Allow components to perform cleanup (with access to the World) when they are removed
LoganDark opened this issue · comments
Description
Allow components one last access to the world when they are deleted. This could be done by either:
- Allowing some sort of call inside of
Dropthat gets the encompassing World (unlikely, probably really unsafe, stinks of globals) - Creating some sort of trait that acts like Drop but allows the component to access the world like a System does. Perhaps it could be named
Cleanupor something?
Motivation
I have a component that represents ownership over some physics objects. When the entity holding that component is deleted, I want the physics objects to also be deleted. The problem is that inside Drop, I have no access to the physics world because it's a resource. :(
Drawbacks
- Is it a breaking change? No, features are only added
- Can it impact performance, learnability, etc? The performance of deleting large amounts of entities with components that require cleanup may be impacted. However, due to monomorphization, if you go with the cleanup trait then there will be no code generated for components without it, just like
Drop. That means only code that explicitly opts into this feature will be impacted
Unresolved questions
None right now
wha twhy does enter submit the issue
give me a minute...
I think the main barrier here is that typically other parts of the world can be borrowed mutably and immutably while removing components from a particular storage.
I think the most effective way to do this is to have a zero-sized 'delete' marker component and a system that runs at the end of the tick to go through entities with this marker performing whatever operation you want to do with them and then deleting them.
I think the most effective way to do this is to have a zero-sized 'delete' marker component and a system that runs at the end of the tick to go through entities with this marker performing whatever operation you want to do with them and then deleting them.
maaaayybbeeee? It feels extremely awkward and very not right but I guess it would technically work?
@LoganDark This is the most correct way to do this within the ECS paradigm. Zero-sized marker components are extremely cheap (they're usually completely allocation-freeze).