fix a few dead links
* Links in `Input` missed a refactor
* `Reflect::downcast` can't use the intra doc link format, as it's not a link to a trait function, but to a function implemented on `dyn Reflect`
noticed in https://github.com/bevyengine/bevy/pull/1781#discussion_r619777879
Fixes#1892
The following code is a cut down version of the issue, and crashes the same way:
```rust
enum AssetLifecycleEvent <T> {
Create(T),
Free
}
fn main() {
let (sender, _receiver) = crossbeam_channel::unbounded();
sender.send(AssetLifecycleEvent::<[u32; 32000]>::Free).unwrap();
}
```
- We're creating a channel that need to be able to hold `AssetLifecycleEvent::Create(T)` which has the size of our type `T`
- The two variants of the enums have a very different size
By keeping `T` boxed while sending through the channel, it doesn't crash
The documentation for `ShouldRun` doesn't completely explain what each of the variants you can return does. For instance, it isn't very clear that looping systems aren't executed again until after all the systems in a stage have had a chance to run.
This PR adds to the documentation for `ShouldRun`, and hopefully clarifies what is happening during a stage's execution when run criteria are checked and systems are being executed.
Some panic messages for systems include the system name, but there's a few panic messages which do not. This PR adds the system name for the remaining panic messages.
This is a continuation of the work done in #1864.
Related: #1846
Fixes#1895
Changed most `println` to `info` in examples, some to `warn` when it was useful to differentiate from other more noisy logs.
Added doc on `LogPlugin`, how to configure it, and why (and how) you may need to disable it
This shrinks breakout from 316k to 310k when using `--feature dynamic`.
I haven't run the ecs benchmark to test performance as my laptop is too noisy for reliable benchmarking.
There are cases where we want an enum variant name. Right now the only way to do that with rust's std is to derive Debug, but this will also print out the variant's fields. This creates the unfortunate situation where we need to manually write out each variant's string name (ex: in #1963), which is both boilerplate-ey and error-prone. Crates such as `strum` exist for this reason, but it includes a lot of code and complexity that we don't need.
This adds a dead-simple `EnumVariantMeta` derive that exposes `enum_variant_index` and `enum_variant_name` functions. This allows us to make cases like #1963 much cleaner (see the second commit). We might also be able to reuse this logic for `bevy_reflect` enum derives.
In bevy_webgl2, the `RenderResourceContext` is created after startup as it needs to first wait for an event from js side:
f31e5d49de/src/lib.rs (L117)
remove `panic` introduced in #1965 and log as a `warn` instead
After running `bevy_core` through `miri`, errors were reported surrounding incorrect memory accesses within the `bytes` test suit.
Specifically:
```
test bytes::tests::test_array_round_trip ... error: Undefined Behavior: accessing memory with alignment 1, but alignment 4 is required
--> crates/bevy_core/src/bytes.rs:55:13
|
55 | (*ptr).clone()
| ^^^^^^ accessing memory with alignment 1, but alignment 4 is required
|
```
and
```
test bytes::tests::test_vec_bytes_round_trip ... error: Undefined Behavior: accessing memory with alignment 2, but alignment 4 is required
--> /home/nward/.rustup/toolchains/nightly-x86_64-unknown-linux-gnu/lib/rustlib/src/rust/library/core/src/slice/raw.rs:95:14
|
95 | unsafe { &*ptr::slice_from_raw_parts(data, len) }
| ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ accessing memory with alignment 2, but alignment 4 is required
|
```
Solution:
The solution is to use `slice::align_to` method to ensure correct alignment.
This implementations allows you
convert std::vec::Vec<T> to VertexAttributeValues::T and back.
# Examples
```rust
use std::convert::TryInto;
use bevy_render::mesh::VertexAttributeValues;
// creating vector of values
let before = vec![[0_u32; 4]; 10];
let values = VertexAttributeValues::from(before.clone());
let after: Vec<[u32; 4]> = values.try_into().unwrap();
assert_eq!(before, after);
```
Co-authored-by: aloucks <aloucks@cofront.net>
Co-authored-by: simens_green <34134129+simensgreen@users.noreply.github.com>
Co-authored-by: Carter Anderson <mcanders1@gmail.com>
This was nowhere documented inside Bevy.
Should I also mention the use case of debugging a project?
Closes#810
Co-authored-by: MinerSebas <66798382+MinerSebas@users.noreply.github.com>
We discussed with @alice-i-cecile privately on iterators and agreed that making a custom ordered iterator over query makes no sense since materialization is required anyway and it's better to reuse existing components or code. Therefore, just adding an example to the documentation as requested.
Fixes#1470.
Co-authored-by: Carter Anderson <mcanders1@gmail.com>
Introduced in #1778, not fixed by #1931
The size of `Lights` buffer currently is :
```rust
16 // (color, `[f32; 4]`)
+ 16 // (number of lights, `f32` encoded as a `[f32; 4]`)
+ 10 // (maximum number of lights)
* ( 16 // (light position, `[f32; 4]`
+ 16 // (color, `[16; 4]`)
+ 4 // (inverse_range_squared, `f32`)
)
-> 392
```
This makes the pbr shader crash when running with Xcode debugger or with the WebGL2 backend. They both expect a buffer sized 512. This can also be seen on desktop by adding a second light to a scene with a color, it's position and color will be wrong.
adding a second light to example `load_gltf`:
```rust
commands
.spawn_bundle(PointLightBundle {
transform: Transform::from_xyz(-3.0, 5.0, -3.0),
point_light: PointLight {
color: Color::BLUE,
..Default::default()
},
..Default::default()
})
.insert(Rotates);
```
before fix:
<img width="1392" alt="Screenshot 2021-04-16 at 19 14 59" src="https://user-images.githubusercontent.com/8672791/115060744-866fb080-9ee8-11eb-8915-f87cc872ad48.png">
after fix:
<img width="1392" alt="Screenshot 2021-04-16 at 19 16 44" src="https://user-images.githubusercontent.com/8672791/115060759-8cfe2800-9ee8-11eb-92c2-d79f39c7b36b.png">
This PR changes `inverse_range_squared` to be a `[f32; 4]` instead of a `f32` to have the expected alignement
Fixes#1921
Buffer was growing with the actual number of lights instead of being limited to the max number of lights.
As it's a query that can be exactly sized, I also switched `count()` to `len()`
This includes a lot of single line comments where either saying more wasn't helpful or due to me not knowing enough about things yet to be able to go more indepth. Proofreading is very much welcome.
Allows render resources to move data to the heap by boxing them. I did this as a workaround to #1892, but it seems like it'd be useful regardless. If not, feel free to close this PR.
I've had problems with compiling and running the pbr example:
```
thread 'main' panicked at 'called `Result::unwrap()` on an `Err` value: Compilation("glslang_shader_preprocess:\nInfo log:\nERROR: 0:40: \'#version\' : must occur first in shader \nERROR: 0:40: \'#version\' : bad profile name; use es, core, or compatibility \nERROR: 0:40: \'#version\' : bad tokens following profile -- expected newline \nERROR: 3 compilation errors. No code generated.\n\n\nDebug log:\n\n")', crates/bevy_render/src/pipeline/pipeline_compiler.rs:161:22
```
I've checked each shader, and only one shader hasn't had `#version` in the first line.
This change fixed my issue.
Fixes#1846
Got scared of the other "Requested resource does not exist" error at line 395 in `system_param.rs`, under `impl<'a, T: Component> SystemParamFetch<'a> for ResMutState<T> {`. Someone with better knowledge of the code might be able to go in and improve that one.
Implements `Byteable` and `RenderResource` for any array containing `Byteable` elements. This allows `RenderResources` to be implemented on structs with arbitrarily-sized arrays, among other things:
```rust
#[derive(RenderResources, TypeUuid)]
#[uuid = "2733ff34-8f95-459f-bf04-3274e686ac5f"]
struct Foo {
buffer: [i32; 256],
}
```
From suggestion from Godot workflows: https://github.com/bevyengine/bevy/issues/1730#issuecomment-810321110
* Add a feature `bevy_debug` that will make Bevy read a debug config file to setup some debug systems
* Currently, only one that will exit after x frames
* Could add option to dump screen to image file once that's possible
* Add a job in CI workflow that will run a few examples using [`swiftshader`](https://github.com/google/swiftshader)
* This job takes around 13 minutes, so doesn't add to global CI duration
|example|number of frames|duration|
|-|-|-|
|`alien_cake_addict`|300|1:50|
|`breakout`|1800|0:44|
|`contributors`|1800|0:43|
|`load_gltf`|300|2:37|
|`scene`|1800|0:44|
Fixes#1809. It makes it also possible to use `derive` for `SystemParam` inside ECS and avoid manual implementation. An alternative solution to macro changes is to use `use crate as bevy_ecs;` in `event.rs`.
The `VertexBufferLayout` returned by `crates\bevy_render\src\mesh\mesh.rs:308` was unstable, because `HashMap.iter()` has a random order. This caused the pipeline_compiler to wrongly consider a specialization to be different (`crates\bevy_render\src\pipeline\pipeline_compiler.rs:123`), causing each mesh changed event to potentially result in a different `PipelineSpecialization`. This in turn caused `Draw` to emit a `set_pipeline` much more often than needed.
This fix shaves off a `BindPipeline` and two `BindDescriptorSets` (for the Camera and for global renderresources) for every mesh after the first that can now use the same specialization, where it didn't before (which was random).
`StableHashMap` was not a good replacement, because it isn't `Clone`, so instead I replaced it with a `BTreeMap` which is OK in this instance, because there shouldn't be many insertions on `Mesh.attributes` after the mesh is created.
related to #1700
This PR:
* documents all methods on `Input<T>`
* adds documentation on the struct about how to use it, and how to implement it for a new input type
* renames method `update` to a easier to understand `clear`
* adds two methods to check for state and clear it after, allowing easier use in the case of #1700
Co-authored-by: Carter Anderson <mcanders1@gmail.com>
- prints glsl compile error message in multiple lines instead of `thread 'main' panicked at 'called Result::unwrap() on an Err value: Compilation("glslang_shader_parse:\nInfo log:\nERROR: 0:335: \'assign\' : l-value required \"anon@7\" (can\'t modify a uniform)\nERROR: 0:335: \'\' : compilation terminated \nERROR: 2 compilation errors. No code generated.\n\n\nDebug log:\n\n")', crates/bevy_render/src/pipeline/pipeline_compiler.rs:161:22`
- makes gltf error messages have more context
New error:
```rust
thread 'Compute Task Pool (5)' panicked at 'Shader compilation error:
glslang_shader_parse:
Info log:
ERROR: 0:12: 'assign' : l-value required "anon@1" (can't modify a uniform)
ERROR: 0:12: '' : compilation terminated
ERROR: 2 compilation errors. No code generated.
', crates/bevy_render/src/pipeline/pipeline_compiler.rs:364:5
```
These changes are a bit unrelated. I can open separate PRs if someone wants that.
