Are nested structarrays valid in numba?

Numba Community Support
1

Hello friendly people :wink:

The documentation seems to be a bit unclear about this, so thought I’ll double check here:

base_dtype = np.dtype([(f'x', np.uint32), (f'y', np.uint16), (f'z', np.uint8)])
n_types = 5
complex_dtype = np.dtype([('global1', np.uint32), ('global2', np.int64), ('comp', base_dtype, n_types)])
test_var_arr = np.empty(10, dtype=complex_dtype)
@nb.njit()
def test(struct_arr):
    struct_arr.view(np.uint8).fill(0)
    for ix in range(n_types):
        value_offset = ix * 10
        struct_arr[0]['comp'][ix]['x'] = value_offset + 1
        struct_arr[0]['comp'][ix]['y'] = value_offset + 2
        struct_arr[0]['comp'][ix]['z'] = value_offset + 3

This seems to compile just fine - am I good to go, or should I be wary of something here?

I want to rely heavily on this feature so don’t want to encounter a hiccup on the way :slight_smile:

Many thanks in advance - Adrian

2

Can you point me toward the docs that were unclear? My understanding was that structured arrays were supported in numba.

3

Many thanks for getting back.

Looking at the docs I’m getting some conflicting messages:

The following scalar types and features are not supported:

  • Arbitrary Python objects
  • Half-precision and extended-precision real and complex numbers
  • Nested structured scalars the fields of structured scalars may not contain other structured scalars

[…]

Array types

NumPy arrays of any of the scalar types above are supported, regardless of the shape or layout.

After doing some digging it seems this has been asked before 4 years ago here, but I think it may be worth getting back to this. Seems the structarrays have later been straightened in this PR .

If structarrays are good, I’d be happy to offer a PR to improve the wording of the documentation, as I believe this is a gotcha for many other users.