprite to this group, it “forgets” about any previous sprites it had. Therefore it always contains only one or zero sprites.
This is a standard group derived from Group. It has a draw() method that draws all the sprites it contains to the screen (or any Surface). For this to work, it requires all sprites it contains to have a “image” and “rect” attributes. It uses these to know what to blit, and where to blit it.
This is derived from the RenderPlain group, and adds a method named clear(). This will erase the previous position of all drawn sprites. It uses a background image to fill in the areas where the sprite were. It is smart enough to handle deleted sprites and properly clear them from the screen when the clear() method is called.
This is the cadillac of rendering Groups. It is inherited from RenderClear, but changes the draw() method to also return a list of pygame Rects, which represent all the areas on screen that have been changed.
That is the list of different groups available We’ll discuss more about these rendering groups in the next section. There’s nothing stopping you from creating your own Group classes as well. They are just python code, so you can inherit from one of these and add/change whatever you want. In the future I hope we can add a couple more Groups to this list. A GroupMulti which is like the GroupSingle, but can hold up to a given number of sprites (in some sort of circular buffer?). Also a super-render group that can clear the position of the old sprites without needing a background image to do it (by grabbing a copy of the screen before blitting). Who knows really, but in the future we can add more useful classes to this list.
From above we can see there are three different rendering groups. We could probably just get away with the RenderUpdates one, but it adds overhead not really needed for something like a scrolling game. So we have a couple tools here, pick the right one for the right job.
For a scrolling type game, where the background completely changes every frame. We obviously don’t need to worry about python’s update rectangles in the call to display.update(). You should definitely go with the RenderPlain group here to manage your rendering.
For games where the background is more stationary, you definitely don’t want pygame updating the entire screen (since it doesn’t need to). This type of game usually involves erasing the old position of each object, then drawing it in a new place for each frame. This way we are only changing what is necessary. Most of the time you will just want to use the RenderUpdates class here. Since you will also want to pass this list of changes to the display.update() function.
The RenderUpdates class also does a good job an minimizing overlapping areas in the list of updated rectangles. If the previous position and current position of an object overlap, it will merge them into a single rectangle. Combine this with the fact that is properly handles deleted objects and this is one powerful Group class. If you’ve written a game that manages the changed rectangles for the objects in a game, you know this the cause for a lot of messy code in your game. Especially once you start to throw in objects that can be deleted at anytime. All this work is reduced down to a clear() and draw() method with this monster class. Plus with the overlap checking, it is likely faster than if you did it yourself.
Also note that there’s nothing stopping you from mixing and matching these render groups in your game. You should definitely use multiple rendering groups when you want to do layering with your sprites. Also if the screen is split into multiple sections, perhaps each section of the screen should use an appropriate render group?
The sprite module also comes with two very generic collision detection functions. For more complex games, these really won’t work for you, but you can easily grab the source code for them, and modify them as needed.
Here’s a summary of what they are, and what they do.
This checks for collisions between a single sprite and the sprites in a group. It requires a “rect” attribute for all the sprites used. It returns a list of all the sprites that overlap with the first sprite. The “dokill” argument is a boolean argument. If it is true, the function will call the kill() method on all the sprites. This means the last reference to each sprite is probably in the returned list. Once the list goes away so do the sprites. A quick example of using this in a loop:
>>> for bomb in sprite.spritecollide(player, bombs, 1): ... boom_sound.play() ... Explosion(bomb, 0)
This finds all the sprites in the “bomb” group that collide with the player. Because of the “dokill” argument it deletes all the crashed bombs. For each bomb that did collide, it plays a “boom” sound effect, and creates a new Explosion where the bomb was. (Note, the Explosion class here knows to add each instance to the appropriate class, so we don’t need to store it in a variable, that last line might feel a little “funny” to you python programmers.
This is similar to the spritecollide function, but a little more complex. It checks for collisions for all the sprites in one group, to the sprites in another. There is a dokill argument for the sprites in each list. When dokill1 is true, the colliding sprites in group1 will be kill()``ed. When ``dokill2 is true, we get the same results for group2. The dictionary it returns works like this; each key in the dictionary is a sprite from group1 that had a collision. The value for that key is a list of the sprites that it collided with. Perhaps another quick code sample explains it best:
>>> for alien in sprite.groupcollide(aliens, shots, 1, 1).keys() ... boom_sound.play() ... Explosion(alien, 0) ... kills += 1
This code checks for the collisions between player bullets and all the aliens they might intersect. In this case we only loop over the dictionary keys, but we could loop over the values() or items() if we wanted to do something to the specific shots that collided with aliens. If we did loop over the values() we would be looping through lists that contain sprites. The same sprite may even appear more than once in these different loops, since the same “shot” could have collided against multiple “aliens”.
Those are the basic collision functions that come with pygame. It should be easy to roll your own that perhaps use something different than the “rect” attribute. Or maybe try to fine-tweak your code a little more by directly effecting the collision object, instead of building a list of the collision? The code in the sprite collision functions is very optimized, but you could speed it up slightly by taking out some functionality you don’t need.
Currently there is one main problem that catches new users. When you derive your new sprite class with the Sprite base, you must call the Sprite.__init__() method from your own class __init__() method. If you forget to call the Sprite.__init__() method, you get a cryptic error, like this:
AttributeError: 'mysprite' instance has no attribute '_Sprite__g'.
Because of speed concerns, the current Group classes try to only do exactly what they need, and not handle a lot of general situations. If you decide you need extra features, you may want to create your own Group class.
The Sprite and Group classes were designed to be extended, so feel free to create your own Group classes to do specialized things. The best place to start is probably the actual python source code for the sprite module. Looking at the current Sprite groups should be enough example on how to create your own.
For example, here is the source code for a rendering Group that calls a render() method for each sprite, instead of just blitting an “image” variable from it. Since we want it to also handle updated areas, we will start with a copy of the original RenderUpdates group, here is the code:
class RenderUpdatesDraw(RenderClear): """call sprite.draw(screen) to render sprites""" def draw(self, surface): dirty = self.lostsprites self.lostsprites =  for s, r in self.spritedict.items(): newrect = s.draw(screen) #Here's the big change if r is 0: dirty.append(newrect) else: dirty.append(newrect.union(r)) self.spritedict[s] = newrect return dirty
Following is more information on how you could create your own Sprite and Group objects from scratch.
The Sprite objects only “require” two methods. “add_internal()” and “remove_internal()”. These are called by the Group classes when they are removing a sprite from themselves. The add_internal() and remove_internal() have a single argument which is a group. Your Sprite will need some way to also keep track of the Groups it belongs to. You will likely want to try to match the other methods and arguments to the real Sprite class, but if you’re not going to use those methods, you sure don’t need them.
It is almost the same requirements for creating your own Group. In fact, if you look at the source you’ll see the GroupSingle isn’t derived from the Group class, it just implements the same methods so you can’t really tell the difference. Again you need an “add_internal()” and “remove_internal()” method that the sprites call when they want to belong or remove themselves from the group. The add_internal() and remove_internal() have a single argument which is a sprite. The only other requirement for the Group classes is they have a dummy attribute named “_spritegroup”. It doesn’t matter what the value is, as long as the attribute is present. The Sprite classes can look for this attribute to determine the difference between a “group” and any ordinary python container. (This is important, because several sprite methods can take an argument of a single group, or a sequence of groups. Since they both look similar, this is the most flexible way to “see” the difference.)
You should through the code for the sprite module. While the code is a bit “tuned”, it’s got enough comments to help you follow along. There’s even a todo section in the source if you feel like contributing.