bevy/examples/ecs/ecs_guide.rs
2020-05-01 13:12:47 -07:00

218 lines
8.6 KiB
Rust

use bevy::prelude::*;
/// This is a guided introduction to Bevy's "Entity Component System" (ECS)
/// All Bevy app logic is built using the ECS pattern, so definitely pay attention!
///
/// Why ECS?
/// * Data oriented: Functionality is driven by data
/// * Clean Architecture: Loose coupling of functionality / prevents deeply nested inheritance
/// * High Performance: Massively parallel and cache friendly
///
/// ECS Definitions:
///
/// Component: just a normal Rust data type. generally scoped to a single piece of functionality
/// Examples: position, velocity, health, color, name
///
/// Entity: a collection of components with a unique id
/// Examples: Entity1 { Name("Alice"), Position(0, 0) }, Entity2 { Name("Bill"), Position(10, 5) }
/// Resource: a shared global piece of data
/// Examples: asset_storage, events, system state
///
/// System: runs logic on entities, components, and resources
/// Examples: move_system, damage_system
///
/// Now that you know a little bit about ECS, lets look at some Bevy code!
// Our Bevy app's entry point
fn main() {
// Bevy apps are created using the builder pattern. Here add our
App::build()
// This plugin runs our app's "system schedule" exactly once. Most apps will run on a loop,
// but we don't want to spam your console with a bunch of example text :)
.add_plugin(ScheduleRunnerPlugin::run_once())
// Resources can be added to our app like this
.add_resource(A { value: 1 })
// Resources that implement the Default or FromResources trait can be added like this:
.init_resource::<B>()
.init_resource::<State>()
// Systems can be added to our app like this
// the system() call converts normal rust functions into ECS systems
.add_system(empty_system.system())
// Startup systems run exactly once BEFORE all other systems. These are generally used for
// app initialization code (adding entities and resources)
.add_startup_system(startup_system)
// Systems that need resources to be constructed can be added like this
.init_system(complex_system)
// Here we just the rest of the example systems
.add_system(resource_system.system())
.add_system(for_each_entity_system.system())
.add_system(resources_and_components_system.system())
.add_system(command_buffer_system.system())
.add_system(thread_local_system)
.add_system(closure_system())
.add_system(stateful_system.system())
.run();
}
// RESOURCES: "global" state accessible by systems
struct A {
value: usize,
}
#[derive(Default)]
struct B {
value: usize,
}
struct C;
// COMPONENTS: pieces of functionality we add to entities
struct X {
value: usize,
}
struct Y {
value: usize,
}
// SYSTEMS: logic that runs on entities, components, and resources
// This is the simplest system. It will run once each time our app updates:
fn empty_system() {
println!("hello!");
}
// Systems can also read and modify resources:
fn resource_system(a: Resource<A>, mut b: ResourceMut<B>) {
b.value += 1;
println!("resource_system: {} {}", a.value, b.value);
}
// This system runs once for each entity with the X and Y component
// NOTE: x is a read-only reference (Ref) whereas y can be modified (RefMut)
fn for_each_entity_system(x: Ref<X>, mut y: RefMut<Y>) {
y.value += 1;
println!("for_each_entity_system: {} {}", x.value, y.value);
}
// This system is the same as the above example, but it also accesses resource A
// NOTE: resources must always come before components in system functions
fn resources_and_components_system(a: Resource<A>, x: Ref<X>, mut y: RefMut<Y>) {
y.value += 1;
println!("resources_and_components:");
println!(" components: {} {}", x.value, y.value);
println!(" resource: {} ", a.value);
}
// This is a "startup" system that runs once when the app starts up. The only thing that distinguishes a
// startup" system from a "normal" system is how it is registered:
// app.add_startup_system(startup_system)
// app.add_system(normal_system)
// With startup systems we can create resources and add entities to our world, which can then be used by
// our other systems:
fn startup_system(world: &mut World, resources: &mut Resources) {
// We already added A and B when we built our App above, so we don't re-add them here
resources.insert(C);
// Add some entities to our world
world.insert(
(),
vec![
(X { value: 0 }, Y { value: 1 }),
(X { value: 2 }, Y { value: 3 }),
],
);
// Add some entities to our world
world.insert(
(),
vec![
(X { value: 0 }, Y { value: 1 }),
(X { value: 2 }, Y { value: 3 }),
],
);
}
// This system uses a command buffer to create a new entity on each iteration
// Normal systems cannot safely access the World instance because they run in parallel
// Command buffers give us the ability to queue up changes to our World without directly accessing it
// NOTE: Command buffers must always come before resources and components in system functions
fn command_buffer_system(command_buffer: &mut CommandBuffer, a: Resource<A>) {
// Creates a new entity with a value read from resource A
command_buffer.insert((), vec![(X { value: a.value },)]);
}
// If you really need full/immediate read/write access to the world or resources, you can use a "thread local system".
// These run on the main app thread (hence the name "thread local")
// WARNING: These will block all parallel execution of other systems until they finish, so they should generally be avoided
// NOTE: You may notice that this looks exactly like the "setup" system above. Thats because they are both thread local!
fn thread_local_system(world: &mut World, _resources: &mut Resources) {
world.insert((), vec![(X { value: 1 },)]);
}
// These are like normal systems, but they also "capture" variables, which they can use as local state.
// This system captures the "counter" variable and uses it to maintain a count across executions
// NOTE: This function returns a Box<dyn Schedulable> type. If you are new to rust don't worry! All you
// need to know for now is that the Box contains our system AND the state it captured.
// You may recognize the .system() call from when we added our system functions to our App in the main()
// function above. Now you know that we are actually converting our functions into the Box<dyn Schedulable> type!
fn closure_system() -> Box<dyn Schedulable> {
let mut counter = 0;
(move |x: Ref<X>, mut y: RefMut<Y>| {
y.value += 1;
println!("closure_system: {} {}", x.value, y.value);
println!(" ran {} times: ", counter);
counter += 1;
})
.system()
}
// Closure systems should be avoided in general because they hide state from the ECS. This makes scenarios
// like "saving", "networking/multiplayer", and "replays" much harder.
// Instead you should use the "state" pattern whenever possible:
#[derive(Default)]
struct State {
counter: usize,
}
fn stateful_system(mut state: RefMut<State>, x: Ref<X>, mut y: RefMut<Y>) {
y.value += 1;
println!("stateful_system: {} {}", x.value, y.value);
println!(" ran {} times: ", state.counter);
state.counter += 1;
}
// If you need more flexibility, you can define complex systems using "system builders".
// SystemBuilder enables scenarios like "multiple queries" and "query filters"
fn complex_system(_resources: &mut Resources) -> Box<dyn Schedulable> {
let mut counter = 0;
SystemBuilder::new("complex_system")
.read_resource::<A>()
.write_resource::<B>()
// this query is equivalent to the system we saw above: system(x: Ref<X>, y: RefMut<Y>)
.with_query(<(Read<X>, Write<Y>)>::query())
// this query only runs on entities with an X component that has changed since the last update
.with_query(<Read<X>>::query().filter(changed::<X>()))
.build(
move |_command_buffer, world, (a, ref mut b), (x_y_query, x_changed_query)| {
println!("complex_system:");
println!(" resources: {} {}", a.value, b.value);
for (x, mut y) in x_y_query.iter_mut(world) {
y.value += 1;
println!(
" processed entity {} times: {} {}",
counter, x.value, y.value
);
counter += 1;
}
for x in x_changed_query.iter(world) {
println!(" x changed: {}", x.value);
}
},
)
}