Are dynamic fields as a struct member supported?

[edoardogiacomello]

Hello!
I'm trying to build a simulation in which I have a dynamic field of particles that belong to different objects. This is needed because in my simulation, new particles and object could spawn at runtime and I want to keep it flexible/efficient enough.

The particles are implemented as follow:

T_part = ti.types.struct(pos=ti.math.vec2, ..., obj_id=int)
S_part = ti.root.dynamic(ti.i, 1000, chunk_size=32)
parts = T_part.field()
S_part.place(parts)

This seems to work, and I can access and add new elements by indexing. As a side note, a for loop iterates over all 1000 elements even if parts contains only a few active elements. I'm satisfied looping over range(parts.length()) for now, but the docs state that for loops should automatically take care of everything (?).

However, these particles belong to some objects that have different properties that are not directly related to the particles.
What I would like to accomplish is to have a dynamic list as a member of the each object struct containing the indices (or pointers?) to the particles belonging to that object. This would allow me to access an object from a particle, but also to access all the particles belonging to a given object:

T_obj = ti.types.struct(center=ti.math.vec2, volume=float, ..., ) # HERE I WOULD WANT A DYNAMIC LIST OF ALL POINTS!
S_obj = ti.root.dynamic(ti.i, 1000, chunk_size=32)
obj = T_obj.field()
S_obj.place(obj)

is that possible?

[hughperkins]

For your point about looping, note that the array is allocated in chunks, in your case 32. So, it's going to loop over every eleement in each entire chunk. eg try the following to see it only loops over activet chunks:

import taichi as ti
import numpy as np

ti.init(arch=ti.cpu, debug=True)

T_part = ti.types.struct(pos=ti.math.vec2, obj_id=int)
S_part = ti.root.dynamic(ti.i, 50, chunk_size=3)
parts = T_part.field()
S_part.place(parts)

@ti.kernel
def loop_parts():
    print("loop parts")
    for i in parts:
        print('i', i)
    print('')

parts[2] = T_part(ti.math.vec2(1, 2), 3)
parts[3] = T_part(ti.math.vec2(1, 2), 3)

loop_parts()

parts[13] = T_part(ti.math.vec2(1, 2), 3)

loop_parts()

oUtput:

loop parts
i 0
i 1
i 2
i 3

loop parts
i 0
i 3
i 11
i 12
i 13
i 4
i 8
i 5
i 9
i 6
i 10
i 1
i 7
i 2