Missing coverage?

[jakirkham]

In the recent CodeCov report, they flag these lines as uncoverged

numcodecs/numcodecs/json.py

Lines 68 to 72 in ad1ad67

	def encode(self, buf):
	try:
	buf = np.asarray(buf)
	except ValueError:
	buf = np.asarray(buf, dtype=object)

It is a little unclear why as those lines were introduced with corresponding tests some time ago in PR ( #417 ). AFAICT recent NumPy versions still need to use that fallback. In fact that behavior is expected and defined in NEP 34

This may need further investigation to resolve

---

Additionally CodeCov flags this line as uncovered.

numcodecs/numcodecs/tests/test_zarr3.py

Lines 51 to 55 in ad1ad67

	@pytest.mark.parametrize("codec_class", ALL_CODECS)
	def test_docstring(codec_class: type[numcodecs.zarr3._NumcodecsCodec]):
	if codec_class.__doc__ is None:
	pytest.skip()
	assert "See :class:`numcodecs." in codec_class.__doc__

Just looking at it, would naively add a # pragma: no cover, but wanted to mention it here in case there is something I'm missing

[jakirkham]

Thanks David! 🙏

[jakirkham]

Reopening as the JSON case is still present

[jakirkham]

This is the test that would ideally cover the JSON case

numcodecs/numcodecs/tests/test_json.py

Lines 61 to 85 in 88660de

	@pytest.mark.parametrize(
	('input_data', 'dtype'),
	[
	([0, 1], None),
	([[0, 1], [2, 3]], None),
	([[0], [1], [2, 3]], object),
	([[[0, 0]], [[1, 1]], [[2, 3]]], None),
	(["1"], None),
	(["11", "11"], None),
	(["11", "1", "1"], None),
	([{}], None),
	([{"key": "value"}, ["list", "of", "strings"]], object),
	([0], None),
	([{'hi': 0}], "object"),
	(["hi"], "object"),
	(0, None),
	],
	)
	def test_non_numpy_inputs(input_data, dtype):
	# numpy will infer a range of different shapes and dtypes for these inputs.
	# Make sure that round-tripping through encode preserves this.
	data = np.array(input_data, dtype=dtype)
	for codec in codecs:
	output_data = codec.decode(codec.encode(data))
	assert input_data == output_data.tolist()

However we coerce to a NumPy array, which bypasses the try...except... handling above

numcodecs/numcodecs/tests/test_json.py

Line 82 in 88660de

data = np.array(input_data, dtype=dtype)

So either that test needs to be adapted to trigger the failure or a new test is needed