Using TAA causing panic when entity is despawned

[PatrickChodowski]

Hello,

I added TAA to my game and I noticed it causes panic whenever I reload the scene or randomly when one of entities gets despawned:

thread 'main' panicked at 'error[B0003]: Could not insert a bundle (of type `bevy_pbr::prepass::PreviousGlobalTransform`) for entity 21v0 because it doesn't exist in this World.', /home/patrick/.cargo/registry/src/index.crates.io-6f17d22bba15001f/bevy_ecs-0.11.2/src/system/commands/mod.rs:957:13
note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace
Encountered a panic when applying buffers for system `bevy_pbr::prepass::update_mesh_previous_global_transforms`!
Encountered a panic in system `bevy_app::main_schedule::Main::run_main`!

Is there a way to avoid it? Should I set the TemporalAntiAliasSettings::reset to true everytime I am about to despawn entities? Or maybe there is a way for TAA to not track certain entities? Thank you for help

[nicopap]

This is probably a bevy bug. I opened an issue for you

[PatrickChodowski]

Thank you!

[alice-i-cecile]

Said issue: #9721

[JMS55]

I worked on TAA, thanks for bringing this up!

I can't figure out why this is happening. Are you deleting any entities in PreUpdate by chance?

[JMS55]

What's the solution to this do you think? Stick an apply_deferred right after update_mesh_previous_global_transforms?

[nicopap]

Hue tough one.

I think this needs a different approach:

• The "insert PreviousGlobalTransform component" step should be separated from the "update PreviousGlobalTransform component" (supposedly in system with a With<Handle<Mesh>>, Without<PreviousGlobalTransform>)
• Now update_mesh_previous_global_transforms only mutates an existing component, no need to worry about command synchronization.

[JMS55]

How does that prevent the crash? The user could still delete the entity after 1., before commands are applied...

[nicopap]

Well, with this, the command only occurs the frame after the entity is spawned, rather than every frame. Which is not often the frame when the entity is despawned… Not optimal, maybe feels like dusting the crap under the rug wrt this issue specifically. But in terms of performance, it's definitively better, since you are only spending time in exclusive-world-access world when the component is added, not every frame for each mesh entity with a global transform.

To fix the final issue, there should be a non-failing insert command. The original Insert is defined there:

bevy/crates/bevy_ecs/src/system/commands/mod.rs

Lines 942 to 960 in 17edf4f

	pub struct Insert<T> {
	/// The entity to which the components will be added.
	pub entity: Entity,
	/// The [`Bundle`] containing the components that will be added to the entity.
	pub bundle: T,
	}
	impl<T> Command for Insert<T>
	where
	T: Bundle + 'static,
	{
	fn apply(self, world: &mut World) {
	if let Some(mut entity) = world.get_entity_mut(self.entity) {
	entity.insert(self.bundle);
	} else {
	panic!("error[B0003]: Could not insert a bundle (of type `{}`) for entity {:?} because it doesn't exist in this World.", std::any::type_name::<T>(), self.entity);
	}
	}
	}

[alice-i-cecile]

A try_insert command was added in #9844 and will be releasing with Bevy 0.12.