Working with component-based entities
:mod:`sdl2.ext` supports a component oriented programming pattern to separate object instances, carried data and processing logic within applications or games. It uses an entity based approach, in which object instances are unique identifiers, while their data is managed within components, which are stored separately. For each individual component type a processing system will take care of all necessary updates on running the application.
Component-based means that - instead of a traditional OOP approach - object information are split up into separate data bags for reusability and that those data bags are separated from any application logic.
Imagine a car class in traditional OOP, which might look like
class Car: def __init__(self): self.color = "red" self.position = 0, 0 self.velocity = 0, 0 self.sprite = get_some_car_image() ... def drive(self, timedelta): self.position = self.velocity * timedelta self.position = self.velocity * timedelta ... def stop(self): self.velocity = 0, 0 ... def render(self, screen): screen.display(self.sprite) mycar = new Car() mycar.color = "green" mycar.velocity = 10, 0
The car features information stored in attributes (color, position, ...) and behaviour (application logic, drive(), stop() ...).
A component-based approach aims to split and reduce the car to a set of information and external systems providing the application logic.
class Car: def __init__(self): self.color = "red" self.position = 0, 0 self.velocity = 0, 0 self.sprite = get_some_car_image() class CarMovement: def drive(self, car, timedelta): car.position = car.velocity * timedelta car.position = car.velocity * timedelta ... def stop(self): car.velocity = 0, 0 class CarRenderer: def render(self, car, screen): screen.display(car.sprite)
At this point of time, there is no notable difference between both approaches, except that the latter one adds additional overhead.
The benefit comes in, when you
- use subclassing in your OOP design
- want to change behavioural patterns on a global scale or based on states
- want to refactor code logic in central locations
- want to cascade application behaviours
The initial Car class from above defines, how it should be displayed on the screen. If you now want to add a feature for rescaling the screen size after the user activates the magnifier mode, you need to refactor the Car and all other classes that render things on the screen, have to consider all subclasses that override the method and so on. Refactoring the CarRenderer code by adding a check for the magnifier mode sounds quite simple in contrast to that, not?
The same applies to the movement logic - inverting the movement logic requires you to refactor all your classes instead of a single piece of application code.
Subclassing with traditional OOP for behavioural changes also might bloat your classes with unnecessary information, causing the memory footprint for your application to rise without any need. Let's assume you have a Truck class that inherits from Car. Let's further assume that all trucks in your application look the same. Why should any of those carry a sprite or color attribute? You would need to refactor your Car class to get rid of those superfluous information, adding another level of subclassing. If at a later point of time you decide to give your trucks different colors, you need to refactor everything again.
Wouldn't it be easier to deal with colors, if they are available on the truck and leave them out, if they are not? We initially stated that the component-based approach aims to separate data (information) from code logic. That said, if the truck has a color, we can handle it easily, if it has not, we will do as usual.
Also, checking for the color of an object (regardless, if it is a truck, car, aeroplane or death star) allows us to apply the same or similar behaviour for every object. If the information is available, we will process it, if it is not, we will not do anything.
All in all
Once we split up the previously OOP-style classes into pure data containers and some separate processing code for the behaviour, we are talking about components and (processing) systems. A component is a data container, ideally grouping related information on a granular level, so that it is easy to (re)use. When you combine different components to build your in-application objects and instantiate those, we are talking about entities.
- provides information (data bag)
- In-application instance that consists of component items
- Application logic for working with Entity items and their component data
- The environment that contains the different System instances and all Entity items with their component data
Within a strict COP design, the application logic (ideally) only knows about data to process. It does not know anything about entities or complex classes and only operates on the data.
To keep things simple, modular and easy to maintain and change, you usually create small processing systems, which perform the necessary operations on the data they shall handle. That said, a MovementSystem for our car entity would only operate on the position and velocity component of the car entity. It does not know anything about the the car's sprite or sounds that the car makes, since this is nothing it has to deal with.
To display the car on the screen, a RenderingSystem might pick up the sprite component of the car, maybe along with the position information (so it knows, where to place the sprite) and render it on the screen.
If you want the car to play sounds, you would add an audio playback system, that can perform the task. Afterwards you can add the necessary audio information via a sound component to the car and it will make noise.
Component-based design with sdl2.ext
This section will deal with the specialities of COP patterns and provide the bare minimum of information. If you are just starting with such a design, it is recommended to read through the :ref:`pong-tutorial` tutorial.
:mod:`sdl2.ext` provides a :class:`World` class in which all other objects will reside. The :class:`World` will maintain both, :class:`Entity` and component items, and allows you to set up the processing logic via the :class:`System` and :class:`Applicator` classes.
>>> appworld = World()
Components can be created from any class that inherits from the :class:`object` type and represent the data bag of information for the entity and application world. Ideally, they should avoid any application logic (except from getter and setter properties).
class Position2D(object): def __init__(self, x=0, y=0): self.x = x self.y = y
class CarEntity(Entity): def __init__(self, world, x=0, y=0): self.position2d = Position2D(x, y)
The world argument in __init__() is necessary. It will be passed to the internal __new__() constructor of the :class:`Entity` and stores a reference to the :class:`World` and also allows the :class:`Entity` to store its information in the :class:`World`.
The :class:`Entity` also requries its attributes to be named exactly as their component class name, but in lowercase letters. If you name a component MyAbsolutelyAwesomeDataContainer, an :class:`Entity` will force you to write the following:
class SomeEntity(Entity): def __init__(self, world): self.myabsolutelyawesomedatacontainer = MyAbsolutelyAwesomeDataContainer()
This is not entirely true. A reference of the object will be stored on a per-class-in-mro basis. This means that if MyAbsolutelyAwesomeDataContainer inherits from ShortName, you can also do:
class SomeEntity(Entity): def __init__(self, world): self.shortname = MyAbsolutelyAwesomeDataContainer()
Components should be as atomic as possible and avoid complex inheritance. Since each value of an :class:`Entity` is stored per class in its mro list, components inheriting from the same class(es) will overwrite each other on conflicting classes:
class Vector(Position2D): def __init__(self, x=0, y=0, z=0): super(Vector, self).__init__(x, y) class SomeEntity(Entity): def __init__(self, world): # This will associate self.position2d with the new Position2D # value, while the previous Vector association is overwritten self.position2d = Position2D(4, 4) # self.vector will also associate a self.position2d attribute # with the Entity, since Vector inherits from Position2D. The # original association will vanish, and each call to # entity.position2d will effectively manipulate the vector! self.vector = Vector(1,2,3)
A processing system within an application world consumes the components of all entities, for which it was set up. At time of processing, the system does not know about any other component type that might be bound to any entity.
Also, the processing system does not know about any specific entity, but only is aware of the data carried by all entities.
An application world defines the combination of application data and processing logic and how the data will be processed. As such, it is a container object in which the application is defined.
The application world maintains a set of entities and their related components as well as a set of systems that process the data of the entities. Each processing system within the application world only operates on a certain set of components, but not all components of an entity at once.
The order in which data is processed depends on the order of the added systems.