Open hymm opened 1 year ago
Previous thinking about atomic systems: https://github.com/bevyengine/rfcs/pull/46
This is different feature from that rfc. This is atomic in the same sense as AtomicBool types that are safe to use from multi threaded contexts
Instead of using atomics, another option would be to store the shared ticks in a resource.
Let's not mess with system internals for this? There are few ways this could be resolved without making all systems more complicated forever...
apply_deferred
, we could do the same for propagate_transforms
(and friends). It could let propagate_transforms
run, save its change tick, and write that to the next instance of propagate_transforms
.propagate_transforms
that all schedules use.I'd similarly like to avoid complicating system internals here if we can.
I think that the systems-as-entities (or other "unified storage") solution is by far my favorite here. I think you also want to be able to share and reuse other local / cached data. Make an app.reuse_system(schedule_label, system)
API that does map-like lookup.
Special-casing this is both a questionable architectural choice, and not feasible for this case due to circular dependencies.
What problem does this solve or what need does it fill?
Sometimes you want to add the transform propagation systems to the schedule in multiple places. The transform systems use change detection to not propagate transforms that haven't changed. This will do extra work because the propagation will run for any transforms changed since the current system has run instead of transforms changed since the last time any transform propagation has run.
What solution would you like?
Probably some type of
AtomicTicks
andAtomicSystem
that are passed inSystem::run
. This is a little tricky since you'd have to clone instances of the system when adding them. For Transform propagation we could potentially store the systems on the Plugin.What alternative(s) have you considered?
I was also thinking of an ArcSystem, but I don't think that can work without locking since we take mutable access to a system to update it's archetypes.