Feature Request NativeArray overloads of Physics.RaycastNonAlloc etc for usability from Burst code

I recently discovered that the

Physics.Raycast

etc functions are directly callable from Burst code, which was a wonderful revelation as it would allow me to significantly tidy up and speed up my character controller, which is currently half Bursted and half not. These hopes were immediately dashed when I realised that the

Physics.RaycastAll

and

Physics.RaycastNonAlloc

versions are not Burstable, as all of their overloads use managed arrays!

Could we have

NativeArray

overloads to enable this please?

I'm talking about all of the cast functions, so ray, box, sphere, etc.

Physics.RaycastAll(Ray ray, Allocator allocator): NativeArray<RaycastHit>

and

Physics.RaycastNonAlloc(Ray ray, NativeArray<RaycastHit> results)

. Same too for

OverlapSphereNonAlloc

etc.

While we're at it, perhaps

NativeList

versions of

RaycastNonAlloc

would also make sense to generally avoid allocations, but guarantee all results.

Perhaps this thread would be better in the Physics forum, but my motivation comes from the Burst experience, and this forum allows feature request tags and that one doesn't.

 

Yeah in hindsight this was the wrong forum, but thank you very much!

 

I just realised that I was a bit silly in requesting

OverlapSphereNonAlloc

as that returns managed colliders, so that's a bit harder than the casts. An unmanaged collider type that would allow more Burstable physics would be a lovely addition though! That would unlock things like

ComputePenetration

too.

I've got quite a few character controllers that are very procedural. The basic flow of each is something like

1. A load of matrix maths
2. A few physics queries
3. A load more matrix maths
4. A Movement integration loop that does about 10 iterations of queries and matrix maths
5. A final batch of matrix maths

Each of the matrix maths heavy stages benefits significantly from being Bursted, but the need to repeatedly drop out of Burst to do the queries is frustrating the whole process and negating the benefits. I have to shuttle a lot of data back and forth across the Burst boundary each time. As I'm iterating on the design I sometimes want to add one extra physics query in one place, but that particular query isn't Burstable, so I have to restructure.

In my opinion, the usefulness of Burst in non ECS projects would be significantly improved by tighter integration with the physics APIs. I'm finding that this frustration comes up a lot. In most places it's ok to do a big batch of data processing in Burst then pass the results to the engine at the end, but physics is one of the few APIs that performance critical code often wants to access rapidly.

Perhaps it would be useful for the Burst and physics teams to work together on this workflow at some point.