[timohl]

Type

• Bug fix (non-breaking change which fixes an issue): Fixes Cannot create pointcloud from numpy array, "array is not writeable" #2557
• New feature (non-breaking change which adds functionality). Resolves #
• Breaking change (fix or feature that would cause existing functionality to not work as expected) Resolves #

Motivation and Context

A C-contiguous, correctly typed NumPy array with writeable=False currently raises ValueError: array is not writeable in an Eigen vector constructor, although the constructor only copies its input. For example:

import numpy as np
import open3d as o3d

points = np.array([[1., 2., 3.], [4., 5., 6.]], dtype=np.float64)
points.setflags(write=False)
vector = o3d.utility.Vector3dVector(points)
np.asarray(vector)[0] = [7., 8., 9.]
np.testing.assert_array_equal(points[0], [1., 2., 3.])
assert not points.flags.writeable

After this fix, construction succeeds and the vector remains independently owned and writable. The input values and readonly flag remain unchanged.

Checklist:

• I have run python util/check_style.py --apply to apply Open3D code style
  to my code.
• This PR changes Open3D behavior or adds new functionality.
  • Both C++ (Doxygen) and Python (Sphinx / Google style) documentation is
    updated accordingly.
  • I have added or updated C++ and / or Python unit tests OR included test
    results (e.g. screenshots or numbers) here.
• I will follow up and update the code if CI fails.
• For fork PRs, I have selected Allow edits from maintainers.

Description

Use unchecked<2>() and Eigen::Map<const EigenVector> in all four array-to-vector conversion helpers. Preserve dtype conversion, contiguity handling from #7343, shape validation and copying semantics. Correct the shared double-helper comment and document readonly input in the example.

Add 60 parameterized cases across five public vector constructors, writable/readonly inputs, empty and C/Fortran/strided/reversed arrays, and dtype conversion. Tests check values, dtype, input flags, independent storage and mutation isolation; existing invalid-shape tests remain in place. The allocator helper variants receive the same const change; the tests exercise the public constructors rather than adding test-only bindings.

Plan and scope: restrict the fix to the copying bindings, demonstrate the readonly failure, then validate the rebuilt bindings and available CPU suites. No algorithm or ownership changes.

Validation

Windows x64 Release, VS2022/MSVC 19.44, Python 3.14.3, NumPy 2.4.6, rebuilt Open3D 0.20.0+306764175. Verified the package and native extension load from the build output.

• Focused python/test/test_open3d_eigen.py: 99 passed, including all 60 new cases and the existing benchmark.
• pytest python/test/ --ignore=python/test/ml_ops --randomly-seed=20261022: 625 passed, 15 skipped. Excluded ML operations because those bindings were not built. Skips cover Linux-only rendering, SYCL/GPU, RealSense, unavailable ML visualization, four existing dataset skips, and optional msgpack.
• C++ gtest: 1,383 passed, 16 skipped, 0 failed (1,399 enabled tests); 200 existing disabled tests. Skips cover GPU/backend parity and the existing NYU loader test.
• The standalone example above passes. Before rebuilding, the existing unpatched binary produced 10 readonly failures in the focused tests (88 passed; benchmark deselected).

Build includes Python bindings, C++ tests, GUI, ISPC and IPP. CUDA/SYCL, TensorFlow/PyTorch ops, bundled ML, WebRTC/Jupyter and camera SDKs were disabled. Tests used eight logical CPUs. Other operating systems and GPU configurations were not validated.

[update-docs]

Thanks for submitting this pull request! The maintainers of this repository would appreciate if you could update the CHANGELOG.md based on your changes.