# Objective
#13152 added support for reflecting functions. Now, we need a way to
register those functions such that they may be accessed anywhere within
the ECS.
## Solution
Added a `FunctionRegistry` type similar to `TypeRegistry`.
This allows a function to be registered and retrieved by name.
```rust
fn foo() -> i32 {
123
}
let mut registry = FunctionRegistry::default();
registry.register("my_function", foo);
let function = registry.get_mut("my_function").unwrap();
let value = function.call(ArgList::new()).unwrap().unwrap_owned();
assert_eq!(value.downcast_ref::<i32>(), Some(&123));
```
Additionally, I added an `AppFunctionRegistry` resource which wraps a
`FunctionRegistryArc`. Functions can be registered into this resource
using `App::register_function` or by getting a mutable reference to the
resource itself.
### Limitations
#### `Send + Sync`
In order to get this registry to work across threads, it needs to be
`Send + Sync`. This means that `DynamicFunction` needs to be `Send +
Sync`, which means that its internal function also needs to be `Send +
Sync`.
In most cases, this won't be an issue because standard Rust functions
(the type most likely to be registered) are always `Send + Sync`.
Additionally, closures tend to be `Send + Sync` as well, granted they
don't capture any `!Send` or `!Sync` variables.
This PR adds this `Send + Sync` requirement, but as mentioned above, it
hopefully shouldn't be too big of an issue.
#### Closures
Unfortunately, closures can't be registered yet. This will likely be
explored and added in a followup PR.
### Future Work
Besides addressing the limitations listed above, another thing we could
look into is improving the lookup of registered functions. One aspect is
in the performance of hashing strings. The other is in the developer
experience of having to call `std::any::type_name_of_val` to get the
name of their function (assuming they didn't give it a custom name).
## Testing
You can run the tests locally with:
```
cargo test --package bevy_reflect
```
---
## Changelog
- Added `FunctionRegistry`
- Added `AppFunctionRegistry` (a `Resource` available from `bevy_ecs`)
- Added `FunctionRegistryArc`
- Added `FunctionRegistrationError`
- Added `reflect_functions` feature to `bevy_ecs` and `bevy_app`
- `FunctionInfo` is no longer `Default`
- `DynamicFunction` now requires its wrapped function be `Send + Sync`
## Internal Migration Guide
> [!important]
> Function reflection was introduced as part of the 0.15 dev cycle. This
migration guide was written for developers relying on `main` during this
cycle, and is not a breaking change coming from 0.14.
`DynamicFunction` (both those created manually and those created with
`IntoFunction`), now require `Send + Sync`. All standard Rust functions
should meet that requirement. Closures, on the other hand, may not if
they capture any `!Send` or `!Sync` variables from its environment.
# Objective
To implement relations we will need to add a `ComponentIndex`, which is
a map from a Component to the list of archetypes that contain this
component.
One of the reasons is that with fragmenting relations the number of
archetypes will explode, so it will become inefficient to create and
update the query caches by iterating through the list of all archetypes.
In this PR, we introduce the `ComponentIndex`, and we update the
`QueryState` to make use of it:
- if a query has at least 1 required component (i.e. something other
than `()`, `Entity` or `Option<>`, etc.): for each of the required
components we find the list of archetypes that contain it (using the
ComponentIndex). Then, we select the smallest list among these. This
gives a small subset of archetypes to iterate through compared with
iterating through all new archetypes
- if it doesn't, then we keep using the current approach of iterating
through all new archetypes
# Implementation
- This breaks query iteration order, in the sense that we are not
guaranteed anymore to return results in the order in which the
archetypes were created. I think this should be fine because this wasn't
an explicit bevy guarantee so users should not be relying on this. I
updated a bunch of unit tests that were failing because of this.
- I had an issue with the borrow checker because iterating the list of
potential archetypes requires access to `&state.component_access`, which
was conflicting with the calls to
```
if state.new_archetype_internal(archetype) {
state.update_archetype_component_access(archetype, access);
}
```
which need a mutable access to the state.
The solution I chose was to introduce a `QueryStateView` which is a
temporary view into the `QueryState` which enables a "split-borrows"
kind of approach. It is described in detail in this blog post:
https://smallcultfollowing.com/babysteps/blog/2018/11/01/after-nll-interprocedural-conflicts/
# Test
The unit tests pass.
Benchmark results:
```
❯ critcmp main pr
group main pr
----- ---- --
iter_fragmented/base 1.00 342.2±25.45ns ? ?/sec 1.02 347.5±16.24ns ? ?/sec
iter_fragmented/foreach 1.04 165.4±11.29ns ? ?/sec 1.00 159.5±4.27ns ? ?/sec
iter_fragmented/foreach_wide 1.03 3.3±0.04µs ? ?/sec 1.00 3.2±0.06µs ? ?/sec
iter_fragmented/wide 1.03 3.1±0.06µs ? ?/sec 1.00 3.0±0.08µs ? ?/sec
iter_fragmented_sparse/base 1.00 6.5±0.14ns ? ?/sec 1.02 6.6±0.08ns ? ?/sec
iter_fragmented_sparse/foreach 1.00 6.3±0.08ns ? ?/sec 1.04 6.6±0.08ns ? ?/sec
iter_fragmented_sparse/foreach_wide 1.00 43.8±0.15ns ? ?/sec 1.02 44.6±0.53ns ? ?/sec
iter_fragmented_sparse/wide 1.00 29.8±0.44ns ? ?/sec 1.00 29.8±0.26ns ? ?/sec
iter_simple/base 1.00 8.2±0.10µs ? ?/sec 1.00 8.2±0.09µs ? ?/sec
iter_simple/foreach 1.00 3.8±0.02µs ? ?/sec 1.02 3.9±0.03µs ? ?/sec
iter_simple/foreach_sparse_set 1.00 19.0±0.26µs ? ?/sec 1.01 19.3±0.16µs ? ?/sec
iter_simple/foreach_wide 1.00 17.8±0.24µs ? ?/sec 1.00 17.9±0.31µs ? ?/sec
iter_simple/foreach_wide_sparse_set 1.06 95.6±6.23µs ? ?/sec 1.00 90.6±0.59µs ? ?/sec
iter_simple/sparse_set 1.00 19.3±1.63µs ? ?/sec 1.01 19.5±0.29µs ? ?/sec
iter_simple/system 1.00 8.1±0.10µs ? ?/sec 1.00 8.1±0.09µs ? ?/sec
iter_simple/wide 1.05 37.7±2.53µs ? ?/sec 1.00 35.8±0.57µs ? ?/sec
iter_simple/wide_sparse_set 1.00 95.7±1.62µs ? ?/sec 1.00 95.9±0.76µs ? ?/sec
par_iter_simple/with_0_fragment 1.04 35.0±2.51µs ? ?/sec 1.00 33.7±0.49µs ? ?/sec
par_iter_simple/with_1000_fragment 1.00 50.4±2.52µs ? ?/sec 1.01 51.0±3.84µs ? ?/sec
par_iter_simple/with_100_fragment 1.02 40.3±2.23µs ? ?/sec 1.00 39.5±1.32µs ? ?/sec
par_iter_simple/with_10_fragment 1.14 38.8±7.79µs ? ?/sec 1.00 34.0±0.78µs ? ?/sec
```
# Objective
- Fix#14629
## Solution
- Make `QueryState::transmute`, `QueryState::transmute_filtered`,
`QueryState::join` and `QueryState::join_filtered` take a `impl
Into<UnsafeWorldCell>` instead of a `&Components` and validate their
`WorldId`
## Migration Guide
- `QueryState::transmute`, `QueryState::transmute_filtered`,
`QueryState::join` and `QueryState::join_filtered` now take a `impl
Into<UnsafeWorldCell>` instead of a `&Components`
---------
Co-authored-by: Carter Anderson <mcanders1@gmail.com>
# Objective
Fixes https://github.com/bevyengine/bevy/issues/14277.
May also fix https://github.com/bevyengine/bevy/issues/14255, needs
verification.
## Solution
Explicitly order `CameraUpdateSystem` before `UiSystem::Prepare`, so
that when the window resizes, `camera_system` will update the `Camera`'s
viewport size before `ui_layout_system` also reacts to the window resize
and tries to read the new `Camera` viewport size to set UI node sizes
accordingly.
## Testing
I tested that explicitly ordering `CameraUpdateSystem` _after_ triggers
the buggy behavior, and explicitly ordering it _before_ does not trigger
the buggy behavior or crash the app (which also demonstrates that the
system sets are ambiguous).
---
## Migration Guide
`CameraUpdateSystem` is now explicitly ordered before
`UiSystem::Prepare` instead of being ambiguous with it.
# Objective
- We previously had a dependency in `bevy_utils`, `hashbrown = 0.14`,
and used the `hashbrown::hash_table` api, which was introduced in
`0.14.2`.
## Solution
- Bump `hashbrown` to `0.14.2`
## Testing
- Now compiles with the minimum declared `hashbrown` version.
---
# Objective
- currently, bevy employs sparse iteration if any of the target
components in the query are stored in a sparse set. it may lead to
increased cache misses in some cases, potentially impacting performance.
- partial fixes#12381
## Solution
- use dense iteration when an archetype and its table have the same
entity count.
- to avoid introducing complicate unsafe noise, this pr only implement
for `for_each ` style iteration.
- added a benchmark to test performance for hybrid iteration.
## Performance
nearly 2x win in specific scenarios, and no performance degradation in
other test cases.
---------
Co-authored-by: Alice Cecile <alice.i.cecile@gmail.com>
Co-authored-by: Christian Hughes <9044780+ItsDoot@users.noreply.github.com>
# Objective
I want to get the visual depth (after view proj matrix stuff) of the
object beneath my cursor.
Even when having a write-back of the depth texture, you would still need
to convert the NDC depth to a logical value.
## Solution
This is done on shader-side by [this
function](e6261b0f5f/crates/bevy_pbr/src/render/view_transformations.wgsl (L151)),
which I ported over to the cpu-side.
I also added `world_to_viewport_with_depth` to get a `Vec3` instead of
`Vec2`.
---
If anyone knows a smarter solution to get the visual depth instead of
going `screen -> viewport ray -> screen`, please let me know :>
# Objective
This idea came up in the context of a hypothetical "text sections as
entities" where text sections are children of a text bundle.
```rust
commands
.spawn(TextBundle::default())
.with_children(|parent} {
parent.spawn(TextSection::from("Hello"));
});
```
This is a bit cumbersome (but powerful and probably the way things are
headed). [`bsn!`](https://github.com/bevyengine/bevy/discussions/14437)
will eventually make this nicer, but in the mean time, this might
improve ergonomics for the common case where there is only one
`TextSection`.
## Solution
Add a `with_child` method to the `BuildChildren` trait that spawns a
single bundle and adds it as a child to the entity.
```rust
commands
.spawn(TextBundle::default())
.with_child(TextSection::from("Hello"));
```
## Testing
I added some tests, and modified the `button` example to use the new
method.
If any potential co-authors want to improve the tests, that would be
great.
## Alternatives
- Some sort of macro. See
https://github.com/tigregalis/bevy_spans_ent/blob/main/examples/macro.rs#L20.
I don't love this, personally, and it would probably be obsoleted by
`bsn!`.
- Wait for `bsn!`
- Add `with_children_batch` that takes an `Into<Iterator>` of bundles.
```rust
with_children_batch(vec![TextSection::from("Hello")])
```
This is maybe not as useful as it sounds -- it only works with
homogeneous bundles, so no marker components or styles.
- If this doesn't seem valuable, doing nothing is cool with me.
# Objective
I can't mutate the dof settings via tools like `bevy_inspector_egui`
## Solution
Add `Reflect` for `DepthOfFieldSettings` and `DepthOfFieldMode`
# Objective
Bevy's direction types have `new` and `new_unchecked` constructors, but
no unchecked variant for the `Dir2::from_xy` and `Dir3::from_xyz`
methods.
For me, this has several times lead to constructing directions like
this, in cases where the components of the direction are already known
to be normalized:
```rust
let normal = Dir2::new_unchecked(Vec2::new(-ray.direction.x.signum(), 0.0));
```
```rust
segment.direction =
Dir2::new_unchecked(Vec2::new(-segment.direction.x, segment.direction.y));
```
For consistency and ergonomics, it would be nice to have unchecked
variants of `Dir2::from_xy` and `Dir3::from_xyz`:
```rust
let normal = Dir2::from_xy_unchecked(-ray.direction.x.signum(), 0.0);
```
```rust
segment.direction = Dir2::from_xy_unchecked(-segment.direction.x, segment.direction.y);
```
## Solution
Add `Dir2::from_xy_unchecked` and `Dir3::from_xyz_unchecked`.
# Objective
- Fix#14295
## Solution
- Early out when `GFBD::get_index_and_compare_data` returns None.
## Testing
- Tested on a selection of examples including `many_foxes` and
`3d_shapes`.
- Resolved the original issue in `bevy_vector_shapes`.
# Objective
Spamming the window close button on window may trigger a panic.
```
thread 'main' panicked at <Bevy repo>\crates\bevy_ecs\src\system\commands\mod.rs:1320:13:
error[B0003]: Could not insert a bundle (of type `bevy_window:🪟:ClosingWindow`) for entity 0v1#4294967296 because it doesn't exist in this World. See: https://bevyengine.org/learn/errors/b0003
note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace
Encountered a panic when applying buffers for system `bevy_window::system::close_when_requested`!
2024-08-01T15:00:29.742612Z WARN bevy_ecs::world::command_queue: CommandQueue has un-applied commands being dropped. Did you forget to call SystemState::apply?
Encountered a panic in system `bevy_app::main_schedule::Main::run_main`!
error: process didn't exit successfully: `target\debug\bevy.exe` (exit code: 101)
```
## Solution
Don't panic when trying to insert the `ClosingWindow` component into a
entity.
## Testing
Found and tested on windows. I haven't checked if this bug happens on
linux or macos.
For testing I ran this code:
```rust
use std::{thread, time::Duration};
use bevy::prelude::*;
fn lag() {
thread::sleep(Duration::from_millis(300));
}
fn main() -> AppExit {
App::new()
.add_plugins(DefaultPlugins)
.add_systems(Update, lag)
.run()
}
```
Then spammed the window close button. The panic no longer occurs.
# Objective
B0003 indicates that you tried to act upon a nonexistant entity, but
does not mention where the error occured:
```
2024-07-31T15:46:25.954840Z WARN bevy_ecs::world: error[B0003]: Could not despawn entity Entity { index: 4294967295, generation: 1 } because it doesn't exist in this World. See: https://bevyengine.org/learn/errors/b0003
```
## Solution
Include caller location:
```
2024-07-31T15:46:25.954840Z WARN bevy_ecs::world: error[B0003]: src/main.rs:18:11: Could not despawn entity Entity { index: 4294967295, generation: 1 } because it doesn't exist in this World. See: https://bevyengine.org/learn/errors/b0003
```
Open question: What should the exact message format be?
## Testing
None, this doesn't change any logic.
# Objective
While scrolling through the animation crate, I was confused by the docs
and code for the two methods. One does nothing for resetting an
animation, the other just resets the weights for whatever reason.
## Solution
Made the functions work accordingly to their documentation.
`start` now replays the animation.
And `play` doesn't reset the weight anymore. I have no clue why it
should. `play` is there to don't do anything to an already existing
animation.
## Testing
I tested the current 0.14 code with bevy playground in the Animated Fox
exampled and changed it such that on pressing space, either `play` or
`start` would be called. Neither changed anything.
I then inlined the function for start there and it restarted the
animation, so it should work.
---
## Migration Guide
`AnimationPlayer::start` now correspondingly to its docs restarts a
running animation.
`AnimationPlayer::play` doesn't reset the weight anymore.
# Objective
Resolve possible ambiguity detection panic between `time_system` and
`event_update_system`.
Fixes#14524
## Solution
Sets `.ambiguous_with(event_update_system)` on `time_system`. This is
slightly new territory for me, so please treat with scepticism.
## Testing
As described in the issue, added
```
.configure_schedules(ScheduleBuildSettings {
ambiguity_detection: LogLevel::Error,
..default()
})
```
to the `time` example and ran it.
# Objective
Fixes#12139
## Solution
See this comment on original issue for my proposal:
https://github.com/bevyengine/bevy/issues/12139#issuecomment-2241915791
This PR is an implementation of this proposal.
I modified the implementation of `fmt::Debug` to instead display
`0v0#12345` to ensure entity index, generation, and raw bits are all
present in the output for debug purposes while still keeping log message
concise.
`fmt::Display` remains as is (`0v0`) to offer an even shorter output.
To me, this is the most non-intrusive fix for this issue.
## Testing
Add `fn entity_debug` test
---------
Co-authored-by: Alice Cecile <alice.i.cecile@gmail.com>
Co-authored-by: Gino Valente <49806985+MrGVSV@users.noreply.github.com>
# Objective
- Fixes#14517.
## Solution
- Replace two instances of `map()` with `inspect()`.
- `#[allow(dead_code)]` on `Bundle` derive macro tests.
## Testing
You need to install the beta toolchain, since these lints are not stable
yet.
```bash
cargo +beta clippy --workspace
cargo +beta test --workspace
```
# Objective
The `SceneInstanceReady` event would be more ergonomic (and potentially
efficient) if it could be delivered to listeners attached to the scene
entities becoming ready rather than into a World-global queue.
This is an evolution of @Shatur's work in #9313.
## Solution
The scene spawner is changed to trigger observers on the scene entity
when it is ready rather than enqueue an event with `EventWriter`.
This addresses the two outstanding feature requests mentioned on #2218,
that i) the events should be "scoped" in some way and ii) that the
`InstanceId` should be included in the event.
## Testing
Modified the `scene_spawner::tests::event` test to use the new
mechanism.
---
## Changelog
- Changed `SceneInstanceReady` to trigger an entity observer rather than
be written to an event queue.
- Changed `SceneInstanceReady` to carry the `InstanceId` of the scene.
## Migration Guide
If you have a system which read `SceneInstanceReady` events:
> ```fn ready_system(ready_events: EventReader<'_, '_,
SceneInstanceReady>) {```
It must be rewritten as an observer:
> ```commands.observe(|trigger: Trigger<SceneInstanceReady>| {```
Or, if you were expecting the event in relation to a specific entity or
entities, as an entity observer:
> ```commands.entity(entity).observe(|trigger:
Trigger<SceneInstanceReady>| {```
# Objective
- Fixes#11219
## Solution
- Scaling calculations use texture dimensions instead of layout
dimensions.
## Testing
- Did you test these changes? If so, how?
All UI examples look fine.
- How can other people (reviewers) test your changes? Is there anything
specific they need to know?
Example in #11219
## Migration Guide
```diff
let ui_node = ExtractedUiNode {
stack_index,
transform,
color,
rect,
image,
- atlas_size: Some(atlas_size * scale_factor),
+ atlas_scaling: Some(Vec2::splat(scale_factor)),
clip,
flip_x,
flip_y,
camera_entity,
border,
border_radius,
node_type,
},
```
```diff
let computed_slices = ComputedTextureSlices {
slices,
- image_size,
}
```
# Objective
- Dynamic plugins were deprecated in #13080 due to being unsound. The
plan was to deprecate them in 0.14 and remove them in 0.15.
## Solution
- Remove all dynamic plugin functionality.
- Update documentation to reflect this change.
---
## Migration Guide
Dynamic plugins were deprecated in 0.14 for being unsound, and they have
now been fully removed. Please consider using the alternatives listed in
the `bevy_dynamic_plugin` crate documentation, or worst-case scenario
you may copy the code from 0.14.
# Objective
- Make it possible to know *what* changed your component or resource.
- Common need when debugging, when you want to know the last code
location that mutated a value in the ECS.
- This feature would be very useful for the editor alongside system
stepping.
## Solution
- Adds the caller location to column data.
- Mutations now `track_caller` all the way up to the public API.
- Commands that invoke these functions immediately call
`Location::caller`, and pass this into the functions, instead of the
functions themselves attempting to get the caller. This would not work
for commands which are deferred, as the commands are executed by the
scheduler, not the user's code.
## Testing
- The `component_change_detection` example now shows where the component
was mutated:
```
2024-07-28T06:57:48.946022Z INFO component_change_detection: Entity { index: 1, generation: 1 }: New value: MyComponent(0.0)
2024-07-28T06:57:49.004371Z INFO component_change_detection: Entity { index: 1, generation: 1 }: New value: MyComponent(1.0)
2024-07-28T06:57:49.012738Z WARN component_change_detection: Change detected!
-> value: Ref(MyComponent(1.0))
-> added: false
-> changed: true
-> changed by: examples/ecs/component_change_detection.rs:36:23
```
- It's also possible to inspect change location from a debugger:
<img width="608" alt="image"
src="https://github.com/user-attachments/assets/c90ecc7a-0462-457a-80ae-42e7f5d346b4">
---
## Changelog
- Added source locations to ECS change detection behind the
`track_change_detection` flag.
## Migration Guide
- Added `changed_by` field to many internal ECS functions used with
change detection when the `track_change_detection` feature flag is
enabled. Use Location::caller() to provide the source of the function
call.
---------
Co-authored-by: BD103 <59022059+BD103@users.noreply.github.com>
Co-authored-by: Gino Valente <49806985+MrGVSV@users.noreply.github.com>
# Objective
Optimize the cloning process for Access-related structs in the ECS
system, specifically targeting the `clone_from` method.
Previously, profiling showed that 1% of CPU time was spent in
`FixedBitSet`'s `drop_in_place`, due to the default `clone_from`
implementation:
```rust
fn clone_from(&mut self, source: &Self) {
*self = source.clone()
}
```
This implementation causes unnecessary allocations and deallocations.
However, [FixedBitSet provides a more optimized clone_from
method](https://github.com/petgraph/fixedbitset/blob/master/src/lib.rs#L1445-L1465)
that avoids these allocations and utilizes SIMD instructions for better
performance.
This PR aims to leverage the optimized clone_from method of FixedBitSet
and implement custom clone_from methods for Access-related structs to
take full advantage of this optimization. By doing so, we expect to
significantly reduce CPU time spent on cloning operations and improve
overall system performance.
## Solution
- Implemented custom `clone` and `clone_from` methods for `Access`,
`FilteredAccess`, `AccessFilters`, and `FilteredAccessSet` structs.
- Removed `#[derive(Clone)]` and manually implemented `Clone` trait to
use optimized `clone_from` method from `FixedBitSet`.
- Added unit tests for cloning and `clone_from` methods to ensure
correctness.
## Testing
- Conducted performance testing comparing the original and optimized
versions.
- Measured CPU time consumption for the `clone_from` method:
- Original version: 1.34% of CPU time
- Optimized version: 0.338% of CPU time
- Compared FPS before and after the changes (results may vary depending
on the run):
Before optimization:
```
2024-07-28T12:49:11.864019Z INFO bevy diagnostic: fps : 213.489463 (avg 214.502488)
2024-07-28T12:49:11.864037Z INFO bevy diagnostic: frame_time : 4.704746ms (avg 4.682251ms)
2024-07-28T12:49:11.864042Z INFO bevy diagnostic: frame_count: 7947.000000 (avg 7887.500000)
```
After optimization:
```
2024-07-28T12:29:42.705738Z INFO bevy diagnostic: fps : 220.273721 (avg 220.912227)
2024-07-28T12:29:42.705762Z INFO bevy diagnostic: frame_time : 4.559127ms (avg 4.544905ms)
2024-07-28T12:29:42.705769Z INFO bevy diagnostic: frame_count: 7596.000000 (avg 7536.500000)
```
---
Reviewers can test these changes by running `cargo run --release
--example ssr`
# Objective
The `ui_layout_system` relies on change detection to sync parent-child
relation to taffy. The children need to by synced before node removal to
avoid trying to set deleted nodes as children (due to how the different
queries collect entities). This however may leave nodes that were
removed set as children to other nodes in special cases.
Fixes#11385
## Solution
The solution is simply to re-sync the changed children after the nodes
are removed.
## Testing
Tested with `sickle_ui` where docking zone highlights would end up
glitched when docking was done in a certain manner:
- run the `docking_zone_splits` example
- pop out a tab from the top
- dock the floating panel in the center right
- grab another tab and try to hover the original static docking zone:
the highlight is semi-stuck
- (NOTE: sometimes it worked even without the fix due to scheduling
order not producing the bugged query results)
After the fix, the issue is no longer present.
NOTE: The performance impact should be minimal, as the child sync relies
on change detection. The change detection was also the reason the parent
nodes remained "stuck" with the phantom children if no other update were
done to them.
# Objective
Clarify that `StatesPlugin` is a prerequisite for state code.
Closes#14329 .
Edit: am I missing a way to link `DefaultPlugins` correctly other than
using the URL? I guess I expected to be able to refer to it with
`bevy::prelude::DefaultPlugins` or some such 🤔
# Objective
Previously, this area of bevy_math used raw translation and rotations to
encode isometries, which did not exist earlier. The goal of this PR is
to make the codebase of bevy_math more harmonious by using actual
isometries (`Isometry2d`/`Isometry3d`) in these places instead — this
will hopefully make the interfaces more digestible for end-users, in
addition to facilitating conversions.
For instance, together with the addition of #14478, this means that a
bounding box for a collider with an isometric `Transform` can be
computed as
```rust
collider.aabb_3d(collider_transform.to_isometry())
```
instead of using manual destructuring.
## Solution
- The traits `Bounded2d` and `Bounded3d` now use `Isometry2d` and
`Isometry3d` (respectively) instead of `translation` and `rotation`
parameters; e.g.:
```rust
/// A trait with methods that return 3D bounding volumes for a shape.
pub trait Bounded3d {
/// Get an axis-aligned bounding box for the shape translated and
rotated by the given isometry.
fn aabb_3d(&self, isometry: Isometry3d) -> Aabb3d;
/// Get a bounding sphere for the shape translated and rotated by the
given isometry.
fn bounding_sphere(&self, isometry: Isometry3d) -> BoundingSphere;
}
```
- Similarly, the `from_point_cloud` constructors for axis-aligned
bounding boxes and bounding circles/spheres now take isometries instead
of separate `translation` and `rotation`; e.g.:
```rust
/// Computes the smallest [`Aabb3d`] containing the given set of points,
/// transformed by the rotation and translation of the given isometry.
///
/// # Panics
///
/// Panics if the given set of points is empty.
#[inline(always)]
pub fn from_point_cloud(
isometry: Isometry3d,
points: impl Iterator<Item = impl Into<Vec3A>>,
) -> Aabb3d { //... }
```
This has a couple additional results:
1. The end-user no longer interacts directly with `Into<Vec3A>` or
`Into<Rot2>` parameters; these conversions all happen earlier now,
inside the isometry types.
2. Similarly, almost all intermediate `Vec3 -> Vec3A` conversions have
been eliminated from the `Bounded3d` implementations for primitives.
This probably has some performance benefit, but I have not measured it
as of now.
## Testing
Existing unit tests help ensure that nothing has been broken in the
refactor.
---
## Migration Guide
The `Bounded2d` and `Bounded3d` traits now take `Isometry2d` and
`Isometry3d` parameters (respectively) instead of separate translation
and rotation arguments. Existing calls to `aabb_2d`, `bounding_circle`,
`aabb_3d`, and `bounding_sphere` will have to be changed to use
isometries instead. A straightforward conversion is to refactor just by
calling `Isometry2d/3d::new`, as follows:
```rust
// Old:
let aabb = my_shape.aabb_2d(my_translation, my_rotation);
// New:
let aabb = my_shape.aabb_2d(Isometry2d::new(my_translation, my_rotation));
```
However, if the old translation and rotation are 3d
translation/rotations originating from a `Transform` or
`GlobalTransform`, then `to_isometry` may be used instead. For example:
```rust
// Old:
let bounding_sphere = my_shape.bounding_sphere(shape_transform.translation, shape_transform.rotation);
// New:
let bounding_sphere = my_shape.bounding_sphere(shape_transform.to_isometry());
```
This discussion also applies to the `from_point_cloud` construction
method of `Aabb2d`/`BoundingCircle`/`Aabb3d`/`BoundingSphere`, which has
similarly been altered to use isometries.
# Objective
When depending on the `bevy_ui` crate specifically and using the
`serialize` feature flag, the compilation fails due to `bevy_math` not
having the serialize flag enabled.
## Solution
Added the `serialize` flag to the `bevy_math` dependency when using that
flag on `bevy_ui`.
## Testing
Tested by adding `bevy_math = { version = "0.14", features =
["serialize"] }` on a small Bevy library to ensure compilation was
successful.
Fixes#14418
Note that this does not add AtomicPtr, which would need its own special
casing support, just the regular value types.
Also, I was forced to be opinionated about which Ordering to use, so I
chose SeqCst as the strictest by default.
# Objective
Previously, our cubic spline constructors would produce
`CubicCurve`/`RationalCurve` output with no data when they themselves
didn't hold enough control points to produce a well-formed curve.
Attempting to sample the resulting empty "curves" (e.g. by calling
`CubicCurve::position`) would crash the program (😓).
The objectives of this PR are:
1. Ensure that the curve output of `bevy_math`'s spline constructions
are never invalid as data.
2. Provide a type-level guarantee that `CubicCurve` and `RationalCurve`
actually function as curves.
## Solution
This has a few pieces. Firstly, the curve generator traits
`CubicGenerator`, `CyclicCubicGenerator`, and `RationalGenerator` are
now fallible — they have associated error types, and the
curve-generation functions are allowed to fail:
```rust
/// Implement this on cubic splines that can generate a cubic curve from their spline parameters.
pub trait CubicGenerator<P: VectorSpace> {
/// An error type indicating why construction might fail.
type Error;
/// Build a [`CubicCurve`] by computing the interpolation coefficients for each curve segment.
fn to_curve(&self) -> Result<CubicCurve<P>, Self::Error>;
}
```
All existing spline constructions use this together with errors that
indicate when they didn't have the right control data and provide curves
which have at least one segment whenever they return an `Ok` variant.
Next, `CubicCurve` and `RationalCurve` have been blessed with a
guarantee that their internal array of segments (`segments`) is never
empty. In particular, this field is no longer public, so that invalid
curves cannot be built using struct instantiation syntax. To compensate
for this shortfall for users (in particular library authors who might
want to implement their own generators), there is a new method
`from_segments` on these for constructing a curve from a list of
segments, failing if the list is empty:
```rust
/// Create a new curve from a collection of segments. If the collection of segments is empty,
/// a curve cannot be built and `None` will be returned instead.
pub fn from_segments(segments: impl Into<Vec<CubicSegment<P>>>) -> Option<Self> { //... }
```
All existing methods on `CyclicCurve` and `CubicCurve` maintain the
invariant, so the direct construction of invalid values by users is
impossible.
## Testing
Run unit tests from `bevy_math::cubic_splines`. Additionally, run the
`cubic_splines` example and try to get it to crash using small numbers
of control points: it uses the fallible constructors directly, so if
invalid data is ever constructed, it is basically guaranteed to crash.
---
## Migration Guide
The `to_curve` method on Bevy's cubic splines is now fallible (returning
a `Result`), meaning that any existing calls will need to be updated by
handling the possibility of an error variant.
Similarly, any custom implementation of `CubicGenerator` or
`RationalGenerator` will need to be amended to include an `Error` type
and be made fallible itself.
Finally, the fields of `CubicCurve` and `RationalCurve` are now private,
so any direct constructions of these structs from segments will need to
be replaced with the new `CubicCurve::from_segments` and
`RationalCurve::from_segments` methods.
---
## Design
The main thing to justify here is the choice for the curve internals to
remain the same. After all, if they were able to cause crashes in the
first place, it's worth wondering why safeguards weren't put in place on
the types themselves to prevent that.
My view on this is that the problem was really that the internals of
these methods implicitly relied on the assumption that the value they
were operating on was *actually a curve*, when this wasn't actually
guaranteed. Now, it's possible to make a bunch of small changes inside
the curve struct methods to account for that, but I think that's worse
than just guaranteeing that the data is valid upstream — sampling is
about as hot a code path as we're going to get in this area, and hitting
an additional branch every time it happens just to check that the struct
contains valid data is probably a waste of resources.
Another way of phrasing this is that even if we're only interested in
solving the crashes, the curve's validity needs to be checked at some
point, and it's almost certainly better to do this once at the point of
construction than every time the curve is sampled.
In cases where the control data is supplied dynamically, users would
already have to deal with empty curve outputs basically not working.
Anecdotally, I ran into this while writing the `cubic_splines` example,
and I think the diff illustrates the improvement pretty nicely — the
code no longer has to anticipate whether the output will be good or not;
it just has to handle the `Result`.
The cost of all this, of course, is that we have to guarantee that the
new invariant is actually maintained whenever we extend the API.
However, for the most part, I don't expect users to want to do much
surgery on the internals of their curves anyway.
# Objective
- The implementation of `update_component_access` for `AnyOf`/`Or` is
kinda weird due to special casing the first filter, let's simplify it;
- Fundamentally we want to fold/reduce the various filters using an OR
operation, however in order to do a proper fold we need a neutral
element for the initial accumulator, which for OR is FALSE. However we
didn't have a way to create a `FilteredAccess` value corresponding to
FALSE and thus the only option was reducing, which special cases the
first element as being the initial accumulator.
This is an alternative to https://github.com/bevyengine/bevy/pull/14026
## Solution
- Introduce `FilteredAccess::empty` as a way to create a
`FilteredAccess` corresponding to the logical proposition FALSE;
- Use it as the initial accumulator for the above operations, allowing
to handle all the elements to fold in the same way.
---
## Migration Guide
- The behaviour of `AnyOf<()>` and `Or<()>` has been changed to match no
archetypes rather than all archetypes to naturally match the
corresponding logical operation. Consider replacing them with `()`
instead.
# Objective
- Fix issue #2611
## Solution
- Add `--generate-link-to-definition` to all the `rustdoc-args` arrays
in the `Cargo.toml`s (for docs.rs)
- Add `--generate-link-to-definition` to the `RUSTDOCFLAGS` environment
variable in the docs workflow (for dev-docs.bevyengine.org)
- Document all the workspace crates in the docs workflow (needed because
otherwise only the source code of the `bevy` package will be included,
making the argument useless)
- I think this also fixes#3662, since it fixes the bug on
dev-docs.bevyengine.org, while on docs.rs it has been fixed for a while
on their side.
---
## Changelog
- The source code viewer on docs.rs now includes links to the
definitions.
# Objective
- `bevy_render` depends on `image 0.25` but uses `image::ImageReader`
which was added only in `image 0.25.2`
- users that have `image 0.25` in their `Cargo.lock` and update to the
latest `bevy_render` may thus get a compilation due to this (at least I
did)
## Solution
- Properly set the correct minimum version of `image` that `bevy_render`
depends on.
# Objective
- `SystemId`'s `Debug` implementation includes its `entity` field twice.
- This was likely an oversight in #11019, since before that PR the
second field was the `PhantomData` one.
## Solution
- Only include it once
Alternatively, this could be changed to match the struct representation
of `SystemId`, thus instructing the formatter to print a named struct
and including the `PhantomData` field.
# Objective
Fix a memory leak in `TextureCache` caused by the internal HashMap never
having unused entries cleared.
This isn't a giant memory leak, given the unused entries are simply
empty vectors. Though, if someone goes and resizes a window a bunch, it
can lead to hundreds/thousands of TextureDescriptor keys adding up in
the hashmap – which isn't ideal.
## Solution
- Only retain hashmap entries that still have textures.
- I also added an `is_empty()` method to `TextureCache`, which is useful
for 3rd-party higher-level caches that might have individual caches by
view entity or texture type, for example.
## Testing
- Verified the examples still work (this is a trivial change)
# Objective
- `bevy_winit` fails to build with just the `serialize` feature.
- Caught by [`flag-frenzy`](https://github.com/TheBevyFlock/flag-frenzy)
in [this
run](https://github.com/TheBevyFlock/flag-frenzy/actions/runs/10087486444/job/27891723948),
using the new, nuanced configuration system!
## Solution
- It was failing because `bevy_winit` did not pass the `serialize` flag
to two of its dependencies: `bevy_input` and `bevy_window`.
- To fix this, add these crates to the feature flag.
## Testing
```bash
# On Linux, you must also specify a backend: `x11` or `wayland`.
# You can do this with `-F serialize,x11`, etc.
cargo check -p bevy_winit --no-default-features -F serialize
```
# Objective
- Fix a confusing panic when the viewport width is non-zero and the
height is 0, `prepare_bloom_textures` tries to create a `4294967295x1`
texture.
## Solution
- Avoid dividing by zero
- Apps still crash after this, but now on a more reasonable error about
the zero-size viewport
## Testing
- I isolated and tested the math. A height of 0 sets `mip_height_ratio`
to `inf`, causing the width to explode if it isn't also 0
# Objective
- `bevy_gltf` does not build with only the
`pbr_multi_layer_material_textures` or `pbr_anisotropy_texture`
features.
- Caught by [`flag-frenzy`](https://github.com/TheBevyFlock/flag-frenzy)
in [this
run](https://github.com/TheBevyFlock/flag-frenzy/actions/runs/10087486444/job/27891723948).
## Solution
- This error was due to the feature not enabling the corresponding
feature in `bevy_pbr`. Adding these flags as a dependency fixes this
error.
## Testing
The following commands fail on `main`, but pass with this PR:
```bash
cargo check -p bevy_gltf --no-default-features -F pbr_multi_layer_material_textures
cargo check -p bevy_gltf --no-default-features -F pbr_anisotropy_texture
```
# Objective
- Made `ViewUniform` fields public so that 3rd-parties can create this
uniform. This is useful for custom pipelines that use custom views (e.g.
views buffered by a particular amount, for example).
# Objective
- Not including bevy's default font shouldn't result in code not
compiling anymore.
- Games may want to load their own default font into the default
`Handle<Font>` and not include bevy's default font, but still use these
convenience impls (https://github.com/bevyengine/bevy/issues/12192
currently makes this a bit inconvenient, but it does work).
## Solution
- Include these impls unconditionally.
- Slightly expand the comment on the `font` field to indicate that a
custom font can be used to override the default font.
- (drive-by: add `#[reflect(Default)]` on `TextSection`, since it was
missing a way to construct it via reflection)
# Objective
- Currently `bevy_ptr::{Ptr, PtrMut}` have `From` implementations from
references.
- These implementations impose an implicit `Sized` bound so `bevy_ptr`
types cannot be created from references to slices and trait objects.
- I ran into this trying to use `Ptr<'static>` as an untyped `&'static
dyn Any`, and [had to work around
it](f32b41512c/src/registry.rs (L214-L219)).
## Solution
- Relax the `Sized` bound on the relevant `From` implementations.
# Objective
- Before this fix, the view query in `prepare_mesh2d_view_bind_groups`
matched all views – leading to 2D view bind groups being prepared for 3D
cameras.
## Solution
- Added `With<Camera2d>` to the views query.
## Testing
- Verified the examples still work.
# Objective
Allow interoperation between `Isometry3d` and the transform types from
bevy_transform. At least in the short term, the primary goal is to allow
the extraction of isometries from transform components by users.
## Solution
- Add explicit `from_isometry`/`to_isometry` methods to `Transform`.
- Add explicit `from_isometry`/`to_isometry` methods to
`GlobalTransform`. The former is hidden (primarily for internal use),
and the latter has the caveats originating in
[`Affine3A::to_scale_rotation_translation`](https://docs.rs/glam/latest/glam/f32/struct.Affine3A.html#method.to_scale_rotation_translation).
- Implement the `TransformPoint` trait for `Isometry3d`.
# Objective
Fix erroneous hue mixing in `Lcha` and `Oklcha`. Purple + Red == Green
is the current behavior.
## Solution
Use `crate::color_ops::lerp_hue` to handle the wrap-around at 360
degrees, the same way that `Hsla`, `Hsva`, and `Hwba` do it.
## Testing
Game jamming, but tested that the workaround below produces
correct-looking colors in my jam game.
