MAINT: Remove uses of scalar aliases #14901
Conversation
| highm1_arr = <np.ndarray>np.empty_like(high_arr, dtype=np.{{nptype}}) | ||
| highm1_arr = <np.ndarray>np.empty_like(high_arr, dtype=np.{{otype}}) |
There was a problem hiding this comment.
A note: nptype is such that npy_{nptype} is valid in C, otype is such that np.{otype} is valid in python.
6638c81 to
b7181a3
Compare
numpy/core/tests/test_ufunc.py
Outdated
| @@ -980,7 +980,7 @@ def test_can_ignore_signature(self): | |||
| assert_array_equal(out, mm_row_col_vec.squeeze()) | |||
|
|
|||
| def test_matrix_multiply(self): | |||
| self.compare_matrix_multiply_results(np.long) | |||
| self.compare_matrix_multiply_results(np.compat.long) | |||
There was a problem hiding this comment.
maybe a more specific dtype would be better than using a builtin?
There was a problem hiding this comment.
How about I make a follow-up that uses parametrize for this test?
|
LGTM. The test failure for b7181a3 is the win32 heisenbug. There are a few places where the use of a builtin for test parametrization next to specific dtypes looks out of place. Maybe they could all be specific dtypes rather than a mix of specific and builtins, but other than that this is a nice cleanup. |
b7181a3 to
6dfe331
Compare
|
A few of these tests really are testing the behavior of builtins. I fixed the two you called out. |
|
Thanks @eric-wieser |
Relates to gh-6103
Proof that this is exhaustive as of writing can be found in the CI of gh-14882, hopefully.
Even if it's not exhaustive, it's still good to merge something that gets us closer.
Split from gh-14882 so that cleanup can go in quickly, assuming a long discussion about deprecation.