Whats an efficient way to make a memory copy of some components from entities

[sonic matrix]

For each frame, I need to compute the directions of my units based on the Position and LinearVelocity of all other units. (Similar to the swarm (boid) simulation).
If I allocate a NativeArray like the one below, can I assume they are in the same order?

Furthermore, how do I get access to a LinearVelocity component? It is part of the RigidBodyAspect but I cannot use it to fill a NativeArray.

NativeArray<LocalTransform> localTransformMemory = m_WalkingAgents.ToComponentDataArray<LocalTransform>(Allocator.TempJob);
        NativeArray<Physics.LinearVelocity> localVelocityMemory = m_WalkingAgents.ToComponentDataArray<LocalTransform>(Allocator.TempJob);

If that assumption does not hold. How do I best create a data structure that allows me to create a copy of these two types of data efficiently.

Other approaches could be:

1. Using a NativeParallelHashMap<> for each data type with Entity as the key. I am not sure how an efficient data allocation would look, similar to ToComponentDataArray.
2. Create another data type in that someone combines the elements.

[steady stratus]

If I allocate a NativeArray like the one below, can I assume they are in the same order?
Yes. Though, I am assuming you aren't using a query with one of those components as optional, I don't know what that would do.

Furthermore, how do I get access to a LinearVelocity component? It is part of the RigidBodyAspect but I cannot use it to fill a NativeArray.
When you inspect the code, you will find the rigidbody aspect uses the PhysicsVelocity component for that, so you could just get that structure instead

to compute the directions of my units based on the Position and LinearVelocity of all other units

if you're tackling this in a brute-force manner, even with parallelization, you will quickly find it becomes very expensive; usually there's some kind of spatial subdivision to make this work efficiently

of course it depends on the number of entities whether this is a real problem or not

usually though, instead of allocating arrays like that, you would create a job to iterate over the data for you; this would save you the cost of creating the array

but if you are just starting out, it's not too complicated to change that later

[sonic matrix]

Hmm ok... I am using the created arrays in a separate parallel scheduled job that computes new directions for each unit very much in a swarm simulation... which is why I want to have the input data separately.

Maybe there is a way to do it with a parallel for-ach loop.

I'll keep experimenting. thank you!!

Ill let you know what worked (for completion and documentation! )