Jump to content
  • Advertisement
  • 10/07/17 10:02 AM

    Game Architecture: Migrating to a Basic Entity Component System

    General and Gameplay Programming

    tgobbens

    Introduction 

    The architecture of software design is a much-debated subject. Every developer has his own opinion about what is good software design and what is not. Most developers agree on what is bad design, on what is good design there are a wide variety of opinions. Unfortunately, due to the nature of software development, there is no silver bullet; there is no one design strategy that always works.

    There are a couple of strategies that have proofed to be successful. These strategies have known strengths and weaknesses. The advantage of using such a strategy is allowing you to focus on building your game, instead of worrying if your codebase will implode after someone decided the game should function a bit different than how the code was originally written.

    Architecture

    In game development, the entity-component-system is an architectural pattern that is used successfully in small, medium and large games. The main strength of this strategy is the usage of the composition over inheritance pattern. This pattern prevents the build-up of complex object inheritance tree’s, that will make your code impossible to refactor without a lot of side-effects.

    So how does this pattern work, at its core, there are three elements; entity, components, and systems guess you didn’t see that coming. ;) Let’s describe these one by one, I’m starting with the smallest and most simple one the “component”:

    Component

     The component represents a single attribute of an entity. Some examples of entities can be:

    • Position
    • Rotation
    • Scale
    • Physics body
    • Texture
    • Health

    Entity

     Entities are the “game-object” some examples:

    • Ball
    • Camera
    • Player
    • Enemy
    • Bullet

    An entity can have multiple components, for example, a ball entity can have the following components: position, rotation, scale, texture and physics body. A camera entity might only have a position. Usually, all entities have a unique-id for fast access, but there can also be other ways of accessing entities.

    System

    The system is responsible for one aspect of the game, a simple game has can, for example, have the following systems:

    • Rendering
    • Physics
    • GUI
    • Sound
    • AI

    What a system does is iterating over all entities that have components of the types defined by the system. For example, the physics system will act only on entities with a physics-component and a position-component. The rendering system will only act on entities that have a position and texture component. 

    For example, if a ball entity has a position and physics and texture component. The physics system will pick up the ball entity, as it has a physics and position component. The physics-system control a physics-engine, that will do its magic and calculate a new position for the ball. The physics system will set this new position on the position component.

    The rendering-system will also pick up the ball entity, as it acts on all entities that have a position and a texture component. The rendering system can render the ball using the texture and the position found with the ball entity (yes, the same component that was just updated by the physics-system).

    Now imagine you spend some time implementing the architecture described above, and you after, running it, realize the ball is not really moving very realistic. That might be because you forgot to take rotation into account.

    To fix it you now only have to create a rotation-component and add it to the ball entity. Now add in the physics system a check if the entity has a rotation component and if so just set the rotation on this component. In the rendering-system also add some code to check if there is a rotation component and if so render with this rotation.

    This is where the power of this architecture emerges, imagine you have not one ball entity but have a lot of entities like the ball, wall, player, ground, etc. and you forgot about the rotation. You only have to modify the rendering system and the physics system. Then by simple adding a rotation component to the entities you want to have rotation on, magically all those objects have a rotation. Now adding rotation seems like a trivial thing to do, but what does this architecture enforces is the separation of concerns (e.g. rendering and physics) while still allows adding new functionality. And important; It does this without the usage of inheritance but rather by composition.

    Possible features

    The architecture described above is in itself quite powerful. But there are some more interesting possibilities with this architecture. I will describe some of them below.

    Game-mechanic tweaking

    Creating a generic mechanic-tuning-utilities; as there are a limit number of component types, you can create a (developer) GUI-overlay that allows you to modify the values of a component. This will allow you to in real-time modify the values, e.g. the position, size, texture, acceleration, the weight of a certain entity his components. This will help you tremendously in fine-tuning game mechanics without the need to keep recompiling and reloading your game.

    Level-editor

    Taking this even a step further you could use the above system to load all entities and relevant components from a file, e.g. XML. This will also help you decrease compile and loading time, letting you focus more one tuning game mechanics. This could then be a very good start for creating a level-editor, letting a none technical team member (game-designers) tweak game mechanics.

    Dynamic loading

    Something else that can be managed using this system, is implementing an entity loading/unloading mechanism. You can define an interface with functions like initializing, loading, starting, stopping, unloading, destructing. Now you can implement a loading mechanism that guarantees the loading and initializing always happen asynchronously. This will allow you to load and unload entities in a uniform and controlled manner. You could also choose to run all the systems in a different thread you need to take some more care about modifying components, but this could allow you to do a lot of performance enhancements, as for example, the AI needs less frequent updates then a renderer.

    Real-world example

    Note this implementation is done in Java, as I’m using libGDX as the platform, but the architecture is certainly not limited to Java and can also be implemented in other languages like C++.

    Enough of the theory, now for a real implementation. As a hobby project, I have been creating a small iOS/Android game, my first implementation of this game was naïve, with one source file containing all logic. No need to explain this is a bad implementation, but this did allow me to check if my idea was fun and create a quick prototype and do some fast iterations from there.

    For reference, the “bad” implementation can still be found here:  https://github.com/tgobbens/fluffybalance/blob/master/core/src/com/balanceball/Balanceball.java  

    After I created this implementation I decided I wanted to implement the same game using a better manageable implementation.

    The “main” entry file can be found here: https://github.com/tgobbens/fluffybalance/blob/master/core/src/com/balanceball/BalanceBallSec.java.

    So, I’ve created my own entity-component-system. If you want to create your own game using an entity-component-system, and want the game to be ready as soon as possible then I wouldn’t recommend writing one yourself. However, if you want to learn about programming or just create something for fun, implementing such a system is easy, and you will learn a lot from doing this. Another reason to implement this yourself is you get a lot of freedom allowing you to add specific tricks and features that can help you improve your codebase.

    The entity component system can be found under https://github.com/tgobbens/fluffybalance/tree/master/core/src/com/sec and yes there are some optimisation and improvements opportunities in this code base. But it does show an easy to understand implementation. When trying to understand make sure you know what Java generic types are. It’s quite common you need to find a certain entity to update or get some info from. As there are a lot of components you know there will be only one instance from. I’ve added a method to get the first entity of a certain type. For example, give me the camera entity, or give me the first “game-world” entity. But there are also helper functions to get all entities of a certain type. The same trick is used for getting components of an entity.

    You will also find a basic type called “engine”, used for binding everything together. This will trigger the updates and holding references to all systems and entities. If you look for a “starting” point for the architecture this is where to start looking.



      Report Article


    User Feedback


    There are no comments to display.



    Create an account or sign in to comment

    You need to be a member in order to leave a comment

    Create an account

    Sign up for a new account in our community. It's easy!

    Register a new account

    Sign in

    Already have an account? Sign in here.

    Sign In Now

  • Advertisement
  • Game Developer Survey

    completed-task.png

    We are looking for qualified game developers to participate in a 10-minute online survey. Qualified participants will be offered a $15 incentive for your time and insights. Click here to start!

    Take me to the survey!

  • Advertisement
  • Latest Featured Articles

  • Featured Blogs

  • Advertisement
  • Popular Now

  • Similar Content

    • By bandages
      So, in real life, incoming dot normal at the silhouette is always 0.  With smooth shaded meshes, it never is, not naturally, not outside of contrived situations.  (Not with flat shaded meshes either, I guess.)
      And incoming dot normal is one of the bedrocks of CG.  Probably the equal of 4x4 matrix multiplication.  Problems with silhouette normals show up in Fresnel, in diffuse lighting, in environment mapping....  everywhere.  But I can't really find anybody talking about it.  (Maybe I'm not Googling the right terms.)
      Obviously, the problem decreases as poly count goes up, eventually reaching a point where it's dwarfed by other silhouette problems (like translucency or micro-occlusion) that CG doesn't handle well either.  But, if I'm reasoning correctly, normal maps don't improve the problem-- they're as likely to exacerbate it as improve it, and the exacerbations are, aesthetically speaking, probably worse than the improvements are better.
      I've tried playing with crude fixes-- basically, rotating normals toward incoming by a percentage, or of course clamping incoming dot normal (like we all have to do) to prevent it from bending behind the mesh.  Nothing I've tried looks good.  I suppose the best option might be to rotate normals to perpendicular to incoming at the silhouette and then interpolate to the nearest inflection point  of something like screen space depth to preserve curvature, but the math for how to do that is beyond me, and I'm not sure it would look any better.  Or maybe, instead, somehow, adjust the drawn silhouette to match the silhouette defined by incoming dot normal?  Not even sure how that would work, not if the normal was pointing away from incoming.
      I don't know-- is this a solvable problem?  Has anyone tried other stuff and given up, pursued anything that was promising but too expensive, anything like that?  Are there any papers I'm missing?  It's really surprising to me that I can't find anyone else talking about this.
      (Apologies if I chose the wrong subforum for this.  I considered art forums, but I felt that people frequenting the programming forums would have more to say on the subject.)
    • By mujina
      What could be a way of avoiding using inheritance and virtual methods when designing components for an entity-component-system?
      I'll be more specific about my design issue:
      I currently have different classes for different kinds of colliders (let's say, CircleCollider and LineCollider).
      My system that checks for collisions and updates the positions and/or velocities of my entities should be something like:
      for entity_i in alive_entities { collider_i = get_collider_of_entity(entity_i) // components of same kind are stored contiguously in separate arrays transform_i = get_transform_of_entity(entity_i) for entity_j in alive_entities { collider_j = get_collider_of_entity(entity_j) transform_j = get_transform_of_entity(entity_j) if check_collision(collider_i, collider_j) { update(transform_i) update(transform_j) } } } my problem is that I don't have a generic `get_collider_of_entity` function, but rather a function `get_circle_collider_of_entity` and a separate one `get_line_collider_of_entity`, and so on. (This happens because under the hood I am keeping a mapping (entity_id -> [transform_id, sprite_id, circle_collider_id, line_collider_id, ...]) that tells me whether an entity is using certain kinds of components and which are the indices of those components in the arrays containing the actual components instances. As you can see, each component class is corresponding to a unique index, namely the index position of the array of the mapping described above. For example, transforms are 0, sprites are 1, circle colliders are 2, line colliders are 3, and so on.)
      I am in need to write a system as the one in the snippet above. I can write several overloaded `check_collision` functions that implement the logic for collision detection between different kinds of geometric primitives, but my problem is that I am not sure how to obtain a generic `get_collider_of_entity` function. I would need something that would get me the collider of an entity, regardless of whether the entity has a circle collider, a line collider, a square collider, etc.
      One solution could be to write a function that checks whether in my internal entity_id -> [components_ids] mapping a certain entity has a collider at any of the indices that correspond to colliders. For example, say that the indices related to the collider classes are indices 10 to 20, then my function would do
      get_collider_of_entity (entity_id) { for comp_type_id in 10..20{ if mapping[entity_id][comp_type_id] not null { return components_arrays[comp_type_id][entity_id] } } return null } This could turn out to be pretty slow, since I have to do a small search for every collider of every entity. Also, it may not be straightforward to handle returned types here. (I'm working with C++, and the first solution - that is not involving inheritance in any way - would be returning a std::variant<CircleCollider, LineCollider, ... all kinds of components>, since I would need to return something that could be of different types).
      Another solution could be having some inheritance among components, e.g. all specific component classes inherit from a base Collider, and overrride some virtual `collide_with(const Collider& other)` method. Then I would redesign my mapping to probably reserve just one index for colliders, and then I would actual colliders in a polymorphic array of pointers to colliders, instead of having a separate array for CircleColliders, another for LineColliders, and so on. But this would destroy any attempt to be cache-friendly in my design, wouldn't it? That's why I am looking for alternatives.
      A third alternative would be to just have a single, only, Collider class. That would internally store the "actual type" ( aka what kind of collider it is ) with dynamic information (like an enum ColliderType). Then I would have all colliders have all members needed by any kind of colliders, and specific collision detection functions which I can dispatch dynamically that only use some of that data. (Practical example: a "Collider" would have a radius, and the coordinate for 2 points, and in case its type was "circle" it would only make use of the radius and of one of the 2 points - used as the center -, while if it was a "segment" it would only make use of the 2 points). My gut feeling is that this would bloat all colliders, and, even if the bloat could be reduced - using unions in some smart way for storing members? I wouldn't know how -, then still the design would be pretty brittle.
      I'm clueless and open for ideas and advice! How do you handle in general situations in which you have components that can be naturally modeled as subclasses of a more generic component class? Inheritance? Smart hacks with variants, templates, macros, custom indexing? Dynamic "internal" type?
×

Important Information

By using GameDev.net, you agree to our community Guidelines, Terms of Use, and Privacy Policy.

GameDev.net is your game development community. Create an account for your GameDev Portfolio and participate in the largest developer community in the games industry.

Sign me up!