# Objective
Fixes#9367.
Yet another follow-up to #16547.
These traits were initially based on `Borrow<Entity>` because that trait
was what they were replacing, and it felt close enough in meaning.
However, they ultimately don't quite match: `borrow` always returns
references, whereas `EntityBorrow` always returns a plain `Entity`.
Additionally, `EntityBorrow` can imply that we are borrowing an `Entity`
from the ECS, which is not what it does.
Due to its safety contract, `TrustedEntityBorrow` is important an
important and widely used trait for `EntitySet` functionality.
In contrast, the safe `EntityBorrow` does not see much use, because even
outside of `EntitySet`-related functionality, it is a better idea to
accept `TrustedEntityBorrow` over `EntityBorrow`.
Furthermore, as #9367 points out, abstracting over returning `Entity`
from pointers/structs that contain it can skip some ergonomic friction.
On top of that, there are aspects of #18319 and #18408 that are relevant
to naming:
We've run into the issue that relying on a type default can switch
generic order. This is livable in some contexts, but unacceptable in
others.
To remedy that, we'd need to switch to a type alias approach:
The "defaulted" `Entity` case becomes a
`UniqueEntity*`/`Entity*Map`/`Entity*Set` alias, and the base type
receives a more general name. `TrustedEntityBorrow` does not mesh
clearly with sensible base type names.
## Solution
Replace any `EntityBorrow` bounds with `TrustedEntityBorrow`.
+
Rename them as such:
`EntityBorrow` -> `ContainsEntity`
`TrustedEntityBorrow` -> `EntityEquivalent`
For `EntityBorrow` we produce a change in meaning; We designate it for
types that aren't necessarily strict wrappers around `Entity` or some
pointer to `Entity`, but rather any of the myriad of types that contain
a single associated `Entity`.
This pattern can already be seen in the common `entity`/`id` methods
across the engine.
We do not mean for `ContainsEntity` to be a trait that abstracts input
API (like how `AsRef<T>` is often used, f.e.), because eliding
`entity()` would be too implicit in the general case.
We prefix "Contains" to match the intuition of a struct with an `Entity`
field, like some contain a `length` or `capacity`.
It gives the impression of structure, which avoids the implication of a
relationship to the `ECS`.
`HasEntity` f.e. could be interpreted as "a currently live entity",
As an input trait for APIs like #9367 envisioned, `TrustedEntityBorrow`
is a better fit, because it *does* restrict itself to strict wrappers
and pointers. Which is why we replace any
`EntityBorrow`/`ContainsEntity` bounds with
`TrustedEntityBorrow`/`EntityEquivalent`.
Here, the name `EntityEquivalent` is a lot closer to its actual meaning,
which is "A type that is both equivalent to an `Entity`, and forms the
same total order when compared".
Prior art for this is the
[`Equivalent`](https://docs.rs/hashbrown/latest/hashbrown/trait.Equivalent.html)
trait in `hashbrown`, which utilizes both `Borrow` and `Eq` for its one
blanket impl!
Given that we lose the `Borrow` moniker, and `Equivalent` can carry
various meanings, we expand on the safety comment of `EntityEquivalent`
somewhat. That should help prevent the confusion we saw in
[#18408](https://github.com/bevyengine/bevy/pull/18408#issuecomment-2742094176).
The new name meshes a lot better with the type aliasing approach in
#18408, by aligning with the base name `EntityEquivalentHashMap`.
For a consistent scheme among all set types, we can use this scheme for
the `UniqueEntity*` wrapper types as well!
This allows us to undo the switched generic order that was introduced to
`UniqueEntityArray` by its `Entity` default.
Even without the type aliases, I think these renames are worth doing!
## Migration Guide
Any use of `EntityBorrow` becomes `ContainsEntity`.
Any use of `TrustedEntityBorrow` becomes `EntityEquivalent`.
# Objective
Unlike for their helper typers, the import paths for
`unique_array::UniqueEntityArray`, `unique_slice::UniqueEntitySlice`,
`unique_vec::UniqueEntityVec`, `hash_set::EntityHashSet`,
`hash_map::EntityHashMap`, `index_set::EntityIndexSet`,
`index_map::EntityIndexMap` are quite redundant.
When looking at the structure of `hashbrown`, we can also see that while
both `HashSet` and `HashMap` have their own modules, the main types
themselves are re-exported to the crate level.
## Solution
Re-export the types in their shared `entity` parent module, and simplify
the imports where they're used.
# Objective
Fixes#18515
After the recent changes to system param validation, the panic message
for a missing resource is currently:
```
Encountered an error in system `missing_resource_error::res_system`: SystemParamValidationError { skipped: false }
```
Add the parameter type name and a descriptive message, improving the
panic message to:
```
Encountered an error in system `missing_resource_error::res_system`: SystemParamValidationError { skipped: false, message: "Resource does not exist", param: "bevy_ecs::change_detection::Res<missing_resource_error::MissingResource>" }
```
## Solution
Add fields to `SystemParamValidationError` for error context. Include
the `type_name` of the param and a message.
Store them as `Cow<'static, str>` and only format them into a friendly
string in the `Display` impl. This lets us create errors using a
`&'static str` with no allocation or formatting, while still supporting
runtime `String` values if necessary.
Add a unit test that verifies the panic message.
## Future Work
If we change the default error handling to use `Display` instead of
`Debug`, and to use `ShortName` for the system name, the panic message
could be further improved to:
```
Encountered an error in system `res_system`: Parameter `Res<MissingResource>` failed validation: Resource does not exist
```
However, `BevyError` currently includes the backtrace in `Debug` but not
`Display`, and I didn't want to try to change that in this PR.
# Objective
This fixes `NonMesh` draw commands not receiving render-world entities
since
- https://github.com/bevyengine/bevy/pull/17698
This unbreaks item queries for queued non-mesh entities:
```rust
struct MyDrawCommand {
type ItemQuery = Read<DynamicUniformIndex<SomeUniform>>;
// ...
}
```
### Solution
Pass render entity to `NonMesh` draw commands instead of
`Entity::PLACEHOLDER`. This PR also introduces sorting of the `NonMesh`
bin keys like other types, which I assume is the intended behavior.
@pcwalton
## Testing
- Tested on a local project that extensively uses `NonMesh` items.
# Objective
- feature `shader_format_wesl` doesn't compile in Wasm
- once fixed, example `shader_material_wesl` doesn't work in WebGL2
## Solution
- remove special path handling when loading shaders. this seems like a
way to escape the asset folder which we don't want to allow, and can't
compile on android or wasm, and can't work on iOS (filesystem is rooted
there)
- pad material so that it's 16 bits. I couldn't get conditional
compilation to work in wesl for type declaration, it fails to parse
- the shader renders the color `(0.0, 0.0, 0.0, 0.0)` when it's not a
polka dot. this renders as black on WebGPU/metal/..., and white on
WebGL2. change it to `(0.0, 0.0, 0.0, 1.0)` so that it's black
everywhere
* `submit_graph_commands` was incorrectly timing the command buffer
generation tasks as well, and not only the queue submission. Moved the
span to fix that.
* Added a new `command_buffer_generation_tasks` span as a parent for all
the individual command buffer generation tasks that don't run as part of
the Core3d span.
# Objective
Fixes https://github.com/bevyengine/bevy/issues/16586.
## Solution
- Free meshes before allocating new ones (so hopefully the existing
allocation is used, but it's not guaranteed since it might end up
getting used by a smaller mesh first).
- Keep track of modified render assets, and have the mesh allocator free
their allocations.
- Cleaned up some render asset code to make it more understandable,
since it took me several minutes to reverse engineer/remember how it was
supposed to work.
Long term we'll probably want to explicitly reusing allocations for
modified meshes that haven't grown in size, or do delta uploads using a
compute shader or something, but this is an easy fix for the near term.
## Testing
Ran the example provided in the issue. No crash after a few minutes, and
memory usage remains steady.
# Objective
Make it easier to short-circuit system parameter validation.
Simplify the API surface by combining `ValidationOutcome` with
`SystemParamValidationError`.
## Solution
Replace `ValidationOutcome` with `Result<(),
SystemParamValidationError>`. Move the docs from `ValidationOutcome` to
`SystemParamValidationError`.
Add a `skipped` field to `SystemParamValidationError` to distinguish the
`Skipped` and `Invalid` variants.
Use the `?` operator to short-circuit validation in tuples of system
params.
# Objective
When introduced, `Single` was intended to simply be silently skipped,
allowing for graceful and efficient handling of systems during invalid
game states (such as when the player is dead).
However, this also caused missing resources to *also* be silently
skipped, leading to confusing and very hard to debug failures. In
0.15.1, this behavior was reverted to a panic, making missing resources
easier to debug, but largely making `Single` (and `Populated`)
worthless, as they would panic during expected game states.
Ultimately, the consensus is that this behavior should differ on a
per-system-param basis. However, there was no sensible way to *do* that
before this PR.
## Solution
Swap `SystemParam::validate_param` from a `bool` to:
```rust
/// The outcome of system / system param validation,
/// used by system executors to determine what to do with a system.
pub enum ValidationOutcome {
/// All system parameters were validated successfully and the system can be run.
Valid,
/// At least one system parameter failed validation, and an error must be handled.
/// By default, this will result in1 a panic. See [crate::error] for more information.
///
/// This is the default behavior, and is suitable for system params that should *always* be valid,
/// either because sensible fallback behavior exists (like [`Query`] or because
/// failures in validation should be considered a bug in the user's logic that must be immediately addressed (like [`Res`]).
Invalid,
/// At least one system parameter failed validation, but the system should be skipped due to [`ValidationBehavior::Skip`].
/// This is suitable for system params that are intended to only operate in certain application states, such as [`Single`].
Skipped,
}
```
Then, inside of each `SystemParam` implementation, return either Valid,
Invalid or Skipped.
Currently, only `Single`, `Option<Single>` and `Populated` use the
`Skipped` behavior. Other params (like resources) retain their current
failing
## Testing
Messed around with the fallible_params example. Added a pair of tests:
one for panicking when resources are missing, and another for properly
skipping `Single` and `Populated` system params.
## To do
- [x] get https://github.com/bevyengine/bevy/pull/18454 merged
- [x] fix the todo!() in the macro-powered tuple implementation (please
help 🥺)
- [x] test
- [x] write a migration guide
- [x] update the example comments
## Migration Guide
Various system and system parameter validation methods
(`SystemParam::validate_param`, `System::validate_param` and
`System::validate_param_unsafe`) now return and accept a
`ValidationOutcome` enum, rather than a `bool`. The previous `true`
values map to `ValidationOutcome::Valid`, while `false` maps to
`ValidationOutcome::Invalid`.
However, if you wrote a custom schedule executor, you should now respect
the new `ValidationOutcome::Skipped` parameter, skipping any systems
whose validation was skipped. By contrast, `ValidationOutcome::Invalid`
systems should also be skipped, but you should call the
`default_error_handler` on them first, which by default will result in a
panic.
If you are implementing a custom `SystemParam`, you should consider
whether failing system param validation is an error or an expected
state, and choose between `Invalid` and `Skipped` accordingly. In Bevy
itself, `Single` and `Populated` now once again skip the system when
their conditions are not met. This is the 0.15.0 behavior, but stands in
contrast to the 0.15.1 behavior, where they would panic.
---------
Co-authored-by: MiniaczQ <xnetroidpl@gmail.com>
Co-authored-by: Dmytro Banin <banind@cs.washington.edu>
Co-authored-by: Chris Russell <8494645+chescock@users.noreply.github.com>
There are two related problems here:
1. Users should be able to change the fallback behavior of *all*
ECS-based errors in their application by setting the
`GLOBAL_ERROR_HANDLER`. See #18351 for earlier work in this vein.
2. The existing solution (#15500) for customizing this behavior is high
on boilerplate, not global and adds a great deal of complexity.
The consensus is that the default behavior when a parameter fails
validation should be set based on the kind of system parameter in
question: `Single` / `Populated` should silently skip the system, but
`Res` should panic. Setting this behavior at the system level is a
bandaid that makes getting to that ideal behavior more painful, and can
mask real failures (if a resource is missing but you've ignored a system
to make the Single stop panicking you're going to have a bad day).
I've removed the existing `ParamWarnPolicy`-based configuration, and
wired up the `GLOBAL_ERROR_HANDLER`/`default_error_handler` to the
various schedule executors to properly plumb through errors .
Additionally, I've done a small cleanup pass on the corresponding
example.
I've run the `fallible_params` example, with both the default and a
custom global error handler. The former panics (as expected), and the
latter spams the error console with warnings 🥲
1. Currently, failed system param validation will result in endless
console spam. Do you want me to implement a solution for warn_once-style
debouncing somehow?
2. Currently, the error reporting for failed system param validation is
very limited: all we get is that a system param failed validation and
the name of the system. Do you want me to implement improved error
reporting by bubbling up errors in this PR?
3. There is broad consensus that the default behavior for failed system
param validation should be set on a per-system param basis. Would you
like me to implement that in this PR?
My gut instinct is that we absolutely want to solve 2 and 3, but it will
be much easier to do that work (and review it) if we split the PRs
apart.
`ParamWarnPolicy` and the `WithParamWarnPolicy` have been removed
completely. Failures during system param validation are now handled via
the `GLOBAL_ERROR_HANDLER`: please see the `bevy_ecs::error` module docs
for more information.
---------
Co-authored-by: MiniaczQ <xnetroidpl@gmail.com>
# Objective
Fixes https://github.com/bevyengine/bevy/issues/18366 which seems to
have a similar underlying cause than the already closed (but not fixed)
https://github.com/bevyengine/bevy/issues/16185.
## Solution
For Windows with the AMD vulkan driver, there was already a hack to
force serial command encoding, which prevented these issues. The Linux
version of the AMD vulkan driver seems to have similar issues than its
Windows counterpart, so I extended the hack to also cover AMD on Linux.
I also removed the mention of `wgpu` since it was already outdated, and
doesn't seem to be relevant to the core issue (the AMD driver being
buggy).
## Testing
- Did you test these changes? If so, how?
- I ran the `3d_scene` example, which on `main` produced the flickering
shadows on Linux with the amdvlk driver, while it no longer does with
the workaround applied.
- Are there any parts that need more testing?
- Not sure.
- How can other people (reviewers) test your changes? Is there anything
specific they need to know?
- Requires a Linux system with an AMD card and the AMDVLK driver.
- If relevant, what platforms did you test these changes on, and are
there any important ones you can't test?
- My change should only affect Linux, where I did test it.
# Objective
WESL was broken on windows.
## Solution
- Upgrade to `wesl_rs` 1.2.
- Fix path handling on windows.
- Improve example for khronos demo this week.
# Objective
Now that #13432 has been merged, it's important we update our reflected
types to properly opt into this feature. If we do not, then this could
cause issues for users downstream who want to make use of
reflection-based cloning.
## Solution
This PR is broken into 4 commits:
1. Add `#[reflect(Clone)]` on all types marked `#[reflect(opaque)]` that
are also `Clone`. This is mandatory as these types would otherwise cause
the cloning operation to fail for any type that contains it at any
depth.
2. Update the reflection example to suggest adding `#[reflect(Clone)]`
on opaque types.
3. Add `#[reflect(clone)]` attributes on all fields marked
`#[reflect(ignore)]` that are also `Clone`. This prevents the ignored
field from causing the cloning operation to fail.
Note that some of the types that contain these fields are also `Clone`,
and thus can be marked `#[reflect(Clone)]`. This makes the
`#[reflect(clone)]` attribute redundant. However, I think it's safer to
keep it marked in the case that the `Clone` impl/derive is ever removed.
I'm open to removing them, though, if people disagree.
4. Finally, I added `#[reflect(Clone)]` on all types that are also
`Clone`. While not strictly necessary, it enables us to reduce the
generated output since we can just call `Clone::clone` directly instead
of calling `PartialReflect::reflect_clone` on each variant/field. It
also means we benefit from any optimizations or customizations made in
the `Clone` impl, including directly dereferencing `Copy` values and
increasing reference counters.
Along with that change I also took the liberty of adding any missing
registrations that I saw could be applied to the type as well, such as
`Default`, `PartialEq`, and `Hash`. There were hundreds of these to
edit, though, so it's possible I missed quite a few.
That last commit is **_massive_**. There were nearly 700 types to
update. So it's recommended to review the first three before moving onto
that last one.
Additionally, I can break the last commit off into its own PR or into
smaller PRs, but I figured this would be the easiest way of doing it
(and in a timely manner since I unfortunately don't have as much time as
I used to for code contributions).
## Testing
You can test locally with a `cargo check`:
```
cargo check --workspace --all-features
```
Extracted from my DLSS branch.
## Changelog
* Added `source_texture` and `destination_texture` to
`PostProcessWrite`, in addition to the existing texture views.
# Objective
In its existing form, the clamping that's done in `camera_system`
doesn't work well when the `physical_position` of the associated
viewport is nonzero. In such cases, it may produce invalid viewport
rectangles (i.e. not lying inside the render target), which may result
in crashes during the render pass.
The goal of this PR is to eliminate this possibility by making the
clamping behavior always result in a valid viewport rectangle when
possible.
## Solution
Incorporate the `physical_position` information into the clamping
behavior. In particular, always cut off enough so that it's contained in
the render target rather than clamping it to the same dimensions as the
target itself. In weirder situations, still try to produce a valid
viewport rectangle to avoid crashes.
## Testing
Tested these changes on my work branch where I encountered the crash.
# Objective
The `ViewportConversionError` error type does not implement `Error`,
making it incompatible with `BevyError`.
## Solution
Derive `Error` for `ViewportConversionError`.
I chose to use `thiserror` since it's already a dependency, but do let
me know if we should be preferring `derive_more`.
## Testing
You can test this by trying to compile the following:
```rust
let error: BevyError = ViewportConversionError::InvalidData.into();
```
# Objective
Prevents duplicate implementation between IntoSystemConfigs and
IntoSystemSetConfigs using a generic, adds a NodeType trait for more
config flexibility (opening the door to implement
https://github.com/bevyengine/bevy/issues/14195?).
## Solution
Followed writeup by @ItsDoot:
https://hackmd.io/@doot/rJeefFHc1x
Removes IntoSystemConfigs and IntoSystemSetConfigs, instead using
IntoNodeConfigs with generics.
## Testing
Pending
---
## Showcase
N/A
## Migration Guide
SystemSetConfigs -> NodeConfigs<InternedSystemSet>
SystemConfigs -> NodeConfigs<ScheduleSystem>
IntoSystemSetConfigs -> IntoNodeConfigs<InternedSystemSet, M>
IntoSystemConfigs -> IntoNodeConfigs<ScheduleSystem, M>
---------
Co-authored-by: Christian Hughes <9044780+ItsDoot@users.noreply.github.com>
Co-authored-by: Alice Cecile <alice.i.cecile@gmail.com>
# Objective
- #16883
- Improve the default behaviour of the exclusive fullscreen API.
## Solution
This PR changes the exclusive fullscreen window mode to require the type
`WindowMode::Fullscreen(MonitorSelection, VideoModeSelection)` and
removes `WindowMode::SizedFullscreen`. This API somewhat intentionally
more closely resembles Winit 0.31's upcoming fullscreen and video mode
API.
The new VideoModeSelection enum is specified as follows:
```rust
pub enum VideoModeSelection {
/// Uses the video mode that the monitor is already in.
Current,
/// Uses a given [`crate::monitor::VideoMode`]. A list of video modes supported by the monitor
/// is supplied by [`crate::monitor::Monitor::video_modes`].
Specific(VideoMode),
}
```
### Changing default behaviour
This might be contentious because it removes the previous behaviour of
`WindowMode::Fullscreen` which selected the highest resolution possible.
While the previous behaviour would be quite easy to re-implement as
additional options, or as an impl method on Monitor, I would argue that
this isn't an implementation that should be encouraged.
From the perspective of a Windows user, I prefer what the majority of
modern games do when entering fullscreen which is to preserve the OS's
current resolution settings, which allows exclusive fullscreen to be
entered faster, and to only have it change if I manually select it in
either the options of the game or the OS. The highest resolution
available is not necessarily what the user prefers.
I am open to changing this if I have just missed a good use case for it.
Likewise, the only functionality that `WindowMode::SizedFullscreen`
provided was that it selected the resolution closest to the current size
of the window so it was removed since this behaviour can be replicated
via the new `VideoModeSelection::Specific` if necessary.
## Out of scope
WindowResolution and scale factor act strangely in exclusive fullscreen,
this PR doesn't address it or regress it.
## Testing
- Tested on Windows 11 and macOS 12.7
- Linux untested
## Migration Guide
`WindowMode::SizedFullscreen(MonitorSelection)` and
`WindowMode::Fullscreen(MonitorSelection)` has become
`WindowMode::Fullscreen(MonitorSelection, VideoModeSelection)`.
Previously, the VideoMode was selected based on the closest resolution
to the current window size for SizedFullscreen and the largest
resolution for Fullscreen. It is possible to replicate that behaviour by
searching `Monitor::video_modes` and selecting it with
`VideoModeSelection::Specific(VideoMode)` but it is recommended to use
`VideoModeSelection::Current` as the default video mode when entering
fullscreen.
PR #17965 mistakenly made the `AsBindGroup` macro no longer emit a bind
group layout entry and a resource descriptor for buffers. This commit
adds that functionality back, fixing the `shader_material_bindless`
example.
Closes#18124.
# Objective
- In `Camera::viewport_to_world_2d`, `Camera::viewport_to_world`,
`Camera::world_to_viewport` and `Camera::world_to_viewport_with_depth`,
the results were incorrect when the `Camera::viewport` field was
configured with a viewport position that was non-zero. This PR attempts
to correct that.
- Fixes#16200
## Solution
- This PR now takes the viewport position into account in the functions
mentioned above.
- Extended `2d_viewport_to_world` example to test the functions with a
dynamic viewport position and size, camera positions and zoom levels. It
is probably worth discussing whether to change the example, add a new
one or just completely skip touching the examples.
## Testing
Used the modified example to test the functions with dynamic camera
transform as well as dynamic viewport size and position.
# Objective
Because `prepare_assets::<T>` had a mutable reference to the
`RenderAssetBytesPerFrame` resource, no render asset preparation could
happen in parallel. This PR fixes this by using an `AtomicUsize` to
count bytes written (if there's a limit in place), so that the system
doesn't need mutable access.
- Related: https://github.com/bevyengine/bevy/pull/12622
**Before**
<img width="1049" alt="Screenshot 2025-02-17 at 11 40 53 AM"
src="https://github.com/user-attachments/assets/040e6184-1192-4368-9597-5ceda4b8251b"
/>
**After**
<img width="836" alt="Screenshot 2025-02-17 at 1 38 37 PM"
src="https://github.com/user-attachments/assets/95488796-3323-425c-b0a6-4cf17753512e"
/>
## Testing
- Tested on a local project (with and without limiting enabled)
- Someone with more knowledge of wgpu/underlying driver guts should
confirm that this doesn't actually bite us by introducing contention
(i.e. if buffer writing really *should be* serial).
# Objective
- Allow bevy and wgpu developers to test newer versions of wgpu without
having to update naga_oil.
## Solution
- Currently bevy feeds wgsl through naga_oil to get a naga::Module that
it passes to wgpu.
- Added a way to pass wgsl through naga_oil, and then serialize the
naga::Module back into a wgsl string to feed to wgpu, allowing wgpu to
parse it using it's internal version of naga (and not the version of
naga bevy_render/naga_oil is using).
## Testing
1. Run 3d_scene (it works)
2. Run 3d_scene with `--features bevy_render/decoupled_naga` (it still
works)
3. Add the following patch to bevy/Cargo.toml, run cargo update, and
compile again (it will fail)
```toml
[patch.crates-io]
wgpu = { git = "https://github.com/gfx-rs/wgpu", rev = "2764e7a39920e23928d300e8856a672f1952da63" }
wgpu-core = { git = "https://github.com/gfx-rs/wgpu", rev = "2764e7a39920e23928d300e8856a672f1952da63" }
wgpu-hal = { git = "https://github.com/gfx-rs/wgpu", rev = "2764e7a39920e23928d300e8856a672f1952da63" }
wgpu-types = { git = "https://github.com/gfx-rs/wgpu", rev = "2764e7a39920e23928d300e8856a672f1952da63" }
```
4. Fix errors and compile again (it will work, and you didn't have to
touch naga_oil)
# Objective
WESL's pre-MVP `0.1.0` has been
[released](https://docs.rs/wesl/latest/wesl/)!
Add support for WESL shader source so that we can begin playing and
testing WESL, as well as aiding in their development.
## Solution
Adds a `ShaderSource::WESL` that can be used to load `.wesl` shaders.
Right now, we don't support mixing `naga-oil`. Additionally, WESL
shaders currently need to pass through the naga frontend, which the WESL
team is aware isn't great for performance (they're working on compiling
into naga modules). Also, since our shaders are managed using the asset
system, we don't currently support using file based imports like `super`
or package scoped imports. Further work will be needed to asses how we
want to support this.
---
## Showcase
See the `shader_material_wesl` example. Be sure to press space to
activate party mode (trigger conditional compilation)!
https://github.com/user-attachments/assets/ec6ad19f-b6e4-4e9d-a00f-6f09336b08a4
# Objective
- Contributes to #15460
- Supersedes #8520
- Fixes#4906
## Solution
- Added a new `web` feature to `bevy`, and several of its crates.
- Enabled new `web` feature automatically within crates without `no_std`
support.
## Testing
- `cargo build --no-default-features --target wasm32v1-none`
---
## Migration Guide
When using Bevy crates which _don't_ automatically enable the `web`
feature, please enable it when building for the browser.
## Notes
- I added [`cfg_if`](https://crates.io/crates/cfg-if) to help manage
some of the feature gate gore that this extra feature introduces. It's
still pretty ugly, but I think much easier to read.
- Certain `wasm` targets (e.g.,
[wasm32-wasip1](https://doc.rust-lang.org/nightly/rustc/platform-support/wasm32-wasip1.html#wasm32-wasip1))
provide an incomplete implementation for `std`. I have not tested these
platforms, but I suspect Bevy's liberal use of usually unsupported
features (e.g., threading) will cause these targets to fail. As such,
consider `wasm32-unknown-unknown` as the only `wasm` platform with
support from Bevy for `std`. All others likely will need to be treated
as `no_std` platforms.
# Objective
- Fixes#15460 (will open new issues for further `no_std` efforts)
- Supersedes #17715
## Solution
- Threaded in new features as required
- Made certain crates optional but default enabled
- Removed `compile-check-no-std` from internal `ci` tool since GitHub CI
can now simply check `bevy` itself now
- Added CI task to check `bevy` on `thumbv6m-none-eabi` to ensure
`portable-atomic` support is still valid [^1]
[^1]: This may be controversial, since it could be interpreted as
implying Bevy will maintain support for `thumbv6m-none-eabi` going
forward. In reality, just like `x86_64-unknown-none`, this is a
[canary](https://en.wiktionary.org/wiki/canary_in_a_coal_mine) target to
make it clear when `portable-atomic` no longer works as intended (fixing
atomic support on atomically challenged platforms). If a PR comes
through and makes supporting this class of platforms impossible, then
this CI task can be removed. I however wager this won't be a problem.
## Testing
- CI
---
## Release Notes
Bevy now has support for `no_std` directly from the `bevy` crate.
Users can disable default features and enable a new `default_no_std`
feature instead, allowing `bevy` to be used in `no_std` applications and
libraries.
```toml
# Bevy for `no_std` platforms
bevy = { version = "0.16", default-features = false, features = ["default_no_std"] }
```
`default_no_std` enables certain required features, such as `libm` and
`critical-section`, and as many optional crates as possible (currently
just `bevy_state`). For atomically-challenged platforms such as the
Raspberry Pi Pico, `portable-atomic` will be used automatically.
For library authors, we recommend depending on `bevy` with
`default-features = false` to allow `std` and `no_std` users to both
depend on your crate. Here are some recommended features a library crate
may want to expose:
```toml
[features]
# Most users will be on a platform which has `std` and can use the more-powerful `async_executor`.
default = ["std", "async_executor"]
# Features for typical platforms.
std = ["bevy/std"]
async_executor = ["bevy/async_executor"]
# Features for `no_std` platforms.
libm = ["bevy/libm"]
critical-section = ["bevy/critical-section"]
[dependencies]
# We disable default features to ensure we don't accidentally enable `std` on `no_std` targets, for example.
bevy = { version = "0.16", default-features = false }
```
While this is verbose, it gives the maximum control to end-users to
decide how they wish to use Bevy on their platform.
We encourage library authors to experiment with `no_std` support. For
libraries relying exclusively on `bevy` and no other dependencies, it
may be as simple as adding `#![no_std]` to your `lib.rs` and exposing
features as above! Bevy can also provide many `std` types, such as
`HashMap`, `Mutex`, and `Instant` on all platforms. See
`bevy::platform_support` for details on what's available out of the box!
## Migration Guide
- If you were previously relying on `bevy` with default features
disabled, you may need to enable the `std` and `async_executor`
features.
- `bevy_reflect` has had its `bevy` feature removed. If you were relying
on this feature, simply enable `smallvec` and `smol_str` instead.
---------
Co-authored-by: Alice Cecile <alice.i.cecile@gmail.com>
## Objective
`insert_or_spawn_batch` is due to be deprecated eventually (#15704), and
removing uses internally will make that easier.
## Solution
Replaced internal uses of `insert_or_spawn_batch` with
`try_insert_batch` (non-panicking variant because
`insert_or_spawn_batch` didn't panic).
All of the internal uses are in rendering code. Since retained rendering
was meant to get rid non-opaque entity IDs, I assume the code was just
using `insert_or_spawn_batch` because `insert_batch` didn't exist and
not because it actually wanted to spawn something. However, I am *not*
confident in my ability to judge rendering code.
# Objective
Component `require()` IDE integration is fully broken, as of #16575.
## Solution
This reverts us back to the previous "put the docs on Component trait"
impl. This _does_ reduce the accessibility of the required components in
rust docs, but the complete erasure of "required component IDE
experience" is not worth the price of slightly increased prominence of
requires in docs.
Additionally, Rust Analyzer has recently started including derive
attributes in suggestions, so we aren't losing that benefit of the
proc_macro attribute impl.
# Objective
Fix unsound query transmutes on queries obtained from
`Query::as_readonly()`.
The following compiles, and the call to `transmute_lens()` should panic,
but does not:
```rust
fn bad_system(query: Query<&mut A>) {
let mut readonly = query.as_readonly();
let mut lens: QueryLens<&mut A> = readonly.transmute_lens();
let other_readonly: Query<&A> = query.as_readonly();
// `lens` and `other_readonly` alias, and are both alive here!
}
```
To make `Query::as_readonly()` zero-cost, we pointer-cast
`&QueryState<D, F>` to `&QueryState<D::ReadOnly, F>`. This means that
the `component_access` for a read-only query's state may include
accesses for the original mutable version, but the `Query` does not have
exclusive access to those components! `transmute` and `join` use that
access to ensure that a join is valid, and will incorrectly allow a
transmute that includes mutable access.
As a bonus, allow `Query::join`s that output `FilteredEntityRef` or
`FilteredEntityMut` to receive access from the `other` query. Currently
they only receive access from `self`.
## Solution
When transmuting or joining from a read-only query, remove any writes
before performing checking that the transmute is valid. For joins, be
sure to handle the case where one input query was the result of
`as_readonly()` but the other has valid mutable access.
This requires identifying read-only queries, so add a
`QueryData::IS_READ_ONLY` associated constant. Note that we only call
`QueryState::as_transmuted_state()` with `NewD: ReadOnlyQueryData`, so
checking for read-only queries is sufficient to check for
`as_transmuted_state()`.
Removing writes requires allocating a new `FilteredAccess`, so only do
so if the query is read-only and the state has writes. Otherwise, the
existing access is correct and we can continue using a reference to it.
Use the new read-only state to call `NewD::set_access`, so that
transmuting to a `FilteredAccessMut` results in a read-only
`FilteredAccessMut`. Otherwise, it would take the original write access,
and then the transmute would panic because it had too much access.
Note that `join` was previously passing `self.component_access` to
`NewD::set_access`. Switching it to `joined_component_access` also
allows a join that outputs `FilteredEntity(Ref|Mut)` to receive access
from `other`. The fact that it didn't do that before seems like an
oversight, so I didn't try to prevent that change.
## Testing
Added unit tests with the unsound transmute and join.
# Objective
As discussed in #14275, Bevy is currently too prone to panic, and makes
the easy / beginner-friendly way to do a large number of operations just
to panic on failure.
This is seriously frustrating in library code, but also slows down
development, as many of the `Query::single` panics can actually safely
be an early return (these panics are often due to a small ordering issue
or a change in game state.
More critically, in most "finished" products, panics are unacceptable:
any unexpected failures should be handled elsewhere. That's where the
new
With the advent of good system error handling, we can now remove this.
Note: I was instrumental in a) introducing this idea in the first place
and b) pushing to make the panicking variant the default. The
introduction of both `let else` statements in Rust and the fancy system
error handling work in 0.16 have changed my mind on the right balance
here.
## Solution
1. Make `Query::single` and `Query::single_mut` (and other random
related methods) return a `Result`.
2. Handle all of Bevy's internal usage of these APIs.
3. Deprecate `Query::get_single` and friends, since we've moved their
functionality to the nice names.
4. Add detailed advice on how to best handle these errors.
Generally I like the diff here, although `get_single().unwrap()` in
tests is a bit of a downgrade.
## Testing
I've done a global search for `.single` to track down any missed
deprecated usages.
As to whether or not all the migrations were successful, that's what CI
is for :)
## Future work
~~Rename `Query::get_single` and friends to `Query::single`!~~
~~I've opted not to do this in this PR, and smear it across two releases
in order to ease the migration. Successive deprecations are much easier
to manage than the semantics and types shifting under your feet.~~
Cart has convinced me to change my mind on this; see
https://github.com/bevyengine/bevy/pull/18082#discussion_r1974536085.
## Migration guide
`Query::single`, `Query::single_mut` and their `QueryState` equivalents
now return a `Result`. Generally, you'll want to:
1. Use Bevy 0.16's system error handling to return a `Result` using the
`?` operator.
2. Use a `let else Ok(data)` block to early return if it's an expected
failure.
3. Use `unwrap()` or `Ok` destructuring inside of tests.
The old `Query::get_single` (etc) methods which did this have been
deprecated.
# Objective
There are currently three ways to access the parent stored on a ChildOf
relationship:
1. `child_of.parent` (field accessor)
2. `child_of.get()` (get function)
3. `**child_of` (Deref impl)
I will assert that we should only have one (the field accessor), and
that the existence of the other implementations causes confusion and
legibility issues. The deref approach is heinous, and `child_of.get()`
is significantly less clear than `child_of.parent`.
## Solution
Remove `impl Deref for ChildOf` and `ChildOf::get`.
The one "downside" I'm seeing is that:
```rust
entity.get::<ChildOf>().map(ChildOf::get)
```
Becomes this:
```rust
entity.get::<ChildOf>().map(|c| c.parent)
```
I strongly believe that this is worth the increased clarity and
consistency. I'm also not really a huge fan of the "pass function
pointer to map" syntax. I think most people don't think this way about
maps. They think in terms of a function that takes the item in the
Option and returns the result of some action on it.
## Migration Guide
```rust
// Before
**child_of
// After
child_of.parent
// Before
child_of.get()
// After
child_of.parent
// Before
entity.get::<ChildOf>().map(ChildOf::get)
// After
entity.get::<ChildOf>().map(|c| c.parent)
```
# Objective
fixes#17896
## Solution
Change ChildOf ( Entity ) to ChildOf { parent: Entity }
by doing this we also allow users to use named structs for relationship
derives, When you have more than 1 field in a struct with named fields
the macro will look for a field with the attribute #[relationship] and
all of the other fields should implement the Default trait. Unnamed
fields are still supported.
When u have a unnamed struct with more than one field the macro will
fail.
Do we want to support something like this ?
```rust
#[derive(Component)]
#[relationship_target(relationship = ChildOf)]
pub struct Children (#[relationship] Entity, u8);
```
I could add this, it but doesn't seem nice.
## Testing
crates/bevy_ecs - cargo test
## Showcase
```rust
use bevy_ecs::component::Component;
use bevy_ecs::entity::Entity;
#[derive(Component)]
#[relationship(relationship_target = Children)]
pub struct ChildOf {
#[relationship]
pub parent: Entity,
internal: u8,
};
#[derive(Component)]
#[relationship_target(relationship = ChildOf)]
pub struct Children {
children: Vec<Entity>
};
```
---------
Co-authored-by: Tim Overbeek <oorbecktim@Tims-MacBook-Pro.local>
Co-authored-by: Tim Overbeek <oorbecktim@c-001-001-042.client.nl.eduvpn.org>
Co-authored-by: Tim Overbeek <oorbecktim@c-001-001-059.client.nl.eduvpn.org>
Co-authored-by: Tim Overbeek <oorbecktim@c-001-001-054.client.nl.eduvpn.org>
Co-authored-by: Tim Overbeek <oorbecktim@c-001-001-027.client.nl.eduvpn.org>
Even though opaque deferred entities aren't placed into the `Opaque3d`
bin, we still want to cache them as though they were, so that we don't
have to re-queue them every frame. This commit implements that logic,
reducing the time of `queue_material_meshes` to near-zero on Caldera.
Currently, the structure-level `#[uniform]` attribute of `AsBindGroup`
creates a binding array of individual buffers, each of which contains
data for a single material. A more efficient approach would be to
provide a single buffer with an array containing all of the data for all
materials in the bind group. Because `StandardMaterial` uses
`#[uniform]`, this can be notably inefficient with large numbers of
materials.
This patch introduces a new attribute on `AsBindGroup`, `#[data]`, which
works identically to `#[uniform]` except that it concatenates all the
data into a single buffer that the material bind group allocator itself
manages. It also converts `StandardMaterial` to use this new
functionality. This effectively provides the "material data in arrays"
feature.
# Objective
- Fixes#17960
## Solution
- Followed the [edition upgrade
guide](https://doc.rust-lang.org/edition-guide/editions/transitioning-an-existing-project-to-a-new-edition.html)
## Testing
- CI
---
## Summary of Changes
### Documentation Indentation
When using lists in documentation, proper indentation is now linted for.
This means subsequent lines within the same list item must start at the
same indentation level as the item.
```rust
/* Valid */
/// - Item 1
/// Run-on sentence.
/// - Item 2
struct Foo;
/* Invalid */
/// - Item 1
/// Run-on sentence.
/// - Item 2
struct Foo;
```
### Implicit `!` to `()` Conversion
`!` (the never return type, returned by `panic!`, etc.) no longer
implicitly converts to `()`. This is particularly painful for systems
with `todo!` or `panic!` statements, as they will no longer be functions
returning `()` (or `Result<()>`), making them invalid systems for
functions like `add_systems`. The ideal fix would be to accept functions
returning `!` (or rather, _not_ returning), but this is blocked on the
[stabilisation of the `!` type
itself](https://doc.rust-lang.org/std/primitive.never.html), which is
not done.
The "simple" fix would be to add an explicit `-> ()` to system
signatures (e.g., `|| { todo!() }` becomes `|| -> () { todo!() }`).
However, this is _also_ banned, as there is an existing lint which (IMO,
incorrectly) marks this as an unnecessary annotation.
So, the "fix" (read: workaround) is to put these kinds of `|| -> ! { ...
}` closuers into variables and give the variable an explicit type (e.g.,
`fn()`).
```rust
// Valid
let system: fn() = || todo!("Not implemented yet!");
app.add_systems(..., system);
// Invalid
app.add_systems(..., || todo!("Not implemented yet!"));
```
### Temporary Variable Lifetimes
The order in which temporary variables are dropped has changed. The
simple fix here is _usually_ to just assign temporaries to a named
variable before use.
### `gen` is a keyword
We can no longer use the name `gen` as it is reserved for a future
generator syntax. This involved replacing uses of the name `gen` with
`r#gen` (the raw-identifier syntax).
### Formatting has changed
Use statements have had the order of imports changed, causing a
substantial +/-3,000 diff when applied. For now, I have opted-out of
this change by amending `rustfmt.toml`
```toml
style_edition = "2021"
```
This preserves the original formatting for now, reducing the size of
this PR. It would be a simple followup to update this to 2024 and run
`cargo fmt`.
### New `use<>` Opt-Out Syntax
Lifetimes are now implicitly included in RPIT types. There was a handful
of instances where it needed to be added to satisfy the borrow checker,
but there may be more cases where it _should_ be added to avoid
breakages in user code.
### `MyUnitStruct { .. }` is an invalid pattern
Previously, you could match against unit structs (and unit enum
variants) with a `{ .. }` destructuring. This is no longer valid.
### Pretty much every use of `ref` and `mut` are gone
Pattern binding has changed to the point where these terms are largely
unused now. They still serve a purpose, but it is far more niche now.
### `iter::repeat(...).take(...)` is bad
New lint recommends using the more explicit `iter::repeat_n(..., ...)`
instead.
## Migration Guide
The lifetimes of functions using return-position impl-trait (RPIT) are
likely _more_ conservative than they had been previously. If you
encounter lifetime issues with such a function, please create an issue
to investigate the addition of `+ use<...>`.
## Notes
- Check the individual commits for a clearer breakdown for what
_actually_ changed.
---------
Co-authored-by: François Mockers <francois.mockers@vleue.com>
Two-phase occlusion culling can be helpful for shadow maps just as it
can for a prepass, in order to reduce vertex and alpha mask fragment
shading overhead. This patch implements occlusion culling for shadow
maps from directional lights, when the `OcclusionCulling` component is
present on the entities containing the lights. Shadow maps from point
lights are deferred to a follow-up patch. Much of this patch involves
expanding the hierarchical Z-buffer to cover shadow maps in addition to
standard view depth buffers.
The `scene_viewer` example has been updated to add `OcclusionCulling` to
the directional light that it creates.
This improved the performance of the rend3 sci-fi test scene when
enabling shadows.
Currently, Bevy's implementation of bindless resources is rather
unusual: every binding in an object that implements `AsBindGroup` (most
commonly, a material) becomes its own separate binding array in the
shader. This is inefficient for two reasons:
1. If multiple materials reference the same texture or other resource,
the reference to that resource will be duplicated many times. This
increases `wgpu` validation overhead.
2. It creates many unused binding array slots. This increases `wgpu` and
driver overhead and makes it easier to hit limits on APIs that `wgpu`
currently imposes tight resource limits on, like Metal.
This PR fixes these issues by switching Bevy to use the standard
approach in GPU-driven renderers, in which resources are de-duplicated
and passed as global arrays, one for each type of resource.
Along the way, this patch introduces per-platform resource limits and
bumps them from 16 resources per binding array to 64 resources per bind
group on Metal and 2048 resources per bind group on other platforms.
(Note that the number of resources per *binding array* isn't the same as
the number of resources per *bind group*; as it currently stands, if all
the PBR features are turned on, Bevy could pack as many as 496 resources
into a single slab.) The limits have been increased because `wgpu` now
has universal support for partially-bound binding arrays, which mean
that we no longer need to fill the binding arrays with fallback
resources on Direct3D 12. The `#[bindless(LIMIT)]` declaration when
deriving `AsBindGroup` can now simply be written `#[bindless]` in order
to have Bevy choose a default limit size for the current platform.
Custom limits are still available with the new
`#[bindless(limit(LIMIT))]` syntax: e.g. `#[bindless(limit(8))]`.
The material bind group allocator has been completely rewritten. Now
there are two allocators: one for bindless materials and one for
non-bindless materials. The new non-bindless material allocator simply
maintains a 1:1 mapping from material to bind group. The new bindless
material allocator maintains a list of slabs and allocates materials
into slabs on a first-fit basis. This unfortunately makes its
performance O(number of resources per object * number of slabs), but the
number of slabs is likely to be low, and it's planned to become even
lower in the future with `wgpu` improvements. Resources are
de-duplicated with in a slab and reference counted. So, for instance, if
multiple materials refer to the same texture, that texture will exist
only once in the appropriate binding array.
To support these new features, this patch adds the concept of a
*bindless descriptor* to the `AsBindGroup` trait. The bindless
descriptor allows the material bind group allocator to probe the layout
of the material, now that an array of `BindGroupLayoutEntry` records is
insufficient to describe the group. The `#[derive(AsBindGroup)]` has
been heavily modified to support the new features. The most important
user-facing change to that macro is that the struct-level `uniform`
attribute, `#[uniform(BINDING_NUMBER, StandardMaterial)]`, now reads
`#[uniform(BINDLESS_INDEX, MATERIAL_UNIFORM_TYPE,
binding_array(BINDING_NUMBER)]`, allowing the material to specify the
binding number for the binding array that holds the uniform data.
To make this patch simpler, I removed support for bindless
`ExtendedMaterial`s, as well as field-level bindless uniform and storage
buffers. I intend to add back support for these as a follow-up. Because
they aren't in any released Bevy version yet, I figured this was OK.
Finally, this patch updates `StandardMaterial` for the new bindless
changes. Generally, code throughout the PBR shaders that looked like
`base_color_texture[slot]` now looks like
`bindless_2d_textures[material_indices[slot].base_color_texture]`.
This patch fixes a system hang that I experienced on the [Caldera test]
when running with `caldera --random-materials --texture-count 100`. The
time per frame is around 19.75 ms, down from 154.2 ms in Bevy 0.14: a
7.8× speedup.
[Caldera test]: https://github.com/DGriffin91/bevy_caldera_scene
This commit restructures the multidrawable batch set builder for better
performance in various ways:
* The bin traversal is optimized to make the best use of the CPU cache.
* The inner loop that iterates over the bins, which is the hottest part
of `batch_and_prepare_binned_render_phase`, has been shrunk as small as
possible.
* Where possible, multiple elements are added to or reserved from GPU
buffers as a batch instead of one at a time.
* Methods that LLVM wasn't inlining have been marked `#[inline]` where
doing so would unlock optimizations.
This code has also been refactored to avoid duplication between the
logic for indexed and non-indexed meshes via the introduction of a
`MultidrawableBatchSetPreparer` object.
Together, this improved the `batch_and_prepare_binned_render_phase` time
on Caldera by approximately 2×.
Eventually, we should optimize the batchable-but-not-multidrawable and
unbatchable logic as well, but these meshes are much rarer, so in the
interests of keeping this patch relatively small I opted to leave those
to a follow-up.
# Objective
Make checked vs unchecked shaders configurable
Fixes#17786
## Solution
Added `ValidateShaders` enum to `Shader` and added
`create_and_validate_shader_module` to `RenderDevice`
## Testing
I tested the shader examples locally and they all worked. I'd like to
write a few tests to verify but am unsure how to start.
---------
Co-authored-by: Alice Cecile <alice.i.cecile@gmail.com>
The `check_visibility` system currently follows this algorithm:
1. Store all meshes that were visible last frame in the
`PreviousVisibleMeshes` set.
2. Determine which meshes are visible. For each such visible mesh,
remove it from `PreviousVisibleMeshes`.
3. Mark all meshes that remain in `PreviousVisibleMeshes` as invisible.
This algorithm would be correct if the `check_visibility` were the only
system that marked meshes visible. However, it's not: the shadow-related
systems `check_dir_light_mesh_visibility` and
`check_point_light_mesh_visibility` can as well. This results in the
following sequence of events for meshes that are in a shadow map but
*not* visible from a camera:
A. `check_visibility` runs, finds that no camera contains these meshes,
and marks them hidden, which sets the changed flag.
B. `check_dir_light_mesh_visibility` and/or
`check_point_light_mesh_visibility` run, discover that these meshes
are visible in the shadow map, and marks them as visible, again
setting the `ViewVisibility` changed flag.
C. During the extraction phase, the mesh extraction system sees that
`ViewVisibility` is changed and re-extracts the mesh.
This is inefficient and results in needless work during rendering.
This patch fixes the issue in two ways:
* The `check_dir_light_mesh_visibility` and
`check_point_light_mesh_visibility` systems now remove meshes that they
discover from `PreviousVisibleMeshes`.
* Step (3) above has been moved from `check_visibility` to a separate
system, `mark_newly_hidden_entities_invisible`. This system runs after
all visibility-determining systems, ensuring that
`PreviousVisibleMeshes` contains only those meshes that truly became
invisible on this frame.
This fix dramatically improves the performance of [the Caldera
benchmark], when combined with several other patches I've submitted.
[the Caldera benchmark]:
https://github.com/DGriffin91/bevy_caldera_scene
PR #17688 broke motion vector computation, and therefore motion blur,
because it enabled retention of `MeshInputUniform`s, and
`MeshInputUniform`s contain the indices of the previous frame's
transform and the previous frame's skinned mesh joint matrices. On frame
N, if a `MeshInputUniform` is retained on GPU from the previous frame,
the `previous_input_index` and `previous_skin_index` would refer to the
indices for frame N - 2, not the index for frame N - 1.
This patch fixes the problems. It solves these issues in two different
ways, one for transforms and one for skins:
1. To fix transforms, this patch supplies the *frame index* to the
shader as part of the view uniforms, and specifies which frame index
each mesh's previous transform refers to. So, in the situation described
above, the frame index would be N, the previous frame index would be N -
1, and the `previous_input_frame_number` would be N - 2. The shader can
now detect this situation and infer that the mesh has been retained, and
can therefore conclude that the mesh's transform hasn't changed.
2. To fix skins, this patch replaces the explicit `previous_skin_index`
with an invariant that the index of the joints for the current frame and
the index of the joints for the previous frame are the same. This means
that the `MeshInputUniform` never has to be updated even if the skin is
animated. The downside is that we have to copy joint matrices from the
previous frame's buffer to the current frame's buffer in
`extract_skins`.
The rationale behind (2) is that we currently have no mechanism to
detect when joints that affect a skin have been updated, short of
comparing all the transforms and setting a flag for
`extract_meshes_for_gpu_building` to consume, which would regress
performance as we want `extract_skins` and
`extract_meshes_for_gpu_building` to be able to run in parallel.
To test this change, use `cargo run --example motion_blur`.
Currently, Bevy rebuilds the buffer containing all the transforms for
joints every frame, during the extraction phase. This is inefficient in
cases in which many skins are present in the scene and their joints
don't move, such as the Caldera test scene.
To address this problem, this commit switches skin extraction to use a
set of retained GPU buffers with allocations managed by the offset
allocator. I use fine-grained change detection in order to determine
which skins need updating. Note that the granularity is on the level of
an entire skin, not individual joints. Using the change detection at
that level would yield poor performance in common cases in which an
entire skin is animated at once. Also, this patch yields additional
performance from the fact that changing joint transforms no longer
requires the skinned mesh to be re-extracted.
Note that this optimization can be a double-edged sword. In
`many_foxes`, fine-grained change detection regressed the performance of
`extract_skins` by 3.4x. This is because every joint is updated every
frame in that example, so change detection is pointless and is pure
overhead. Because the `many_foxes` workload is actually representative
of animated scenes, this patch includes a heuristic that disables
fine-grained change detection if the number of transformed entities in
the frame exceeds a certain fraction of the total number of joints.
Currently, this threshold is set to 25%. Note that this is a crude
heuristic, because it doesn't distinguish between the number of
transformed *joints* and the number of transformed *entities*; however,
it should be good enough to yield the optimum code path most of the
time.
Finally, this patch fixes a bug whereby skinned meshes are actually
being incorrectly retained if the buffer offsets of the joints of those
skinned meshes changes from frame to frame. To fix this without
retaining skins, we would have to re-extract every skinned mesh every
frame. Doing this was a significant regression on Caldera. With this PR,
by contrast, mesh joints stay at the same buffer offset, so we don't
have to update the `MeshInputUniform` containing the buffer offset every
frame. This also makes PR #17717 easier to implement, because that PR
uses the buffer offset from the previous frame, and the logic for
calculating that is simplified if the previous frame's buffer offset is
guaranteed to be identical to that of the current frame.
On Caldera, this patch reduces the time spent in `extract_skins` from
1.79 ms to near zero. On `many_foxes`, this patch regresses the
performance of `extract_skins` by approximately 10%-25%, depending on
the number of foxes. This has only a small impact on frame rate.
The GPU can fill out many of the fields in `IndirectParametersMetadata`
using information it already has:
* `early_instance_count` and `late_instance_count` are always
initialized to zero.
* `mesh_index` is already present in the work item buffer as the
`input_index` of the first work item in each batch.
This patch moves these fields to a separate buffer, the *GPU indirect
parameters metadata* buffer. That way, it avoids having to write them on
CPU during `batch_and_prepare_binned_render_phase`. This effectively
reduces the number of bits that that function must write per mesh from
160 to 64 (in addition to the 64 bits per mesh *instance*).
Additionally, this PR refactors `UntypedPhaseIndirectParametersBuffers`
to add another layer, `MeshClassIndirectParametersBuffers`, which allows
abstracting over the buffers corresponding indexed and non-indexed
meshes. This patch doesn't make much use of this abstraction, but
forthcoming patches will, and it's overall a cleaner approach.
This didn't seem to have much of an effect by itself on
`batch_and_prepare_binned_render_phase` time, but subsequent PRs
dependent on this PR yield roughly a 2× speedup.
## Objective
There's no general error for when an entity doesn't exist, and some
methods are going to need one when they get Resultified. The closest
thing is `EntityFetchError`, but that error has a slightly more specific
purpose.
## Solution
- Added `EntityDoesNotExistError`.
- Contains `Entity` and `EntityDoesNotExistDetails`.
- Changed `EntityFetchError` and `QueryEntityError`:
- Changed `NoSuchEntity` variant to wrap `EntityDoesNotExistError` and
renamed the variant to `EntityDoesNotExist`.
- Renamed `EntityFetchError` to `EntityMutableFetchError` to make its
purpose clearer.
- Renamed `TryDespawnError` to `EntityDespawnError` to make it more
general.
- Changed `World::inspect_entity` to return `Result<[ok],
EntityDoesNotExistError>` instead of panicking.
- Changed `World::get_entity` and `WorldEntityFetch::fetch_ref` to
return `Result<[ok], EntityDoesNotExistError>` instead of `Result<[ok],
Entity>`.
- Changed `UnsafeWorldCell::get_entity` to return
`Result<UnsafeEntityCell, EntityDoesNotExistError>` instead of
`Option<UnsafeEntityCell>`.
## Migration Guide
- `World::inspect_entity` now returns `Result<impl Iterator<Item =
&ComponentInfo>, EntityDoesNotExistError>` instead of `impl
Iterator<Item = &ComponentInfo>`.
- `World::get_entity` now returns `EntityDoesNotExistError` as an error
instead of `Entity`. You can still access the entity's ID through the
error's `entity` field.
- `UnsafeWorldCell::get_entity` now returns `Result<UnsafeEntityCell,
EntityDoesNotExistError>` instead of `Option<UnsafeEntityCell>`.
Appending to these vectors is performance-critical in
`batch_and_prepare_binned_render_phase`, so `RawBufferVec`, which
doesn't have the overhead of `encase`, is more appropriate.
The `collect_buffers_for_phase` system tries to reuse these buffers, but
its efforts are stymied by the fact that
`clear_batched_gpu_instance_buffers` clears the containing hash table
and therefore frees the buffers. This patch makes
`clear_batched_gpu_instance_buffers` stop doing that so that the
allocations can be reused.
