Entity Component System focused on usability and flexibility.
This release took quite some time but I'm very happy of all the changes. The api is now macro free. It took a few tries to get to this api for workloads but it works now. There is just a rustc weird behavior that makes it impossible to use values for workload systems. It's an acceptable price to get rid of the macro but it'd be great if it could be fixed.
Most notable changes:
Remove system!
macro
This macro was used to hide what information workloads need. For example using system!(check)
expended to (|world: &World| world.try_run(check), check))
.
The scheduler would use the function on the right to define what the function on the left borrows. If the two functions didn't match you could run into runtime errors.
The new api (.with_system(&check)
) does not have this problem.
Remove packs
Packs are removed temporarily, this implementation had too many limitations. The next implementation should not have the same problems but requires a fairly big amount of work so this version ships without packs and the next version should add them without modifying any existing api.
remove
was the most impacted, to remove two or more components you had to write Remove::<(usize, u32)>::remove((&mut usizes, &mut u32s), entity);
.
With 0.5 it's just (&mut usizes, &mut u32s).remove(entity)
.
Bulk add entity
This new method is faster than adding entities one by one.
Remove Shiperator
0.4 used a custom Iterator
trait to modify some methods' behavior like filter
.
With 0.5, iterating an exclusive storage yields a Mut<T>
by default which removes the need for a custom trait.
Accurate modification tracking by default
In 0.4 when you tracked modification on a storage it would tag all components accessed exclusively regardless if they were modified or not.
This is not true anymore, this code would tag exactly what you would expect:
fn increment_even(mut u32s: ViewMut<u32>) {
for mut i in (&mut u32s).iter() {
if *i % 2 == 0 {
*i += 1;
}
}
}
No more try_*
In 0.4 all fallible functions had two versions, one that panicked and one prefixed with try_
that returned a Result
.
Now all functions that can fail because of storage access return a Result
while almost all others panic.
Workload building is more flexible
In 0.4 you could only create workloads directly with World
and it was borrowed for the entire building time.
world
.add_workload("Add & Check")
.with_system(system!(add))
.with_system(system!(check))
.build();
In 0.5 you only need to borrow World
to actually add the workload to it.
Workload::builder("Add & Check")
.with_system(&add)
.with_system(&check)
.add_to_world(&world)
.unwrap();
You can also "prepare" systems in advance, merge WorkloadBuilder
s, extend a WorkloadBuilder
with a system iterator or even nest multiple workloads.
Workload debug
When you create a workload you can print information about how it will be scheduled and why two systems can't run at the same time.
Custom view
A lot of internal traits, types and functions are exposed in 0.5. One of them is Borrow
, it allows user defined views. Custom views are very important to build clean apis on top of Shipyard.
For example this custom view (https://github.com/dakom/shipyard-scenegraph/blob/master/crate/src/views.rs#L10-L28) borrows 11 storages. Users won't have to borrow 11 views though, just a single one.
Creating a custom view is currently fairly verbose, I want to add a proc macro that would automatically derive the necessary traits.
Custom storage
Shipyard comes with two default storages: SparseSet
and Unique
. 0.5 allows you to define your own storages. You can then build a view for them and access them in systems for example.
The advantage compared to simply storing your type in Unique
is to be able to define what your storage should do when an entity is deleted.
Fixes a bug reported by @hasenbanck where adding multiple workloads with the same name would cause a panic when running them.
AddWorkload
errorSetDefaultWorkload
errorIterator
and IntoIterator
are now supportedFor 0.3 users there's a migration guide to help with all the changes.
User guide
Learning to use a crate with the documentation can be hard. There's now a guide to explain what can be done with Shipyard and how.
No need to register components anymore
Components had to be registered before accessing them by using World::new
or World::register
.
Storages are now automatically created when they are first accessed.
!Send
and !Sync
components
All components had to be Send + Sync
, this is a strict limitation to be sure storages can use the parallel feature.
With 0.3 !Send
and !Sync
types can be stored and accessed while still following Rust's rules. These rules limit threading for these types but doesn't always prevent it.
Unique
components
When we only need a single instance of a component, keeping an id around to access it can be annoying. A Unique
component won't be attached to any entity but will have the storage all for itself.
Components sorting
No std support
And more 😉