Testing the Python Library

Test Suite Overview

Python tests live in test/test_python_api.py and run with pytest. The test file exercises the public API exposed by the voxelpic package — encode, decode, image_shape, voxel_size, PointCloud, and DecodeResult — which delegate to the _voxelpic C extension module.

Installing and Running

pip install -e .[test]    # editable install with test dependencies
pytest -vv                # run all tests from repo root

The [test] extra installs pytest, pytest-md, pytest-emoji, and pytest-cov.

Useful Commands

pytest test/test_python_api.py -vv                 # run only Python tests
pytest test/test_python_api.py::test_transcode -vv # single test function
pytest test/test_python_api.py::test_transcode[9]  # single parametrized case
pytest --cov=voxelpic --cov-report=term-missing    # with coverage

Python API Surface

The C extension _voxelpic exposes four functions consumed by src/voxelpic.py:

Key Types

• PointCloud — NamedTuple with positions (Nx4 float32) and colors (Nx4 uint8). Has to3() / to4() conversion methods and save() / load() for binary I/O.
• DecodeResult — NamedTuple with positions, colors, count, depth. Has a .cloud property that returns a PointCloud.

Array Constraints (validated in the C extension)

• Positions: Nx4, dtype float32, C-contiguous.
• Colors: Nx4, dtype uint8, C-contiguous.
• Images: HxWx4, dtype uint8, C-contiguous.
• Position and color arrays must have the same N dimension.

Conventions for Writing a New Test

File and Imports

Add tests to test/test_python_api.py (or create a new test/test_*.py file).

import numpy as np
import voxelpic as vp
import pytest

Generating Test Data

Tests generate random data inline rather than reading from files:

positions = np.random.rand(1000, 4).astype(np.float32) * 2 - 1
positions[:, 3] = 1  # homogeneous w coordinate
colors = np.random.randint(0, 255, (1000, 4)).astype(np.uint8)
colors[:, 3] = 255   # alpha channel
cloud = vp.PointCloud(positions, colors)

Parametrization

Use @pytest.mark.parametrize for depth variation:

@pytest.mark.parametrize("depth", [7, 8, 9])
def test_something(depth):
    ...

Assertions

Use NumPy testing assertions for array comparisons:

np.testing.assert_array_equal(actual, expected)          # exact (uint8, int)
np.testing.assert_array_almost_equal(actual, expected)    # float tolerance

Use standard assert for scalar values:

assert result.depth == depth
assert image.shape == (height, width, 4)

Fixtures

Use pytest built-in fixtures. tmp_path provides a temporary directory per test, useful for I/O tests:

def test_io(tmp_path):
    path = str(tmp_path / "cloud.dat")
    cloud.save(path)
    loaded = vp.PointCloud.load(path)

No custom fixtures or conftest.py exist — tests are self-contained.

Round-Trip Testing Pattern

The primary pattern is encode → decode → re-encode, verifying that the second encode produces an identical image:

image = vp.encode(cloud, depth=depth)
result = vp.decode(image)
image2 = vp.encode(result.cloud, depth=depth)
np.testing.assert_array_equal(image, image2)

Testing Preallocated Buffers

Both encode and decode accept optional preallocated output arrays. Test both the default allocation path and the preallocated path:

# Default allocation
image = vp.encode(cloud, depth=depth)

# Preallocated output
h, w = vp.image_shape(depth)
out = np.empty((h, w, 4), dtype=np.uint8)
image2 = vp.encode(cloud, depth=depth, out_image=out)

Configuration

There is no pytest.ini, conftest.py, or [tool.pytest.ini_options] section. Pytest uses default discovery: files matching test_*.py with functions matching test_*.