Edict is a fast, powerful and ergonomic ECS crate that expands traditional ECS feature set. Written in Rust by your fellow π¦
use edict::prelude::*;
// Create world instance.
let mut world = World::new();
// Declare some components.
#[derive(Component)]
struct Pos(f32, f32);
// Declare some more.
#[derive(Component)]
struct Vel(f32, f32);
// Spawn entity with components.
world.spawn((Pos(0.0, 0.0), Vel(1.0, 1.0)));
// Query components and iterate over views.
for (pos, vel) in world.view::<(&mut Pos, &Vel)>() {
pos.0 += vel.0;
pos.1 += vel.1;
}
// Define functions that will be used as systems.
#[edict::system::system] // This attribute is optional, but it catches if function is not a system.
fn move_system(pos_vel: View<(&mut Pos, &Vel)>) {
for (pos, vel) in pos_vel {
pos.0 += vel.0;
pos.1 += vel.1;
}
}
// Create scheduler to run systems. Requires "scheduler" feature.
use edict::scheduler::Scheduler;
let mut scheduler = Scheduler::new();
scheduler.add_system(move_system);
// Run systems without parallelism.
scheduler.run_sequential(&mut world);
// Run systems using threads. Requires "std" feature.
scheduler.run_threaded(&mut world);
// Or use custom thread pool.
In Entity Component Systems we create entities and address them to fetch associated data.
Edict provides EntityId
type to address entities.
EntityId
as a world-unique identifier of an entity.
Edict uses IDs without generation and recycling, for this purpose it employs u64
underlying type with a few niches.
It is enough to create IDs non-stop for hundreds of years before running out of them.
This greatly simplifying serialization of the World
's state as it doesn't require any processing of entity IDs.
By default entity IDs are unique only within one World
.
For multi-world scenarios Edict provides a way to make entity IDs unique between any required combination of worlds.
IDs are allocated in sequence from IdRange
s that are allocated by IdRangeAllocator
.
By default IdRange
that spans from 1 to u64::MAX - 1
is used. This makes default ID allocation extremely fast.
Custom IdRangeAllocator
can be provided to WorldBuilder
to use custom ID ranges.
For example in client-server architecture, server and client may use non-overlapping ID ranges. Thus allowing state serialized on server to be transferred to client without ID mapping, which can be cumbersome when components reference entities.
In multi-server or p2p architecture IdRangeAllocator
would need to communicate to allocate disjoint ID ranges for each server.
Using ECS may lead to lots of .unwrap()
calls or excessive error handling.
There a lot of situations when entity is guaranteed to exist (for example it just returned from a view).
To avoid handling NoSuchEntity
error when it is unreachable, Edict provides AliveEntity
trait that extends Entity
trait.
Various methods require AliveEntity
handle and skip existence check.
Entity
and AliveEntity
traits implemented for number of entity types.
EntityId
implements only Entity
as it doesn't provide any guaranties.
EntityBound
is guaranteed to be alive, allowing using it in methods that doesn't handle entity absence.
It keeps lifetime of World
borrow, making it impossible to despawn any entity from the world.
Using it with wrong World
may cause panic.
EntityBound
can be acquire from relation queries.
EntityLoc
not only guarantees entity existence but also contains location of the entity in the archetypes,
allowing to skip lookup step when accessing its components.
Similarly to EntityBound
, it keeps lifetime of World
borrow, making it impossible to despawn any entity from the world.
Using it with wrong World
may cause panic.
EntityLoc
can be acquire from Entities
query.
EntityRef
is special.
It doesn't implement Entity
or AliveEntity
traits since it should not be used in world methods.
Instead it provides direct access to entity's data and allows mutations such as inserting/removing components.
Support for !Send
and !Sync
components and resources with some limitations.
World
itself is not sendable but shareable between threads.
Thread owning World
is referred as "main" thread in documentation.
Components and resources that are !Send
can be fetched mutably only from "main" thread.
Components and resources that are !Sync
can be fetched immutably only from "main" thread.
Since reference to World
may exist outside "main" thread, WorldLocal
reference should be used,
it can be created using mutable reference to World
.
Optional Component
trait that allows implicit component type registration when component is inserted first time.
Implicit registration uses behavior defined by Component
implementation as-is.
When needed, explicit registration can be done using WorldBuilder
to override component behavior.
Non Component
types require explicit registration and
few methods with _external
suffix is used with them instead of normal ones.
Only default registration is possible when World
is already built.
When needed, explicit registration can be done using WorldBuilder
to override component behavior.
A relation can be added to pair of entities, binding them together.
Queries may fetch relations and filter entities by their relations to other entities.
When either of the two entities is despawned, relation is dropped.
Relation
type may further configure behavior of the bounded entities.
Powerful Query
mechanism that can filter entities by components, relations and other criteria and fetch entity data.
Queries can be mutable or immutable, sendable or non-sendable, stateful or stateless.
Using query on World
creates Views.
Views can be used to iterate over entities that match the query yielding query items.
Or fetch single entity data.
ViewRef
and ViewMut
are convenient type aliases to view types returned from World
methods.
Runtime checks are available for query mutable aliasing avoidance.
ViewRef
and ViewCell
do runtime checks allowing multiple views with aliased access coexist,
deferring checks to runtime that prevents invalid aliasing to occur.
When this is not required, ViewMut
and View
s with compile time checks should be used instead.
When View
is expected ViewRef
and ViewCell
can be locked to make a View
.
Component type may define borrowing operations to borrow another type from it. Borrowed type may be not sized, allowing slices and dyn traits to be borrowed. A macro to help define borrowing operations is provided. Queries that tries to borrow type from suitable components are provided:
BorrowAll
borrows from all components that implement borrowing requested type. Yields aVec
with borrowed values since multiple components of the entity may provide it. Skips entities if none of the components provide the requested type.BorrowAny
borrows from first suitable component that implements borrowing requested type. Yields a single value. Skips entities if none of the components provide the requested type.BorrowOne
is configured withTypeId
of component from which it should borrow requested type. Panics if component doesn't provide the requested type. Skips entities without the component.
Built-in type-map for singleton values called "resources".
Resources can be inserted into/fetched from World
.
Resources live separately from entities and their components.
Use ActionEncoder
for recording actions and run them later with mutable access to World
.
Or LocalActionEncoder
instead when action is not Send
.
Or convenient WorldLocal::defer*
methods to defer actions to internal LocalActionEncoder
.
Each component instance is equipped with epoch counter that tracks last potential mutation of the component.
Queries may read and update components epoch to track changes.
Queries to filter recently changed components are provided with Modified
type.
Last epoch can be obtained with World::epoch
.
Systems is convenient way to build logic that operates on World
.
Edict defines System
trait to run logic on World
.
And IntoSystem
trait for types convertible to System
.
Functions may implement IntoSystem
automatically -
it is required to return ()
and accept arguments that implement FnArg
trait.
There are FnArg
implementations:
View
andViewCell
to iterate over entities and their components. UseView
unlessViewCell
is required to handle intra-system views conflict.Res
andResMut
to access resources.ResLocal
andResMutLocal
to access no-thread-safe resources. This will make system non-sendable and force it to run on main thread.ActionEncoder
to record actions that mutateWorld
state, such as entity spawning, inserting and removing components or resources.State
to store system's local state between runs.
Scheduler
is provided to run System
s.
Systems added to the Scheduler
run in parallel where possible,
however they act as if executed sequentially in order they were added.
If systems do not conflict they may be executed in parallel.
If systems conflict, the one added first will be executed before the one added later can start.
std
threads or rayon
can be used as an executor.
User may provide custom executor by implementing ScopedExecutor
trait.
Requires "scheduler"
feature which is enabled by default.
Component replace/drop hooks are called automatically when component is replaced or dropped.
When component is registered it can be equipped with hooks to be called when component value is replaced or dropped.
Implicit registration of Component
types will register hooks defined on the trait impl.
Drop hook is called when component is dropped via World::drop
or entity is despawned and is not
called when component is removed from entity.
Replace hook is called when component is replaced e.g. component is inserted into entity and entity already has component of the same type. Replace hook returns boolean value that indicates if drop hook should be called for replaced component.
Hooks can record actions into provided LocalActionEncoder
that will be executed
before World
method that caused the hook to be called returns.
When component implements Component
trait, hooks defined on the trait impl are registered automatically to call
Component::on_drop
and Component::on_replace
methods.
They may be overridden with custom hooks using WorldBuilder
.
For non Component
types hooks can be registered only via WorldBuilder
.
Default registration with World
will not register any hooks.
Futures executor to run logic that requires waiting for certain conditions or events or otherwise spans for multiple ticks.
Logic that requires waiting can be complex to implement using systems. Systems run in loop and usually work on every entity with certain components. Implementing waiting logic would require adding waiting state to existing or new components and logic would be spread across many system runs or even many systems.
Futures may use await
syntax to wait for certain conditions or events.
Futures that can access ECS data are referred in Edict as "flows".
Flows can be spawned in the World
using World::spawn_flow
or FlowWorld::spawn_flow
method.
Flows
type is used as an executor to run spawned flows.
Flows can be bound to an entity and spawned using World::spawn_flow_for
, FlowWorld::spawn_flow_for
, EntityRef::spawn_flow
or FlowEntity::spawn_flow
method.
Such flows will be cancelled if entity is despawned.
Functions that return futures may serve as flows.
For World::spawn_flow
use function or closure with signature FnOnce(FlowWorld) -> Future
For World::spawn_flow_for
use function or closure with signature FnOnce(FlowEntity) -> Future
User may implement low-level futures using poll*
methods of FlowWorld
and FlowEntity
to access tasks Context
.
Edict provides only a couple of low-level futures that will do the waiting:
yield_now!
yields control to the executor once and resumes on next execution.FlowEntity::wait_despawned
waits until entity is despawned.FlowEntity::wait_has_component
waits until entity get a component.
WakeOnDrop
component can be used when despawning entity should wake a task.
It is recommended to use flows for high-level logic that spans multiple ticks and use systems to do low-level logic that runs every tick. Flows may request systems to perform operations by adding special components to entities. And systems may spawn flows to do long-running operations.
Requires "flow"
feature which is enabled by default.
Edict can be used in no_std
environment but requires alloc
crate.
"std"
feature is enabled by default.
If "std" feature is not enabled error types will not implement std::error::Error
.
When "flow" feature is enabled and "std" is not, extern functions are used to implement TLS. Application must provide implementation for these functions or linking will fail.
When "scheduler" feature is enabled and "std" is not, external functions are used to implement thread parking. Application must provide implementation for these functions or linking will fail.
Licensed under either of
- Apache License, Version 2.0, (license/APACHE or http://www.apache.org/licenses/LICENSE-2.0)
- MIT license (license/MIT or http://opensource.org/licenses/MIT)
at your option.
Unless you explicitly state otherwise, any contribution intentionally submitted for inclusion in the work by you, as defined in the Apache-2.0 license, shall be dual licensed as above, without any additional terms or conditions.