conform7 VCV propagation raises ValueError with NumPy ≥ 2.x #181

When a vcv is passed to conform7 and the transformation has a tf_sd, the call fails while building the Jacobian. Every function that calls conform7 with a vcv fails too: conform14, plate_motion_transformation, transform_atrf2014_to_gda2020 and transform_gda2020_to_atrf2014, and both MGA94↔MGA2020 functions.

To reproduce

import numpy as np
from geodepy.transform import conform7
from geodepy.constants import gda94_to_gda2020

conform7(-4052051.7643, 4212836.2017, -2545106.0245,
         gda94_to_gda2020, vcv=np.diag([1e-4, 1e-4, 1e-4]))

TypeError: only 0-dimensional arrays can be converted to Python scalars
...
  File "geodepy/transform.py", line 98, in conform7
    j_mat[0, 3] = rot_xyz[0]
ValueError: setting an array element with a sequence.

Environment: NumPy 2.4.2, Python 3.14.

Cause

xyz_before and rot_xyz are 3×1 column vectors, so rot_xyz[0] and xyz_before[2] return 1-element arrays of shape (1,) rather than scalars. Assigning them to single cells of j_mat (transform.py:98-108) depended on NumPy converting size-1 arrays to scalars implicitly. NumPy deprecated that in 1.25 and now raises an error. The transformed coordinates are unaffected because lines 59–61 already index with [i][0].

The existing tests didn't catch this because none of them passes a vcv.

Proposed fix

Index the scalars explicitly (rot_xyz[i, 0], xyz_before[i, 0]), or work with 1-D vectors throughout:

xyz_before = np.array([x, y, z])
rot_xyz = rotation @ xyz_before
...
j_mat[:, 0:3] = scale * rotation
j_mat[:, 3] = rot_xyz
j_mat[0, 5], j_mat[0, 6] = -scale * z, scale * y
j_mat[1, 4], j_mat[1, 6] = scale * z, -scale * x
j_mat[2, 4], j_mat[2, 5] = -scale * y, scale * x
j_mat[:, 7:10] = np.eye(3)

After the fix, the Jacobian matches a finite-difference check to within about 5×10⁻⁷ per column, so only the indexing needs to change.

Suggested regression test (ALIC, GDA2020 Tech Manual §3.1.1, input VCV = diag(1e-4) m²)

Expected vcv_out (m²):

[[ 1.007199e-04  -1.221569e-07   7.888910e-08]
 [-1.221569e-07   1.006026e-04  -7.672700e-08]
 [ 7.888910e-08  -7.672700e-08   1.006438e-04]]

The transformation adds about 0.85, 0.78 and 0.80 mm (1σ) in X, Y and Z.

Bug Report Created with Claude 🤖