# Objective
The table [here](https://github.com/nagisa/rust_tracy_client) shows
which versions of [Tracy](https://github.com/wolfpld/tracy) should be
used combined with which Rust deps.
Reading `bevy_log`'s `Cargo.toml` can be slightly confusing since the
exact versions are not picked from the same row.
Reading the produced `Cargo.lock` when building a Bevy example shows
that it's the most recent row that is resolved, but this should be more
clearly understood without needing to check the lock file.
## Solution
- Specify versions from the compatibility table including patch version
Signed-off-by: Torstein Grindvik <torstein.grindvik@muybridge.com>
Co-authored-by: Torstein Grindvik <torstein.grindvik@muybridge.com>
# Objective
- `bevy_gizmos` cannot work if both `bevy_sprite` and `bevy_pbr` are
disabled.
- It silently fails to render, making it difficult to debug.
- Fixes#10984
## Solution
- Log an error message when `GizmoPlugin` is registered.
## Alternatives
I chose to log an error message, since it seemed the least intrusive of
potential solutions. Some alternatives include:
- Choosing one dependency as the default, neglecting the other. (#11035)
- Raising a compile error when neither dependency is enabled. ([See my
original
comment](https://github.com/bevyengine/bevy/issues/10984#issuecomment-1873420426))
- Raising a compile warning using a macro hack. ([Pre-RFC - Add
compile_warning!
macro](https://internals.rust-lang.org/t/pre-rfc-add-compile-warning-macro/9370/7?u=bd103))
- Logging a warning instead of an error.
- _This might be the better option. Let me know if I should change it._
---
## Changelog
- `bevy_gizmos` will now log an error if neither `bevy_pbr` nor
`bevy_sprite` are enabled.
# Objective
Unify flycam-style camera controller from the examples.
`parallax_mapping` controller was kept as is.
## Solution
Fixed some mouse movement & cursor grabbing related issues too.
# Objective
- Implementing `Default` for
[`CubicCurve`](https://docs.rs/bevy/latest/bevy/math/cubic_splines/struct.CubicCurve.html)
does not make sense because it cannot be mutated after creation.
- Closes#11209.
- Alternative to #11211.
## Solution
- Remove `Default` from `CubicCurve`'s derive statement.
Based off of @mockersf comment
(https://github.com/bevyengine/bevy/pull/11211#issuecomment-1880088036):
> CubicCurve can't be updated once created... I would prefer to remove
the Default impl as it doesn't make sense
---
## Changelog
- Removed the `Default` implementation for `CubicCurve`.
## Migration Guide
- Remove `CubicCurve` from any structs that implement `Default`.
- Wrap `CubicCurve` in a new type and provide your own default.
```rust
#[derive(Deref)]
struct MyCubicCurve<P: Point>(pub CubicCurve<P>);
impl Default for MyCubicCurve<Vec2> {
fn default() -> Self {
let points = [[
vec2(-1.0, -20.0),
vec2(3.0, 2.0),
vec2(5.0, 3.0),
vec2(9.0, 8.0),
]];
Self(CubicBezier::new(points).to_curve())
}
}
```
Based on discussion after #11268 was merged:
Instead of panicking should the impl of `TypeId::hash` change
significantly, have a fallback and detect this in a test.
# Objective
In #11330 I found out that `Parent::get` didn't get inlined, **even with
LTO on**!
This means that just to access a field, we have an instruction cache
invalidation, we will move some registers to the stack, will jump to new
instructions, move the field into a register, then do the same dance in
the other direction to go back to the call site.
## Solution
Mark trivial functions as `#[inline]`.
`inline(always)` may increase compilation time proportional to how many
time the function is called **and the size of the function marked with
`inline`**. Since we mark as `inline` functions that consists in a
single instruction, the cost is absolutely negligible.
I also took the opportunity to `inline` other functions. I'm not as
confident that marking functions calling other functions as `inline`
works similarly to very simple functions, so I used `inline` over
`inline(always)`, which doesn't have the same downsides as
`inline(always)`.
More information on inlining in rust:
https://nnethercote.github.io/perf-book/inlining.html
Not always, but skip it if the new length is smaller.
For context, `path_cache` is a `Vec<Vec<Option<Entity>>>`.
# Objective
Previously, when setting a new length to the `path_cache`, we would:
1. Deallocate all existing `Vec<Option<Entity>>`
2. Deallocate the `path_cache`
3. Allocate a new `Vec<Vec<Option<Entity>>>`, where each item is an
empty `Vec`, and would have to be allocated when pushed to.
This is a lot of allocations!
## Solution
Use
[`Vec::resize_with`](https://doc.rust-lang.org/stable/std/vec/struct.Vec.html#method.resize_with).
With this change, what occurs is:
1. We `clear` each `Vec<Option<Entity>>`, keeping the allocation, but
making the memory of each `Vec` re-usable
2. We only append new `Vec` to `path_cache` when it is too small.
* Fixes#11328
### Note on performance
I didn't benchmark it, I just ran a diff on the generated assembly (ran
with `--profile stress-test` and `--native`). I found this PR has 20
less instructions in `apply_animation` (out of 2504).
Though on a purely abstract level, I can deduce this leads to less
allocation.
More information on profiling allocations in rust:
https://nnethercote.github.io/perf-book/heap-allocations.html
## Future work
I think a [jagged vec](https://en.wikipedia.org/wiki/Jagged_array) would
be much more pertinent. Because it allocates everything in a single
contiguous buffer.
This would avoid dancing around allocations, and reduces the overhead of
one `*mut T` and two `usize` per row, also removes indirection,
improving cache efficiency. I think it would both improve code quality
and performance.
# Objective
`TypeId` contains a high-quality hash. Whenever a lookup based on a
`TypeId` is performed (e.g. to insert/remove components), the hash is
run through a second hash function. This is unnecessary.
## Solution
Skip re-hashing `TypeId`s.
In my
[testing](https://gist.github.com/SpecificProtagonist/4b49ad74c6b82b0aedd3b4ea35121be8),
this improves lookup performance consistently by 10%-15% (of course, the
lookup is only a small part of e.g. a bundle insertion).
# Objective
While working on #10832, I found this code very dense and hard to
understand.
I was not confident in my fix (or the correctness of the existing code).
## Solution
Clean up, test and document the code used in the `apply_animation`
system.
I also added a pair of length-related utility methods to `Keyframes` for
easier testing. They seemed generically useful, so I made them pub.
## Changelog
- Added `VariableCurve::find_current_keyframe` method.
- Added `Keyframes::len` and `is_empty` methods.
---------
Co-authored-by: Alice Cecile <alice.i.cecil@gmail.com>
# Objective
The purpose of this PR is to begin putting together a unified identifier
structure that can be used by entities and later components (as
entities) as well as relationship pairs for relations, to enable all of
these to be able to use the same storages. For the moment, to keep
things small and focused, only `Entity` is being changed to make use of
the new `Identifier` type, keeping `Entity`'s API and
serialization/deserialization the same. Further changes are for
follow-up PRs.
## Solution
`Identifier` is a wrapper around `u64` split into two `u32` segments
with the idea of being generalised to not impose restrictions on
variants. That is for `Entity` to do. Instead, it is a general API for
taking bits to then merge and map into a `u64` integer. It exposes
low/high methods to return the two value portions as `u32` integers,
with then the MSB masked for usage as a type flag, enabling entity kind
discrimination and future activation/deactivation semantics.
The layout in this PR for `Identifier` is described as below, going from
MSB -> LSB.
```
|F| High value | Low value |
|_|_______________________________|________________________________|
|1| 31 | 32 |
F = Bit Flags
```
The high component in this implementation has only 31 bits, but that
still leaves 2^31 or 2,147,483,648 values that can be stored still, more
than enough for any generation/relation kinds/etc usage. The low part is
a full 32-bit index. The flags allow for 1 bit to be used for
entity/pair discrimination, as these have different usages for the
low/high portions of the `Identifier`. More bits can be reserved for
more variants or activation/deactivation purposes, but this currently
has no use in bevy.
More bits could be reserved for future features at the cost of bits for
the high component, so how much to reserve is up for discussion. Also,
naming of the struct and methods are also subject to further
bikeshedding and feedback.
Also, because IDs can have different variants, I wonder if
`Entity::from_bits` needs to return a `Result` instead of potentially
panicking on receiving an invalid ID.
PR is provided as an early WIP to obtain feedback and notes on whether
this approach is viable.
---
## Changelog
### Added
New `Identifier` struct for unifying IDs.
### Changed
`Entity` changed to use new `Identifier`/`IdentifierMask` as the
underlying ID logic.
---------
Co-authored-by: Alice Cecile <alice.i.cecile@gmail.com>
Co-authored-by: vero <email@atlasdostal.com>
# Objective
- It is common to run a system only when the clock is paused or not
paused, but this run condition doesn't exist.
## Solution
- Add the "paused" run condition.
---
## Changelog
- Systems can now be scheduled to run only if the clock is paused or not
using `.run_if(paused())` or `.run_if(not(paused()))`.
---------
Co-authored-by: radiish <cb.setho@gmail.com>
# Objective
Fixes https://github.com/bevyengine/bevy/issues/11222
## Solution
SSAO's sample_mip_level was always giving negative values because it was
in UV space (0..1) when it needed to be in pixel units (0..resolution).
Fixing it so it properly samples lower mip levels when appropriate is a
pretty large speedup (~3.2ms -> ~1ms at 4k, ~507us-> 256us at 1080p on a
6800xt), and I didn't notice any obvious visual quality differences.
---------
Co-authored-by: Alice Cecile <alice.i.cecile@gmail.com>
# Objective
Implement bounding volume trait and the 4 types from
https://github.com/bevyengine/bevy/issues/10570. I will add intersection
tests in a future PR.
## Solution
Implement mostly everything as written in the issue, except:
- Intersection is no longer a method on the bounding volumes, but a
separate trait.
- I implemented a `visible_area` since it's the most common usecase to
care about the surface that could collide with cast rays.
- Maybe we want both?
---
## Changelog
- Added bounding volume types to bevy_math
---------
Co-authored-by: Alice Cecile <alice.i.cecile@gmail.com>
# Objective
- Example in error B0003 is not failing anymore after #9822
## Solution
- Update the example code so that is always fail
- Also update logs and instructions on how to debug as it's easier now
---------
Co-authored-by: Alice Cecile <alice.i.cecile@gmail.com>
# Objective
In the past `winit:🪟:Window` was not Send + Sync on web.
https://github.com/rust-windowing/winit/pull/2834 made
`winit:🪟:Window` Sync + Send so Bevy's `unsafe impl` is no longer
necessary.
## Solution
Remove the unsafe impls.
# Objective
This dependency is seemingly no longer used directly after #7267.
Unfortunately, this doesn't fix us having versions of `event-listener`
in our tree.
Closes#10654
## Solution
Remove it, see if anything breaks.
# Objective
- Since #11227, Bevy doesn't work on mobile anymore. Windows are not
created.
## Solution
- Create initial window on mobile after the initial `Resume` event.
macOS is included because it's excluded from the other initial window
creation and I didn't want it to feel alone. Also, it makes sense. this
is needed for Android
cfcb6885e3/crates/bevy_winit/src/lib.rs (L152)
- request redraw during plugin initialisation (needed for WebGPU)
- request redraw when receiving `AboutToWait` instead of at the end of
the event handler. request to redraw during a `RedrawRequested` event
are ignored on iOS
# Objective
Issue #10243: rendering multiple triangles in the same place results in
flickering.
## Solution
Considered these alternatives:
- `depth_bias` may not work, because of high number of entities, so
creating a material per entity is practically not possible
- rendering at slightly different positions does not work, because when
camera is far, float rounding causes the same issues (edit: assuming we
have to use the same `depth_bias`)
- considered implementing deterministic operation like
`query.par_iter().flat_map(...).collect()` to be used in
`check_visibility` system (which would solve the issue since query is
deterministic), and could not figure out how to make it as cheap as
current approach with thread-local collectors (#11249)
So adding an option to sort entities after `check_visibility` system
run.
Should not be too bad, because after visibility check, only a handful
entities remain.
This is probably not the only source of non-determinism in Bevy, but
this is one I could find so far. At least it fixes the repro example.
## Changelog
- `DeterministicRenderingConfig` option to enable deterministic
rendering
## Test
<img width="1392" alt="image"
src="https://github.com/bevyengine/bevy/assets/28969/c735bce1-3a71-44cd-8677-c19f6c0ee6bd">
---------
Co-authored-by: Alice Cecile <alice.i.cecile@gmail.com>
# Objective
I want to run a system once after a given delay.
- First, I tried using the `on_timer` run condition, but it uses a
repeating timer, causing the system to run multiple times.
- Next, I tried combining the `on_timer` with the `run_once` run
condition. However, this causes the timer to *tick* only once, so the
system is never executed.
## Solution
- ~~Replace `on_timer` by `on_time_interval` and `on_real_timer` by
`on_real_time_interval` to clarify the meaning (the old ones are
deprecated to avoid a breaking change).~~ (Reverted according to
feedback)
- Add `once_after_delay` and `once_after_real_delay` to run the system
exactly once after the delay, using `TimerMode::Once`.
- Add `repeating_after_delay` and `repeating_after_real_delay` to run
the system indefinitely after the delay, using `Timer::finished` instead
of `Timer::just_finished`.
---
## Changelog
### Added
- `once_after_delay` and `once_after_real_delay` run conditions to run
the system exactly once after the delay, using `TimerMode::Once`.
- `repeating_after_delay` and `repeating_after_real_delay` run
conditions to run the system indefinitely after the delay, using
`Timer::finished` instead of `Timer::just_finished`.
# Objective
- Implements change described in
https://github.com/bevyengine/bevy/issues/3022
- Goal is to allow Entity to benefit from niche optimization, especially
in the case of Option<Entity> to reduce memory overhead with structures
with empty slots
## Discussion
- First PR attempt: https://github.com/bevyengine/bevy/pull/3029
- Discord:
https://discord.com/channels/691052431525675048/1154573759752183808/1154573764240093224
## Solution
- Change `Entity::generation` from u32 to NonZeroU32 to allow for niche
optimization.
- The reason for changing generation rather than index is so that the
costs are only encountered on Entity free, instead of on Entity alloc
- There was some concern with generations being used, due to there being
some desire to introduce flags. This was more to do with the original
retirement approach, however, in reality even if generations were
reduced to 24-bits, we would still have 16 million generations available
before wrapping and current ideas indicate that we would be using closer
to 4-bits for flags.
- Additionally, another concern was the representation of relationships
where NonZeroU32 prevents us using the full address space, talking with
Joy it seems unlikely to be an issue. The majority of the time these
entity references will be low-index entries (ie. `ChildOf`, `Owes`),
these will be able to be fast lookups, and the remainder of the range
can use slower lookups to map to the address space.
- It has the additional benefit of being less visible to most users,
since generation is only ever really set through `from_bits` type
methods.
- `EntityMeta` was changed to match
- On free, generation now explicitly wraps:
- Originally, generation would panic in debug mode and wrap in release
mode due to using regular ops.
- The first attempt at this PR changed the behavior to "retire" slots
and remove them from use when generations overflowed. This change was
controversial, and likely needs a proper RFC/discussion.
- Wrapping matches current release behaviour, and should therefore be
less controversial.
- Wrapping also more easily migrates to the retirement approach, as
users likely to exhaust the exorbitant supply of generations will code
defensively against aliasing and that defensive code is less likely to
break than code assuming that generations don't wrap.
- We use some unsafe code here when wrapping generations, to avoid
branch on NonZeroU32 construction. It's guaranteed safe due to how we
perform wrapping and it results in significantly smaller ASM code.
- https://godbolt.org/z/6b6hj8PrM
## Migration
- Previous `bevy_scene` serializations have a high likelihood of being
broken, as they contain 0th generation entities.
## Current Issues
- `Entities::reserve_generations` and `EntityMapper` wrap now, even in
debug - although they technically did in release mode already so this
probably isn't a huge issue. It just depends if we need to change
anything here?
---------
Co-authored-by: Natalie Baker <natalie.baker@advancednavigation.com>
Update to `glam` 0.25, `encase` 0.7 and `hexasphere` to 10.0
## Changelog
Added the `FloatExt` trait to the `bevy_math` prelude which adds `lerp`,
`inverse_lerp` and `remap` methods to the `f32` and `f64` types.
# Objective
When you have no idea what to put after `#` when loading an asset, error
message may help.
## Solution
Add all labels to the error message.
## Test plan
Modified `anti_alias` example to put incorrect label, the error is:
```
2024-01-08T07:41:25.462287Z ERROR bevy_asset::server: The file at 'models/FlightHelmet/FlightHelmet.gltf' does not contain the labeled asset 'Rrrr'; it contains the following 25 assets: 'Material0', 'Material1', 'Material2', 'Material3', 'Material4', 'Material5', 'Mesh0', 'Mesh0/Primitive0', 'Mesh1', 'Mesh1/Primitive0', 'Mesh2', 'Mesh2/Primitive0', 'Mesh3', 'Mesh3/Primitive0', 'Mesh4', 'Mesh4/Primitive0', 'Mesh5', 'Mesh5/Primitive0', 'Node0', 'Node1', 'Node2', 'Node3', 'Node4', 'Node5', 'Scene0'
```
# Objective
`Column` unconditionally requires three separate allocations: one for
the data, and two for the tick Vecs. The tick Vecs aren't really needed
for Resources, so we're allocating a bunch of one-element Vecs, and it
costs two extra dereferences when fetching/inserting/removing resources.
## Solution
Drop one level lower in `ResourceData` and directly store a `BlobVec`
and two `UnsafeCell<Tick>`s. This should significantly shrink
`ResourceData` (exchanging 6 usizes for 2 u32s), removes the need to
dereference two separate ticks when inserting/removing/fetching
resources, and can significantly decrease the number of small
allocations the ECS makes by default.
This tentatively might have a non-insignificant impact on the CPU cost
for rendering since we're constantly fetching resources in draw
functions, depending on how aggressively inlined the functions are.
This requires reimplementing some of the unsafe functions that `Column`
wraps, but it also allows us to delete a few Column APIs that were only
used for Resources, so the total amount of unsafe we're maintaining
shouldn't change significantly.
---------
Co-authored-by: Joseph <21144246+JoJoJet@users.noreply.github.com>
# Objective
When creating a normalized direction from a vector, it can be useful to
get both the direction *and* the original length of the vector.
This came up when I was recreating some Parry APIs using bevy_math, and
doing it manually is quite painful. Nalgebra calls this method
[`Unit::try_new_and_get`](https://docs.rs/nalgebra/latest/nalgebra/base/struct.Unit.html#method.try_new_and_get).
## Solution
Add a `new_and_length` method to `Direction2d` and `Direction3d`.
Usage:
```rust
if let Ok((direction, length)) = Direction2d::new_and_length(Vec2::X * 10.0) {
assert_eq!(direction, Vec2::X);
assert_eq!(length, 10.0);
}
```
I'm open to different names, couldn't come up with a perfectly clear one
that isn't too long. My reasoning with the current name is that it's
like using `new` and calling `length` on the original vector.
# Objective
Re this comment:
https://github.com/bevyengine/bevy/pull/11141#issuecomment-1872455313
Since https://github.com/bevyengine/bevy/pull/9822, Bevy automatically
inserts `apply_deferred` between systems with dependencies where needed,
so manually inserted `apply_deferred` doesn't to anything useful, and in
current state this example does more harm than good.
## Solution
The example can be modified with removal of automatic `apply_deferred`
insertion, but that would immediately upgrade this example from beginner
level, to upper intermediate. Most users don't need to disable automatic
sync point insertion, and remaining few who do probably already know how
it works.
CC @hymm
# Objective
In #9604 we removed the ability to define an `EntityCommand` as
`fn(Entity, &mut World)`. However I have since realized that `fn(Entity,
&mut World)` is an incredibly expressive and powerful way to define a
command for an entity that may or may not exist (`fn(EntityWorldMut)`
only works on entities that are alive).
## Solution
Support `EntityCommand`s in the style of `fn(Entity, &mut World)`, as
well as `fn(EntityWorldMut)`. Use a marker generic on the
`EntityCommand` trait to allow multiple impls.
The second commit in this PR replaces all of the internal command
definitions with ones using `fn` definitions. This is mostly just to
show off how expressive this style of command is -- we can revert this
commit if we'd rather avoid breaking changes.
---
## Changelog
Re-added support for expressively defining an `EntityCommand` as a
function that takes `Entity, &mut World`.
## Migration Guide
All `Command` types in `bevy_ecs`, such as `Spawn`, `SpawnBatch`,
`Insert`, etc., have been made private. Use the equivalent methods on
`Commands` or `EntityCommands` instead.
# Objective
- Make it possible to react to arbitrary state changes
- this will be useful regardless of the other changes to states
currently being discussed
## Solution
- added `StateTransitionEvent<S>` struct
- previously, this would have been impossible:
```rs
#[derive(States, Eq, PartialEq, Hash, Copy, Clone, Default)]
enum MyState {
#[default]
Foo,
Bar(MySubState),
}
enum MySubState {
Spam,
Eggs,
}
app.add_system(Update, on_enter_bar);
fn on_enter_bar(trans: EventReader<StateTransition<MyState>>){
for (befoare, after) in trans.read() {
match before, after {
MyState::Foo, MyState::Bar(_) => info!("detected transition foo => bar");
_, _ => ();
}
}
}
```
---
## Changelog
- Added
- `StateTransitionEvent<S>` - Fired on state changes of `S`
## Migration Guide
N/A no breaking changes
---------
Co-authored-by: Federico Rinaldi <gisquerin@gmail.com>
# Motivation
When spawning entities into a scene, it is very common to create assets
like meshes and materials and to add them via asset handles. A common
setup might look like this:
```rust
fn setup(
mut commands: Commands,
mut meshes: ResMut<Assets<Mesh>>,
mut materials: ResMut<Assets<StandardMaterial>>,
) {
commands.spawn(PbrBundle {
mesh: meshes.add(Mesh::from(shape::Cube { size: 1.0 })),
material: materials.add(StandardMaterial::from(Color::RED)),
..default()
});
}
```
Let's take a closer look at the part that adds the assets using `add`.
```rust
mesh: meshes.add(Mesh::from(shape::Cube { size: 1.0 })),
material: materials.add(StandardMaterial::from(Color::RED)),
```
Here, "mesh" and "material" are both repeated three times. It's very
explicit, but I find it to be a bit verbose. In addition to being more
code to read and write, the extra characters can sometimes also lead to
the code being formatted to span multiple lines even though the core
task, adding e.g. a primitive mesh, is extremely simple.
A way to address this is by using `.into()`:
```rust
mesh: meshes.add(shape::Cube { size: 1.0 }.into()),
material: materials.add(Color::RED.into()),
```
This is fine, but from the names and the type of `meshes`, we already
know what the type should be. It's very clear that `Cube` should be
turned into a `Mesh` because of the context it's used in. `.into()` is
just seven characters, but it's so common that it quickly adds up and
gets annoying.
It would be nice if you could skip all of the conversion and let Bevy
handle it for you:
```rust
mesh: meshes.add(shape::Cube { size: 1.0 }),
material: materials.add(Color::RED),
```
# Objective
Make adding assets more ergonomic by making `Assets::add` take an `impl
Into<A>` instead of `A`.
## Solution
`Assets::add` now takes an `impl Into<A>` instead of `A`, so e.g. this
works:
```rust
commands.spawn(PbrBundle {
mesh: meshes.add(shape::Cube { size: 1.0 }),
material: materials.add(Color::RED),
..default()
});
```
I also changed all examples to use this API, which increases consistency
as well because `Mesh::from` and `into` were being used arbitrarily even
in the same file. This also gets rid of some lines of code because
formatting is nicer.
---
## Changelog
- `Assets::add` now takes an `impl Into<A>` instead of `A`
- Examples don't use `T::from(K)` or `K.into()` when adding assets
## Migration Guide
Some `into` calls that worked previously might now be broken because of
the new trait bounds. You need to either remove `into` or perform the
conversion explicitly with `from`:
```rust
// Doesn't compile
let mesh_handle = meshes.add(shape::Cube { size: 1.0 }.into()),
// These compile
let mesh_handle = meshes.add(shape::Cube { size: 1.0 }),
let mesh_handle = meshes.add(Mesh::from(shape::Cube { size: 1.0 })),
```
## Concerns
I believe the primary concerns might be:
1. Is this too implicit?
2. Does this increase codegen bloat?
Previously, the two APIs were using `into` or `from`, and now it's
"nothing" or `from`. You could argue that `into` is slightly more
explicit than "nothing" in cases like the earlier examples where a
`Color` gets converted to e.g. a `StandardMaterial`, but I personally
don't think `into` adds much value even in this case, and you could
still see the actual type from the asset type.
As for codegen bloat, I doubt it adds that much, but I'm not very
familiar with the details of codegen. I personally value the user-facing
code reduction and ergonomics improvements that these changes would
provide, but it might be worth checking the other effects in more
detail.
Another slight concern is migration pain; apps might have a ton of
`into` calls that would need to be removed, and it did take me a while
to do so for Bevy itself (maybe around 20-40 minutes). However, I think
the fact that there *are* so many `into` calls just highlights that the
API could be made nicer, and I'd gladly migrate my own projects for it.
# Objective
This pull request implements *reflection probes*, which generalize
environment maps to allow for multiple environment maps in the same
scene, each of which has an axis-aligned bounding box. This is a
standard feature of physically-based renderers and was inspired by [the
corresponding feature in Blender's Eevee renderer].
## Solution
This is a minimal implementation of reflection probes that allows
artists to define cuboid bounding regions associated with environment
maps. For every view, on every frame, a system builds up a list of the
nearest 4 reflection probes that are within the view's frustum and
supplies that list to the shader. The PBR fragment shader searches
through the list, finds the first containing reflection probe, and uses
it for indirect lighting, falling back to the view's environment map if
none is found. Both forward and deferred renderers are fully supported.
A reflection probe is an entity with a pair of components, *LightProbe*
and *EnvironmentMapLight* (as well as the standard *SpatialBundle*, to
position it in the world). The *LightProbe* component (along with the
*Transform*) defines the bounding region, while the
*EnvironmentMapLight* component specifies the associated diffuse and
specular cubemaps.
A frequent question is "why two components instead of just one?" The
advantages of this setup are:
1. It's readily extensible to other types of light probes, in particular
*irradiance volumes* (also known as ambient cubes or voxel global
illumination), which use the same approach of bounding cuboids. With a
single component that applies to both reflection probes and irradiance
volumes, we can share the logic that implements falloff and blending
between multiple light probes between both of those features.
2. It reduces duplication between the existing *EnvironmentMapLight* and
these new reflection probes. Systems can treat environment maps attached
to cameras the same way they treat environment maps applied to
reflection probes if they wish.
Internally, we gather up all environment maps in the scene and place
them in a cubemap array. At present, this means that all environment
maps must have the same size, mipmap count, and texture format. A
warning is emitted if this restriction is violated. We could potentially
relax this in the future as part of the automatic mipmap generation
work, which could easily do texture format conversion as part of its
preprocessing.
An easy way to generate reflection probe cubemaps is to bake them in
Blender and use the `export-blender-gi` tool that's part of the
[`bevy-baked-gi`] project. This tool takes a `.blend` file containing
baked cubemaps as input and exports cubemap images, pre-filtered with an
embedded fork of the [glTF IBL Sampler], alongside a corresponding
`.scn.ron` file that the scene spawner can use to recreate the
reflection probes.
Note that this is intentionally a minimal implementation, to aid
reviewability. Known issues are:
* Reflection probes are basically unsupported on WebGL 2, because WebGL
2 has no cubemap arrays. (Strictly speaking, you can have precisely one
reflection probe in the scene if you have no other cubemaps anywhere,
but this isn't very useful.)
* Reflection probes have no falloff, so reflections will abruptly change
when objects move from one bounding region to another.
* As mentioned before, all cubemaps in the world of a given type
(diffuse or specular) must have the same size, format, and mipmap count.
Future work includes:
* Blending between multiple reflection probes.
* A falloff/fade-out region so that reflected objects disappear
gradually instead of vanishing all at once.
* Irradiance volumes for voxel-based global illumination. This should
reuse much of the reflection probe logic, as they're both GI techniques
based on cuboid bounding regions.
* Support for WebGL 2, by breaking batches when reflection probes are
used.
These issues notwithstanding, I think it's best to land this with
roughly the current set of functionality, because this patch is useful
as is and adding everything above would make the pull request
significantly larger and harder to review.
---
## Changelog
### Added
* A new *LightProbe* component is available that specifies a bounding
region that an *EnvironmentMapLight* applies to. The combination of a
*LightProbe* and an *EnvironmentMapLight* offers *reflection probe*
functionality similar to that available in other engines.
[the corresponding feature in Blender's Eevee renderer]:
https://docs.blender.org/manual/en/latest/render/eevee/light_probes/reflection_cubemaps.html
[`bevy-baked-gi`]: https://github.com/pcwalton/bevy-baked-gi
[glTF IBL Sampler]: https://github.com/KhronosGroup/glTF-IBL-Sampler
# Objective
- After https://github.com/bevyengine/bevy/pull/11227, example showcase
timeouts
- `ReactiveLowPower` now can wait indefinitely depending on "platform
specifics"
## Solution
- Patch desktop mode in example showcase to use default mode which is
always `Continuous`
# Objective
- Describe the objective or issue this PR addresses.
- If you're fixing a specific issue, say "Fixes #X".
Removes the old third party plugin guidelines section and related links
since it is being moved to the Bevy book in the site per
https://github.com/bevyengine/bevy-website/pull/848 and
https://github.com/bevyengine/bevy-website/issues/298
## Solution
- Describe the solution used to achieve the objective above.
Removes the old links and files.
# Objective
`SceneSpawner::spawn_dynamic_sync` currently returns `()` on success,
which is inconsistent with the other `SceneSpawner::spawn_` methods that
all return an `InstanceId`. We need this ID to do useful work with the
newly-created data.
## Solution
Updated `SceneSpawner::spawn_dynamic_sync` to return `Result<InstanceId,
SceneSpawnError>` instead of `Result<(), SceneSpawnError>`
# Objective
Different platforms use their own implementations of several
mathematical functions (especially transcendental functions like sin,
cos, tan, atan, and so on) to provide hardware-level optimization using
intrinsics. This is good for performance, but bad when you expect
consistent outputs across machines.
[`libm`](https://github.com/rust-lang/libm) is a widely used crate that
provides mathematical functions that don't use intrinsics like `std`
functions. This allows bit-for-bit deterministic math across hardware,
which is crucial for things like cross-platform deterministic physics
simulation.
Glam has the `libm` feature for using [`libm` for the
math](d2871a151b/src/f32/math.rs (L35))
in its own types. This would be nice to expose as a feature in
`bevy_math`.
## Solution
Add `libm` feature to `bevy_math`. We could name it something like
`enhanced-determinism`, but this wouldn't be accurate for the rest of
Bevy, so I think just `libm` is more fitting and explicit.
# Objective
- Since #10702, the way bevy updates the window leads to major slowdowns
as seen in
- #11122
- #11220
- Slow is bad, furthermore, _very_ slow is _very_ bad. We should fix
this issue.
## Solution
- Move the app update code into the `Event::WindowEvent { event:
WindowEvent::RedrawRequested }` branch of the event loop.
- Run `window.request_redraw()` When `runner_state.redraw_requested`
- Instead of swapping `ControlFlow` between `Poll` and `Wait`, we always
keep it at `Wait`, and use `window.request_redraw()` to schedule an
immediate call to the event loop.
- `runner_state.redraw_requested` is set to `true` when
`UpdateMode::Continuous` and when a `RequestRedraw` event is received.
- Extract the redraw code into a separate function, because otherwise
I'd go crazy with the indentation level.
- Fix#11122.
## Testing
I tested the WASM builds as follow:
```sh
cargo run -p build-wasm-example -- --api webgl2 bevymark
python -m http.server --directory examples/wasm/ 8080
# Open browser at http://localhost:8080
```
On main, even spawning a couple sprites is super choppy. Even if it says
"300 FPS". While on this branch, it is smooth as butter.
I also found that it fixes all choppiness on window resize (tested on
Linux/X11). This was another issue from #10702 IIRC.
So here is what I tested:
- On `wasm`: `many_foxes` and `bevymark`, with `argh::from_env()`
commented out, otherwise we get a cryptic error.
- Both with `PresentMode::AutoVsync` and `PresentMode::AutoNoVsync`
- On main, it is consistently choppy.
- With this PR, the visible frame rate is consistent with the diagnostic
numbers
- On native (linux/x11) I ran similar tests, making sure that
`AutoVsync` limits to monitor framerate, and `AutoNoVsync` doesn't.
## Future work
Code could be improved, I wanted a quick solution easy to review, but we
really need to make the code more accessible.
- #9768
- ~~**`WinitSettings::desktop_app()` is completely borked.**~~ actually
broken on main as well
### Review guide
Consider enable the non-whitespace diff to see the _real_ change set.
# Objective
When `BlobVec::reserve` is called with an argument causing capacity
overflow, in release build capacity overflow is ignored, and capacity is
decreased.
I'm not sure it is possible to exploit this issue using public API of
`bevy_ecs`, but better fix it anyway.
## Solution
Check for capacity overflow.
# Objective
- Since #10520, assets are unloaded from RAM by default. This breaks a
number of scenario:
- using `load_folder`
- loading a gltf, then going through its mesh to transform them /
compute a collider / ...
- any assets/subassets scenario should be `Keep` as you can't know what
the user will do with the assets
- android suspension, where GPU memory is unloaded
- Alternative to #11202
## Solution
- Keep assets on CPU memory by default
