Fix panic when a mesh has skinning attributes but no `SkinnedMesh` component #18074

greeble-dev · 2025-02-27T21:02:54Z

PR is in draft mode as it needs to be redone following other changes. Currently waiting on #18762 and #18763 to land.

Objective

Fix panic when mesh has skinning attributes but no SkinnedMesh component.
- Fixes Bevy Panics when trying to load a single mesh from gltf file containing armature #16929
Also fix the inverse bug (entity has SkinnedMesh component but mesh has no skinning attributes).
As a bonus, the fixes are a small optimisation (saves 8.5us on many_foxes).
Add a test that reproduces most of these bugs and is suitable for CI.

Summary

My goal was to fix #16929 - a wgpu panic when a mesh with skinning attributes is instantiated as a Mesh3d but without a SkinnedMesh component.

The problem was:

setup_morph_and_skinning_defs chooses a skinning bind group layout if the mesh has skinning attributes.
SetMeshBindGroup::render chooses a skinning bind group if the mesh is in skin_uniforms.
This breaks if a mesh has skinning attributes but isn't in skin_uniforms (because it doesn't have a SkinnedMesh component).

The fix is to make skin_uniforms the source of truth.

This was mostly the case already - the exception was setup_morph_and_skinning_defs. So I changed the pipeline specialization functions to check skin_uniforms, set a new MeshPipelineKey::SKINNED, and then pick that up in setup_morph_and_skinning_defs.

While working this through I noticed a few more issues.

Incorrect Pipeline Caches

The pipeline specialization used RenderMeshInstanceFlags::HAS_PREVIOUS_SKIN to decide if the mesh needed motion blur. But that flag is set by the set_mesh_motion_vector_flags system, which can run after specialization. So in some cases the wrong pipeline was cached and skinned meshes had no motion blur.

This is fixed as a side effect of using skin_uniforms as the source of truth. It also means RenderMeshInstanceFlags::HAS_PREVIOUS_SKIN can be removed, and set_mesh_motion_vector_flags no longer has to iterate over all skinned entities.

The bad news is that HAS_PREVIOUS_MORPH follows a similar pattern, so I suspect motion blur is broken on morph targets in some cases.

Previous Buffers Too Small

When testing with skins_use_uniform_buffers forced on I got a panic about SkinUniforms::prev_buffer being too small for the mapped range. I think the problem is that prepare_skins grows current_buffer without also growing prev_buffer, so it's too small for one frame.

I've changed prepare_skins to grow both buffers. However, this will cause a single frame visual glitch as the new prev_buffer is initialised with the current frame's joints. For now I think reducing a panic to a glitch is sufficient.

I suspect there's also a bug where prev_buffer has wrong transforms the first frame after a skin is added. And at the last minute I realised MeshPipelineKey::HAS_PREVIOUS_SKIN is probably obsolete, since all skins now have a previous. I might chase these down in a follow-up PR.

System Parameter Limits

When adding a skin_uniforms parameter to systems like specialize_material_meshes I hit a limit on the number of parameters. I fixed this by moving some parameters into a SystemParam shared between specialize systems (85d1a00).

New Test

The PR adds a test that repros most of the issues. It's designed to be run on CI to detect panics, but can be visually inspected as well. I haven't actually added it to the CI - I wasn't sure whether to do that in this PR or a follow-up.

The test currently spams warnings as it repros a bug not addressed by this PR (SkinnedMesh without skinning attributes - the inverse of #16929).

Follow Ups

#9360 made the glTF importer try to skip any meshes that would triggered the bug. Arguably this is no longer necessary, but that's best handled in a separate PR as it may be contentious.

Performance

Tested with many_foxes. The cost of set_mesh_motion_vector_flags is much reduced (8.97us -> 0.3us).

Testing

cargo run --example invalid_skinned_mesh
cargo run --example animated_mesh
cargo run --example custom_skinned_mesh

Tested on Win10/Vulkan/Nvidia, Wasm/Chrome/WebGL2/Nvidia.

…fixed forward mode missing passes required for motion blur.

- Removed render mode cycling and motion blur toggling - all renderer paths can be tested just by enabling motion blur. - Added a text overlay documenting each mesh. - Various tidy ups.

…h for whether a mesh is skinned. This meant adding `MeshPipelineKey::SKINNED`. - Fixed warning spam by making `extract_skins` check if the mesh has skinning attributes. - Fixed panic when using uniform buffers for skinning. If `SkinUniforms::current_buffer` is grown the `SkinUniforms::prev_buffer` should grow too.

crates/bevy_pbr/src/material.rs

crates/bevy_pbr/src/render/light.rs

greeble-dev · 2025-02-27T21:07:37Z

crates/bevy_pbr/src/render/skin.rs

+            size: new_size,
+            mapped_at_creation: false,
+        });
+        // TODO: This is wrong. We should be using last frame's data for prev_buffer.


Awkward TODO - see "Previous Buffers Too Small" section.

alice-i-cecile · 2025-02-27T21:39:46Z

The fix is to make extract_skins check the attributes. Unfortunately that adds complication and probably a performance cost (see below), which is annoying when it only fixes a case that's not actually useful. I can drop/separate this change if it's controversial.

Let's drop that; the rest of this is a clear improvement that we shouldn't let get bogged down.

… but no skinning attributes.

greeble-dev · 2025-02-27T23:22:20Z

I've reverted the attribute checks in extract_skins. This removes the performance issue. It also means the invalid_skinned_mesh example will now spam warnings, but I'm assuming that's a problem for a follow up PR.

I'll investigate using SystemParams tomorrow.

…ther.

…ms`.

greeble-dev · 2025-02-28T09:28:05Z

I've changed the specialize systems to take a SpecializeParams struct. This is mainly to dodge the limit on system parameters, but is arguably a bit cleaner too. I could have put more shared params in there but decided to keep things minimal for this PR.

Note that the struct is pub because specialize_material_meshes is pub. I suspect that's wrong - it should be pub(crate) at most? But again, keeping things minimal.

greeble-dev · 2025-04-04T08:23:05Z

I've pulled one of the fixes out to #18715. Also put this PR into draft mode - once the other PR shakes out I'll need to update the description and fix conflicts.

## Objective Fix motion blur not working on skinned meshes. ## Solution `set_mesh_motion_vector_flags` can set `RenderMeshInstanceFlags::HAS_PREVIOUS_SKIN` after specialization has already cached the material. This can lead to `MeshPipelineKey::HAS_PREVIOUS_SKIN` never getting set, disabling motion blur. The fix is to make sure `set_mesh_motion_vector_flags` happens before specialization. Note that the bug is fixed in a different way by #18074, which includes other fixes but is a much larger change. ## Testing Open the `animated_mesh` example and add these components to the `Camera3d` entity: ```rust MotionBlur { shutter_angle: 5.0, samples: 2, #[cfg(all(feature = "webgl2", target_arch = "wasm32", not(feature = "webgpu")))] _webgl2_padding: Default::default(), }, #[cfg(all(feature = "webgl2", target_arch = "wasm32", not(feature = "webgpu")))] Msaa::Off, ``` Tested on `animated_mesh`, `many_foxes`, `custom_skinned_mesh`, Win10/Nvidia with Vulkan, WebGL/Chrome, WebGPU/Chrome. Note that testing `many_foxes` WebGL requires #18715.

## Objective Fix #18714. ## Solution Make sure `SkinUniforms::prev_buffer` is resized at the same time as `current_buffer`. There will be a one frame visual glitch when the buffers are resized, since `prev_buffer` is incorrectly initialised with the current joint transforms. Note that #18074 includes the same fix. I'm assuming this smaller PR will land first. ## Testing See repro instructions in #18714. Tested on `animated_mesh`, `many_foxes`, `custom_skinned_mesh`, Win10/Nvidia with Vulkan, WebGL/Chrome, WebGPU/Chrome.

## Objective Add a test that would have caught #16929 and #18712. ## Solution The PR adds a `test_invalid_skinned_mesh` example that creates various valid and invalid skinned meshes. This is designed to catch panics via CI, and can be inspected visually. It also tests skinned meshes + motion blur. ![417958065-37df8799-b068-48b8-87a4-1f144e883c9c](https://github.com/user-attachments/assets/22b2fa42-628b-48a4-9013-76d6524cecf5) The screenshot shows all the tests, but two are currently disabled as they cause panics. #18074 will re-enable them. ### Concerns - The test is not currently suitable for screenshot comparison. - I didn't add the test to CI. I'm a bit unsure if this should be part of the PR or a follow up discussion. - Visual inspection requires understanding why some meshes are deliberately broken and what that looks like. - I wasn't sure about naming conventions. I put `test` in the name so it's not confused with a real example. ## Testing ``` cargo run --example test_invalid_skinned_mesh ``` Tested on Win10/Nvidia, across Vulkan, WebGL/Chrome, WebGPU/Chrome.

greeble-dev added 4 commits February 25, 2025 18:06

Added a test that triggers bevy/bevyengine#16929.

f2b8c76

Fixed unskinned mesh missing indices (doh!). Rearranged ordering and …

c4a32a0

…fixed forward mode missing passes required for motion blur.

Updated test.

70d46cf

- Removed render mode cycling and motion blur toggling - all renderer paths can be tested just by enabling motion blur. - Added a text overlay documenting each mesh. - Various tidy ups.

greeble-dev commented Feb 27, 2025

View reviewed changes

crates/bevy_pbr/src/material.rs Outdated Show resolved Hide resolved

greeble-dev commented Feb 27, 2025

View reviewed changes

crates/bevy_pbr/src/render/light.rs Outdated Show resolved Hide resolved

greeble-dev commented Feb 27, 2025

View reviewed changes

greeble-dev added 2 commits February 27, 2025 21:27

Merge branch 'bevyengine:main' into issue-16929-skinned-mesh-crash-v2

9f71ec6

Fixed CI fail following bevyengine#17905 changes to ChildOf.

a2fc0c2

alice-i-cecile added C-Bug An unexpected or incorrect behavior A-Rendering Drawing game state to the screen P-Crash A sudden unexpected crash A-Animation Make things move and change over time S-Needs-Review Needs reviewer attention (from anyone!) to move forward labels Feb 27, 2025

alice-i-cecile requested a review from pcwalton February 27, 2025 21:38

github-project-automation bot added this to Rendering Feb 27, 2025

alice-i-cecile added this to the 0.16 milestone Feb 27, 2025

greeble-dev added 2 commits February 27, 2025 21:55

Reverted fix for warning spam when mesh has a SkinnedMesh component…

9def15d

… but no skinning attributes.

Fixed docs warning ("extract_morphs is private").

0e4e9c1

greeble-dev added 2 commits February 28, 2025 09:02

Tweaked example visuals, mainly to distinguish the meshes from each o…

02a52fb

…ther.

Refactored specialize system parameters into a shared `SpecializePara…

85d1a00

…ms`.

Fixed test failing on WebGL - MSAA and MotionBlur can't be combined.

37b5a8d

alice-i-cecile removed this from the 0.16 milestone Mar 9, 2025

greeble-dev mentioned this pull request Apr 4, 2025

Fix motion blur on skinned meshes #18712

Merged

greeble-dev marked this pull request as draft April 4, 2025 08:19

greeble-dev mentioned this pull request Apr 4, 2025

Fix many_foxes + motion blur = crash on WebGL #18715

Merged

This was referenced Apr 8, 2025

Refactor shared specialize params into a SystemParam #18762

Open

Add test for invalid skinned meshes #18763

Merged

greeble-dev added S-Waiting-on-Author The author needs to make changes or address concerns before this can be merged and removed S-Needs-Review Needs reviewer attention (from anyone!) to move forward labels May 5, 2025

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Uh oh!

Fix panic when a mesh has skinning attributes but no `SkinnedMesh` component #18074

Fix panic when a mesh has skinning attributes but no `SkinnedMesh` component #18074

greeble-dev commented Feb 27, 2025 •

edited

Loading

Uh oh!

Uh oh!

Uh oh!

greeble-dev Feb 27, 2025

Uh oh!

alice-i-cecile commented Feb 27, 2025

Uh oh!

greeble-dev commented Feb 27, 2025

Uh oh!

greeble-dev commented Feb 28, 2025

Uh oh!

greeble-dev commented Apr 4, 2025

Uh oh!

Uh oh!

Uh oh!

Fix panic when a mesh has skinning attributes but no SkinnedMesh component #18074

Are you sure you want to change the base?

Fix panic when a mesh has skinning attributes but no SkinnedMesh component #18074

Conversation

greeble-dev commented Feb 27, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Objective

Summary

Incorrect Pipeline Caches

Previous Buffers Too Small

System Parameter Limits

New Test

Follow Ups

Performance

Testing

Uh oh!

Uh oh!

Uh oh!

greeble-dev Feb 27, 2025

Choose a reason for hiding this comment

Uh oh!

alice-i-cecile commented Feb 27, 2025

Uh oh!

greeble-dev commented Feb 27, 2025

Uh oh!

greeble-dev commented Feb 28, 2025

Uh oh!

greeble-dev commented Apr 4, 2025

Uh oh!

Uh oh!

Fix panic when a mesh has skinning attributes but no `SkinnedMesh` component #18074

Fix panic when a mesh has skinning attributes but no `SkinnedMesh` component #18074

greeble-dev commented Feb 27, 2025 •

edited

Loading