Module evenio::tutorial

source ·
Expand description

§Why ECS?

Entity Component System (often shortened to ‘ECS’) is a programming design pattern in which the structure of the application is broken up into Entities, Components, and Systems.

  • Entities are the “things” in the application, such as a player, UI widget, scene, etc. Entities are identified with simple index-based EntityId handles.
  • Components represent the capabilities that an entity possesses. Entities are described as a set of components. For instance, an asteroid entity flying through space might have a Position and Velocity component to describe its flight path over time. In evenio, components are Rust data types.
  • Systems are procedures which act on all entities with certain combinations of components. A physics system might need an entity’s Position, Velocity, and Mass components in order to calculate the object’s next position. This typically means you have a list of systems that run in sequence (or parallel) every frame or time step of your application.

ECS helps solve a number of problems that often arise in traditional object-oriented software designs. Different “types” of entities are able to share components, subsuming the functionality of inheritance. Additionally, ECS can be very efficient due to the way that components are organized in memory.

ECS is especially useful in Rust because it can, in some sense, ergonomically sidestep Rust’s aliasing rules. Concurrent entity component access is made possible so long as it conforms to Rust’s aliasing XOR mutability demands.

§Why evenio?

Interactive applications are largely event driven. When the player attacks a monster, reduce its health. When an egg hits the ground, break it open. When the “play” button is clicked, start the game. Reifying events in code is a useful design tool because it gives downstream code something to hook into and subsequently change the behavior of the program. In other words, extensibility.

However, many prior Rust ECS libraries have trouble expressing the control flow that events require. Some frameworks will suggest writing events to a temporary buffer and then handling those events in a later system. This is a “pull” style of event propagation in which event consumers must regularly poll the buffer for work. This works well in some cases, but runs into some problems:

  • Creates system ordering considerations. Consider what happens if an event is written to the buffer after the event handling systems run. The event might not be handled until the next frame, or never.
  • Funneling a sequence of events into separate event buffers loses the order between buffers. This is a problem when handling a stream of incoming network packets, for instance.
  • Systems waste time polling the event buffer every frame even when there are no events available.

The conclusion, then, is that traditional batch-oriented run-once-per-frame systems alone are insufficient for expressing many kinds of application logic. This may not come as a surprise in hindsight, but due to Rust’s constraints, it’s tempting to apply this design pattern everywhere. When all you have is a hammer, everything looks like a nail.

Evenio solves these problems by introducing “push” style events as a first-class citizen. As you’ll see later, all systems are generalized as event handlers and run immediately in response to events. The goal is to make it easy to write correct code by making the rules around event propagation easy to understand. Events need to be fast, too!

§The World

To begin using the library, we’ll need a World.

use evenio::prelude::*; // Import the most commonly used items.

let mut world = World::new(); // Create an empty world.

The World is essentially a database of all the data that the ECS can access. It contains all the entities, components, and event handlers. Later, we’ll see how to add data to the world and query for data within it.

An application usually only needs a single World, but multiple worlds can be used if the need arises.

Also note that Worlds are not Send or Sync because they are allowed to contain data that is not Send or Sync. This means that you cannot share an entire World across threads (but parallel data access is still possible, as we’ll see later).

§Handlers and Events

Handlers are callbacks that run in response to events.

To get started, let’s create a “hello world” handler that runs in response to our custom Message event.

use evenio::prelude::*;

struct Message<'a>(&'a str);

let mut world = World::new();

// Add the event handler to our world.

// Send our event.
world.send(Message("Hello, World!"));

fn my_handler(r: Receiver<Message>) {
    println!("The message is: \"{}\"", r.event.0);


The message is: "Hello, World!"

Receiver<E> is a handler parameter which tells the handler that we would like to listen for events of type E.

Note that handlers must listen for exactly one event type. Attempting to listen for more than one event or no events at all will panic.

// Panics. (lambdas are also handlers)
world.add_handler(|| {});

§Handler Ordering

When multiple handlers listen for the same event, we’ll need to consider the order those handlers should run when the event is handled.

Handler order is first determined by the handler’s HandlerPriority. This is a enum with three states: High, Medium, and Low. Medium is the default. If handlers have the same priority, then we fall back on the order the handlers were added to the World to decide the order.

use evenio::prelude::*;

let mut world = World::new();

// Give this handler the `High` priority.


struct MyEvent;

fn handler_a(_: Receiver<MyEvent>) {
    println!("handler A");

fn handler_b(_: Receiver<MyEvent>) {
    println!("handler B");

fn handler_c(_: Receiver<MyEvent>) {
    println!("handler C");


handler C
handler A
handler B

Although handler_c was added to the world last, it was given Priority::High, so it ran first.

§Event Mutation

So far, we’ve only had immutable access to events from within handlers. But sometimes it’s desirable to mutate events so that later handlers will see the event differently.

We can achieve this by using the ReceiverMut handler parameter. ReceiverMut provides a mutable reference to the event.

struct MyEvent(i32);

world.add_handler(|mut r: ReceiverMut<MyEvent>| {
    println!("Event is: {}", r.event.0);
    r.event.0 += 10;

world.add_handler(|r: Receiver<MyEvent>| {
    println!("Event is now: {}", r.event.0);


This prints:

Event is: 42
Event is now: 52

As a general rule of thumb, use Receiver when you can, and ReceiverMut when you have to.

§Event Ownership

We can take this a step further and entirely take ownership of the event. Doing so will stop the event from broadcasting and later handlers will not receive the event. This is done with EventMut::take.

struct MyEvent(u32);

world.add_handler(|r: ReceiverMut<MyEvent>| {
    if r.event.0 % 2 == 0 {

world.add_handler(|r: Receiver<MyEvent>| {
    println!("got event: {}", r.event.0);

for i in 0..10 {


got event: 1
got event: 3
got event: 5
got event: 7
got event: 9

This can be useful in a number of different scenarios. Consider a TakeDamage event used to signal an entity taking damage in a game. With event consumption, we can take ownership of the event before it has a chance to propagate, effectively creating an “invincibility” effect.

§Sending Events From Handlers

Previously, we’ve seen how to send events using the World::send method. But to send events from within a handler, we’ll need to use the Sender handler parameter:

struct A;
struct B;
struct C;

world.add_handler(|_: Receiver<A>, mut sender: Sender<(B, C)>| {
    println!("sent B and C!");

world.add_handler(|_: Receiver<B>| println!("got B!"));
world.add_handler(|_: Receiver<C>| println!("got C!"));



sent B and C!
got B!
got C!

In the parameter Sender<(B, C)>, the (B, C) is the set of events the sender is allowed to send. Attempting to send an event that is not in this set will fail.

Note that Sender::send does not immediately send events, but rather adds them to the front of the event queue in reverse order. The next event in the queue begins broadcasting once all handlers for the current event have finished.

use evenio::prelude::*;

struct A;
#[derive(GlobalEvent, Debug)]
struct B(i32);
#[derive(GlobalEvent, Debug)]
struct C(i32);

fn get_a_send_b(_: Receiver<A>, mut sender: Sender<B>) {
    println!("got A, sending B twice!");

fn get_b_send_c(r: Receiver<B>, mut sender: Sender<C>) {
    sender.send(C(r.event.0 + 1));
    sender.send(C(r.event.0 + 2));
    println!("got {:?}, sending C twice!", r.event);

fn get_c(r: Receiver<C>) {
    println!("got {:?}!", r.event);

let mut world = World::new();


println!("sending A!");


sending A!
got A, sending B twice!
got B(0), sending C twice!
got C(1)!
got C(2)!
got B(3), sending C twice!
got C(4)!
got C(5)!

§Entities and Components

Entities and components are the bread and butter of any ECS framework.

Entities make up the things in your application, such as monsters, players, cameras, GUI elements, or items. By itself, an entity is just a unique identifier (EntityId). It is the entity’s set of components that differentiate a player from a monster, or a camera from an item. Components hold the data while handlers operate on entities with certain sets of components.

Let’s see how we might model player and monster entities in a game with evenio.

use evenio::prelude::*;

struct Health(i32);

struct Position {
    x: f32,
    y: f32,

// A zero-sized component, often called a "marker" or "tag".
struct Player;

struct Monster;

let mut world = World::new();

// Spawn three entities without any components.
// `spawn()` returns a lightweight handle we can use to look up the entity later.
let player = world.spawn();
let monster = world.spawn();
let something_else = world.spawn();

// Insert the player components.
world.insert(player, Health(100));
world.insert(player, Position { x: 0.0, y: 0.0 });
world.insert(player, Player);

// Insert the monster components.
world.insert(monster, Health(20));
world.insert(monster, Position { x: 100.0, y: 100.0 });
world.insert(monster, Monster);

// Something else?
world.insert(something_else, Position { x: 42.0, y: 42.0 });

We can visualize our data using a table where the rows are entities and columns are components:

Entity IDHealthPositionPlayerMonster
0100(0, 0)
120(100, 100)
2(42, 42)

§Bundling Components

If our game has lots of monsters with lots of components, we might grow tired of inserting the components individually. To fix this, we can create a new event and handler to do the inserting for us.

#[derive(GlobalEvent, Clone, Copy)]
struct InitMonster {
    entity: EntityId,
    pos: [f32; 2],

fn init_monster_handler(
    r: Receiver<InitMonster>,
    mut s: Sender<(Insert<Health>, Insert<Position>, Insert<Monster>)>
) {
    let InitMonster {
        pos: [x, y]
    } = *r.event;

    s.insert(entity, Health(20));
    s.insert(entity, Position { x, y });
    s.insert(entity, Monster);


// Test our new handler:
let entity = world.spawn();
world.send(InitMonster {
    pos: [24.0, 24.0],


Whenever we insert a component on an entity, what we’re actually doing is sending the special Insert event for that component. Because of this, we have to specify the correct Insert events in the handler’s Sender above.


To remove an entity from the world, we send the Despawn event. This will drop all of the entity’s components and make its EntityId invalid. The EntityId will never be reused by a new entity.

Alternatively, if we just want to remove components without deleting the entire entity, we send the Remove event for the specific component(s).

struct A;

let e = world.spawn();
world.insert(e, A);

// Component exists.


// No more component.

// Sending the `Remove` event again has no effect.


// Entity no longer exists.

§Performance Considerations

Internally, all entities with the same set of components are organized into groups called archetypes. Archetypes enable fast entity iteration and cleanup among other benefits.

However, this design means that adding or removing a component on an entity will force all components of that entity to move somewhere else in memory, which can be slow. If components are being rapidly added and removed, consider using a bool or an Option inside the component instead.


In the previous section, we’ve seen how to create entities and add components to them. But components aren’t very useful unless we have some way to access them from within handlers.

This is where the Fetcher handler parameter comes in. Fetcher<Q> allows for both random access entity lookups using an EntityId and iteration over all entities matching some Query Q.

struct E;

#[derive(Component, Debug)]
struct A(&'static str);

#[derive(Component, Debug)]
struct B(i32);

let e1 = world.spawn();
world.insert(e1, A("foo"));
world.insert(e1, B(123));

let e2 = world.spawn();
world.insert(e2, A("bar"));

let e3 = world.spawn();
world.insert(e3, B(456));

world.add_handler(move |_: Receiver<E>, fetcher: Fetcher<&A>| {
    // Get a reference to the `A` component on entity `e2`.
    // Returns `None` if the query doesn't match.
    let a = fetcher.get(e1).unwrap();
    println!("fetcher.get(): {a:?}");
    // `e3` doesn't have component `A`.

    // Iterate over all entities with the `A` component.
    // Entities are visited in a deterministic but unspecified order.
    for a in fetcher.iter() {
        println!("fetcher.iter(): {a:?}");



fetcher.get(): A("foo")
fetcher.iter(): A("bar")
fetcher.iter(): A("foo")


Queries are expressions used to filter and access data from entities. In the previous example, we used the &A query to match entities with the A component, but there are many other queries to choose from:

&CMatches entities with component C and returns an immutable reference to C.
&mut CMatches entities with component C and returns a mutable reference to C.
(Q₁, Q₂, …, Qₙ)Matches if all queries Q₁ through Qₙ match. Returns a tuple of the results.
()The empty tuple matches all entities and returns nothing.
Option<Q>Returns the result of query Q as Some, or None if Q does not match. Matches all entities.
Or<Q₁, Q₂>Matches if Q₁ or Q₂ matches. Returns the matched query results.
Xor<Q₁, Q₂>Matches if Q₁ or Q₂ matches, but not both. Returns the matched query results.
EntityIdReturns the EntityId of the matched entity. Matches all entities.
Has<Q>Returns a boolean indicating whether the query Q matches. Matches all entities.
With<Q>Matches if Q matches. Returns nothing and does not access the result of Q.
Not<Q>Matches if Q does not match. Returns nothing.

§Derived Queries

Query can be derived on structs so long as every field is a Query. Derived queries act like tuples — every field must match in order for the whole query to match.

Compared to tuples, however, derived query structs offer some advantages. Fields are named and methods can be implemented on the struct – handy for complex or frequently used queries.

#[derive(Query, Debug)]
struct MyQuery<'a> {
    foo: &'a A,
    bar: &'a mut B,
    _baz: With<&'static C>,

impl MyQuery<'_> {
    fn print(&self) {
        println!("foo: {:?}, bar: {:?}",,;

world.add_handler(|_: Receiver<E>, fetcher: Fetcher<MyQuery>| {
    for my_query in fetcher {

§Access Conflicts

Within a query or set of queries, it is possible to create component accesses that violate Rust’s aliasing rules. Consider:

// Panics!
world.add_handler(|_: Receiver<E>, _: Fetcher<&mut A>, _: Fetcher<&A>| {});

Because both fetchers are capable of accessing the same entity, it is possible to create &mut A and &A references to the same component, thus triggering Undefined Behavior. add_handler guards against this by panicking if the handler parameters have any possibility of causing unsoundness.

There are a number of ways to resolve this situation, but let’s look at one way using With and Not:

// Doesn't panic.
world.add_handler(|_: Receiver<E>, _: Fetcher<(&mut A, With<&B>)>, _: Fetcher<(&A, Not<&B>)>| {});

The set of entities matched by both queries are now disjoint. Therefore, there is no possibility of overlapping access to A, and the handler can be added to the world without panicking.

§Singleton Entities

Through the course of development, we may find ourselves wanting to store global data in a location that is easily accessed by handlers.

To facilitate this, evenio has the Single handler parameter. Single is parameterized by a query that must match a single entity in the world. If the query does not match exactly one entity, then a runtime panic occurs.

To create our global variable, we

  1. create a component to hold the data
  2. initialize the component and attach it to an entity
  3. let Single match against it.
struct MyGlobalData {
    foo: i32,
    bar: &'static str,

let e = world.spawn();
world.insert(e, MyGlobalData { foo: 123, bar: "data" });

world.add_handler(|_: Receiver<E>, g: Single<&MyGlobalData>| {
    println!("foo: {}, bar: {}",,;


For handling the situation where Single fails, see TrySingle.

For global data scoped to a single handler, see Local.

§Targeted Events

Events in evenio come in two flavors: global and targeted.

  • Global events will simply be received by all handlers listening for the event. Use .send(...) to send a global event.
  • Targeted events are directed at a particular entity. Use .send_to(...) to choose the entity to send the event to. Only handlers whose filter matches the target entity will receive the event.

The advantage of targeted events is efficiency. Only the handlers whose chosen query matches the target entity will run. Other handlers won’t event be considered.

To crate a targeted event, use the TargetedEvent derive macro.

struct MyTargetedEvent {
    data: i32,

To listen for a targeted event, we must use the second type parameter of Receiver or ReceiverMut. If the target matches the query, the query data is accessible in the query field of the receiver. Otherwise, the handler is not run.

struct Health(i32);

struct Stamina(i32);

world.add_handler(|mut receiver: Receiver<MyTargetedEvent, (&mut Health, &Stamina)>| {
    let (health, stamina) = receiver.query;

    if stamina.0 == 100 {
        health.0 +=;

let e = world.spawn();
world.insert(e, Health(20));
world.insert(e, Stamina(100));

world.send_to(e, MyTargetedEvent { data: 10 });

assert_eq!(world.get::<Health>(e).unwrap().0, 30);

The special Insert, Remove, and Despawn events are targeted.

§Immutable Components

When defining a component, we have the option to mark the component as immutable. An immutable component disallows direct &mut access, but permits manipulation through Insert and Remove events. This gives the component author more control over its usage.

Let’s say we had a component to represent UUIDs for entities. UUIDs are usually not meant to change during the lifetime of the entity, so let’s see how we can prevent mutation entirely.

To start, we’ll mark the component as immutable.

/// A universally unique identifier for an entity.
struct Uuid(u128);

Attempting to get a mutable reference to the UUID component will fail at compile time:

let e = world.spawn();
world.insert(e, Uuid(12345));

// Does not compile.

The UUID could still be changed using the Insert event, so let’s raise an error if the UUID would be overwritten.

Note that handlers listening for an Insert event will run before the component is inserted on the entity.


let e = world.spawn();
world.insert(e, Uuid(123)); // Doesn't panic
world.insert(e, Uuid(456)); // Panics

fn detect_uuid_overwrite(r: Receiver<Insert<Uuid>, (EntityId, With<&Uuid>)>) {
    panic!("attempt to overwrite UUID on {:?}", r.query.0)

To make this completely airtight, we’ll also need to protect against removing the component with Remove.


let e = world.spawn();
world.remove::<Uuid>(e); // Doesn't panic
world.insert(e, Uuid(123));
world.remove::<Uuid>(e); // Panics

fn detect_uuid_remove(r: Receiver<Remove<Uuid>, (EntityId, With<&Uuid>)>) {
    panic!("attempt to remove UUID on {:?}", r.query.0)

§Immutable Events

Like components, events can be marked as immutable. Doing so will prevent users from mutating or consuming the event.

struct MyEvent;

// Compile error.
world.add_handler(|_: ReceiverMut<MyEvent>| {});