test_isdtype

[honno]

New territory here, but is currently passing for data-apis/array-api-compat#32

[asmeurer]

Not sure how I feel about nesting tests under if api_version like this instead of just skipping them.

[honno]

Yeah, doesn't feel good. I (seemingly) have to do this for a test like test_conj because we're using the >=2022.12 feature of Hypothesis xps.complex_dtypes(),

array-api-tests/array_api_tests/test_operators_and_elementwise_functions.py

Lines 927 to 934 in a533680

	if api_version >= "2022.12":
	@given(xps.arrays(dtype=xps.complex_dtypes(), shape=hh.shapes()))
	def test_conj(x):
	out = xp.conj(x)
	ph.assert_dtype("conj", x.dtype, out.dtype)
	ph.assert_shape("conj", out.shape, x.shape)
	unary_assert_against_refimpl("conj", x, out, operator.methodcaller("conjugate"))

so kinda just did this to be consistent with the pattern. But I think the ultimate goal with test_conj/etc. would be to not use this pattern, so now skipping this test instead in my force-push.

[asmeurer]

This implicitly assumes dtype equality works, meaning if it's broken or incorrect this might not notice. You can see the test I added in my compat PR compare the known value using dtype string names. Maybe it isn't so important here since we also test dtype.__eq__ anyway.

[honno]

Yeah. I think we're just out of luck if that's the case. Fortunately this would get checked via a sanity check I wrote in EqualityMapping

array-api-tests/array_api_tests/dtype_helpers.py

Lines 62 to 63 in a533680

	if not (key == key): # specifically checking __eq__, not __neq__
	raise ValueError(f"Key {key!r} does not have equality with itself")

[asmeurer]

If this passes my compat PR that sounds good to me.