This repository is my first attempt at creating an ECS (Entity Component System) to work alongside the Godot Game Engine. I've been researching it quite a bit recently and am very interested in its philosophy. It is by no means perfect, but I'm enjoying the workflow so far, so it could be worth sharing. If anything, it's an interesting paradigm shift that others might be interested in.
Important Note: This ECS attempt is not going to and not meant to magically make your Godot Game run faster. I also understand that there is plenty of disagreement about what "ECS" actually means. When I was doing my initial research into the topic, it seems that there are some sticklers for a very specific definition of ECS that believe that ECS ONLY refers to the original use-case for cache optimizations and having data loaded contiguously in memory in order to speed up processing of related components. However, after doing plenty of research, I will respectfully disagree.
Just a select few articles that support my case that ECS has expanded beyond its original scope and can be utilized even if it's not for its intended original purpose of cache optimization.
- Tobias Stein
- "memory efficiency" is only ONE of the listed benefits. Even if we don't get the memory efficiency benefit, we still get the rest of them by following an ECS-like architecture.
- Richard Lord's post on the topic.
- Erik Hazzard
For now, clone the repository (which is just a sample project). I may refactor the distribution method later to turn it into more of an "addon" if this repository picks up any traction.
gd-ecs, like every other ECS, consists of Entities, Components, and Systems. My core philosophy behind creating this system was to integrate well with Godot's pre-existing concepts and to sort-of allow for the "best of both worlds" of Godot's built-in Object-oriented Inheritance model and the benefits of writing code in ECS fashion.
By default, your game should have at least one SystemManager Node. A SystemManager manages all of your Systems and glues all of gd-ecs together. It is worth noting that it is not necessary for Entities to be children (or sub-children) of the SystemManager. Internally the SystemManager has a QueryManager that will listen for all changes and update its internal registry of Entity/Component requirements even at runtime. You just use add_child()
and remove_child()
as you would normally, but if you happen to be adding/removing an Entity or a Component, gd-ecs will know about it automatically. (no need to call special methods like add_component()
or something)
Components in gd-ecs are pretty traditional. They are data-only and represent state.
gd-ecs uses duck-typing to determine whether a Node is a Component. It simply checks that the component_name
property exists and is not empty.
Here is an example of the C_KinematicMotion2D
Component. It simply contains a few exported variables and state variables. Very standard stuff.
class_name C_KinematicMotion2D
extends Node
const component_name := "C_KinematicMotion2D"
export var acceleration_default := Vector2(5000.0, 0.0)
export var max_speed_default := Vector2(400.0, 1200.0)
var acceleration: Vector2
var on_floor: bool
var on_wall: bool
var velocity: Vector2
var max_speed: Vector2
func _ready() -> void:
acceleration = acceleration_default
max_speed = max_speed_default
That being said, any pre-existing Node can be a Component, even if it has more than just data and is self-contained in functionality. Want to attach a Sprite to your Entity but don't want to build out a System that handles loading the textures and actually rendering it? Create a Sprite Component! It might not immediately be obvious why you would create a Sprite Component rather than just attaching a normal Sprite (which would work too). You probably want to create a component like this because even if you don't have a System to specifically act upon and provide functionality for the Component since it is self-contained, creating a Component like this will allow for querying from Systems (More on this later)
class_name C_Sprite
extends Sprite
const component_name := "C_Sprite"
You don't need to worry about the rendering or the transform of the Sprite Component or anything in your Systems -- Godot will handle it all for you. (Note that if you want to add something like a Sprite Component, it still needs to follow Godot's built-in hierarchy, so your Entity's type should also be a derivative of Node2D
. If you try to add the above C_Sprite
Component to an Entity that is just a normal Node, it won't work out of the box because children Node2Ds of a regular Node don't inherit their parent Node's transform properties. (and probably other things)
In gd-ecs, all Entities are not strictly "just an id referencing a container of components". Any Node can be an Entity. All you need to do is attach the Entity.gd
script to a Node to turn it into an Entity.
Before you flip out about my not following ECS to the letter, hear me out -- I am effectively treating ALL Godot-provided Nodes as built-in types. Similarly to how you wouldn't re-implement the standard library in C++ even if you were building a more traditional ECS, I'm going to treat Godot-provided Nodes as a standard library that is available anywhere in the project. This means that even if TECHNICALLY, the code I'm executing lives inside of one of my entities rather than in a System, I'm going to be calling the code from inside of a System. The ECS part of gd-ecs does NOT apply to the Engine, but only for the game-logic in the project itself. Rendering, Sound, etc are all abstracted away to the Engine and I don't worry about it.
Since you declare an Entity by creating a base Node, attaching the Entity.gd
script, and then adding just a bunch of children Components, it is perfect for saving in a PackedScene .tscn
file. See the repo-included Player.tscn
file for an example.
When processing an entity, you usually want to be able to access its Components. The easiest ways to do this is with get_component(component_name: String) -> Node
and get_components(component_name: String) -> Array
. They're pretty self-explanatory. get_component()
returns the first matching Component, and get_components()
returns all matching Components. There is a little bit of additional complexity here since an Entity can have duplicates of the same Component type, but I'll worry about this problem in the future :)
Systems are managed by a SystemManager Node. They too, are determined by duck-typing and gd-ecs checks for the existence of a system_name
property. There is a built-in System.gd
Node that can be extended for most use-cases, but a completely custom System node can be built since gd-ecs relies strictly on duck-typing to determine whether a System is valid or not.
However, at runtime, there are a few additional check for a few more things:
- Systems must have a
_system_init()
method that returnstrue
. The SystemManager will call_system_init()
on all of its children and "register" only the ones that returntrue
here. It will pass a reference to itself as a parameter. Here is an example of the default_system_init()
.
var system_manager: SystemManager
func _system_init(_system_manager: SystemManager) -> bool:
system_manager = _system_manager
return true
- Systems must have a non-empty Array
requirements
that is an Array of Strings containing Component requirements for the System. These requirements define which entities a System will process. Here is an example of a system that requires the ComponentsC_Input
andC_Player
.
class_name S_PlayerInput
extends System
func _init() -> void:
system_name = "S_PlayerInput"
requirements = ["C_Input", "C_Player"]
You can additionally negate requirements with a !
. In the below example, the System would register itself to only process Entities that have the C_Input
Component, but do NOT have the C_Player
Component.
func _init() -> void:
system_name = "S_OtherInput"
requirements = ["C_Input", "!C_Player"]
-
Systems have an OPTIONAL Array
entity_filter
that is an Array of Strings containing a filter for the entities that the System will retrieve during its_system_process()
and_system_physics_process()
cycles. Since in gd-ecs, Entities can be any node and aren't just an id, it might be more accurate to describe them as "Assemblages" or "Archetypes". Essentially, think ofEntity.gd
as just a normal entity that's a container of components, but attaching theEntity.gd
script to for example, to Godot'sKinematicBody2D
really just turns it into an Entity with aKinematicBody2D
Component that can't be removed. This filter string matches against Godot'sget_class()
method, so it can only match Godot's built-in Nodes, not custom classes. Similarly to the regular component requirements, you can negate it with a!
. In the below example, let's look at what is happening:class_name S_Jump extends System func _init() -> void: entity_filter = ["KinematicBody2D"] system_name = "S_Jump" requirements = ["C_Jump", "C_KinematicMotion2D"] func _system_physics_process(entities: Array, delta: float) -> void: for e in entities: print(e)
This Jump system requires an Entity to have
C_Jump
andC_KinematicMotion2D
components to even consider looking at the Entity. However, because I've provided an optionalentity_filter
, it now will do a check on theclass_name()
of the Entity before passing it into the system ticks.The following code is completely equivalent to the above.
class_name S_Jump extends System func _init() -> void: system_name = "S_Jump" requirements = ["C_Jump", "C_KinematicMotion2D"] func _system_physics_process(entities: Array, delta: float) -> void: for e in entities: if e.get_class() == "KinematicBody2D": print(e)
I know that this is probably going to be one of the more debatable features of gd-ecs, so I've made it purely optional. If you want to be a little bit more "ECS", you can instead attach an empty Component called
C_IsKinematicBody2D
to your Entity and instead filter withrequirements = ["C_IsKinematicBody2D", "C_Jump", "C_KinematicMotion2D"]
Bonus Note: Because the entity filter runs on every tick rather than rebuilding the query manager's internal registry with Component requirements, it can be changed on the fly at runtime and there will be no issues with it. If for whatever reason, you had a System that normally acts on 2 different entity types, but perhaps you wanted to filter out one of them for some reason, if you change the
entity_filter
, theentities
passed in the tick parameters will be filtered out in the next tick. That being said, I don't really see much of a use case for this... I'm just documenting it because i guess it's a possible feature that's really just a side effect of the implementation.
Systems have 3 virtual methods that get called automatically by the SystemManager.
_system_ready() -> void
is called after the SystemManager completes the validation of all of its children Systems and has registered all their requirements._system_process(entities: Array, delta: float) -> void
is called every_process()
loop of the SystemManager.entities
will contain an Array of only the Entities that match the System's requirements.- Bonus Note: If the System has a
tps
property, you can define the "ticks per second" that the_system_process
loop is executed. For example, if I want to poll for player input 30 times per second, I could do
andfunc _init() -> void: system_name = "S_PlayerInput" requirements = ["C_Input", "C_Player"] tps = 30
_system_process
will only be called 30 times per second.- Bonus Note: If the System has a
_system_physics_process(entities: Array, delta: float) -> void:
does exactly what you might guess. It is called on every physics frame and passes along an Array of Entities that meet the System's requirements.
Since Systems are just Nodes, you can still use Godot's built-in virtual methods of _process
and_physics_process
, but the difference is that the _system_*
ones are only going to be called on "active" systems that passed the initial validation mentioned previously.
You may want to communicate from one System to another System without necessarily knowing the names of the systems or storing references to them. (Standard Decoupling). You can solve this by using an Event Singleton. Alternatively, you can use the built in signaling system in the SystemManager that dynamically creates signals and allows for arbitrary bulletin-board like communication since Systems are passed a reference to their SystemManager on _system_init
.
SystemManager emit(destination: String, payload) -> void
and subscribe(destination: String, system: Node, callback_name: String) -> void
methods that let you emit or subscribe to an arbitrary "destination" string. Any System that subscribes to a string will receive a payload emitted by another System.
NOTE: payload
is a Variant, but if you pass in an Array, it will be split into additional parameters when the signal is emitted rather than being passed as an Array. If you want to emit an Array payload, you need to nest the Array inside of an Array.
Here's an example (assume that both SystemA and SystemB have the following init code):
var system_manager: SystemManager
func _system_init(_system_manager: SystemManager) -> bool:
system_manager = _system_manager
System A:
func _system_process(entities: Array, delta: float) -> void:
for e in entities:
var fake_condition: = true
if fake_condition:
system_manager.emit("fake_signal", [e, "hello])
System B:
func _system_ready() -> void:
system_manager.subscribe("fake_signal", self, "_on_fake_signal")
func _on_fake_signal(entity: Entity, message: String) -> void:
prints(entity, message)
Here, SystemB and SystemA are able to communicate by emitting/subscribing to "fake_signal". No need to predefine your signals. That being said, it may be a good idea to do so in order to better keep track of where all your signals are coming from. Perhaps in the future, there will be debug functionality added to the SystemManager to help trace unused signals or to offer to option of requiring predefined signals.
Please open a Github Issue! I'd love to get Tests set up on all of this but have been too lazy... I'd love it if someone were to contribute some... particularly for the QueryManager since it's pretty important for it to pick up changes properly (when adding/removing entities/components)
- Jonathan Chun Author
- Buch Demo Art from OpenGameArt.org
This project is licensed under the MIT License - see the LICENSE file for details