Don't panic when applying an Insert command

[SludgePhD]

Objective

• Fixes Potential surprising "entity does not exist" panic due to asynchronous command execution #7118

Solution

• Remove the delayed panic caused by EntityCommands::insert, by ignoring the command if its entity has been despawned
• Make all builtin commands behave consistently when acting on an entity that has since been despawned

---

Changelog

• The Insert command now no longer panics when its entity no longer exists.

[alice-i-cecile]

I think we should at least warn here: this is pretty important potential bug.

My preferred solution though is to create a class of try_verb commands though, which fail silently.

[killercup]

I would at least make it log an error, so that you can still see when this is happening.

[SludgePhD]

If we log an error or warning here, we'd basically be indicating that any occurrence of this situation is a bug that should be resolved, and I'm not convinced that that is correct.

I'm currently running into this panic because I am despawning entities in the PostUpdate schedule, and the bevy_sprite::calculate_bounds_2d system is trying to insert the Aabb component there as well.

The panic of course does not happen reliably, because the entity is only rarely despawned on the same frame as the bevy_sprite::calculate_bounds_2d system inserts the Aabb.

From what I can tell, most people running into this panic end up in a similar situation, where the correct resolution is to simply ignore the insert and despawn the entity. It does not really sound like treating this as a bug would be the right thing to do.

Perhaps the "bug" would be that bevy's internals need to use .try_insert instead of .insert pretty much everywhere, but then this strikes me as poor API design: the non-prefixed, shortest method name should ideally be the correct one to use, but here it causes nondeterministic bugs.

[JMS55]

This would also fix #9721

[alice-i-cecile]

This was added as opt-in behavior in #9844.

[SludgePhD]

Were bevy's internal systems ever fixed to use it?

[alice-i-cecile]

No, I would appreciate issues for the places where this is the correct choice :) Or a single issue might be fine and the implementor can track them down.

We just ran into one over in #9721, and I suspect that there's more to fix.