diff options
| author | Sebastian Berg <sebastianb@nvidia.com> | 2023-02-25 14:02:10 +0100 |
|---|---|---|
| committer | Sebastian Berg <sebastianb@nvidia.com> | 2023-02-25 14:27:29 +0100 |
| commit | 07f603d2585e155f5642d665ab75173501d71afc (patch) | |
| tree | b811cbfab01c00acd46cb3d026c1a11b95151be3 /numpy/array_api/tests/test_validation.py | |
| parent | 5f04e748b8d4c92bc6ad4d7a38ee6601dc0ccf04 (diff) | |
| download | numpy-07f603d2585e155f5642d665ab75173501d71afc.tar.gz | |
MAINT: Use strong references/copies for sorting buffer
I accidentally included a start for a cleanup in the other PR which
was incorrect because we actually sorted weak references.
Sorting weak references buffer is correct, but seems a bit trickier
to reason about and also potentially to generalize.
This makes sure we have strong references everywhere and fixes the
issue seen by pandas. Also adds a (slightly complex) test to cover
both the sort and argsort path.
Diffstat (limited to 'numpy/array_api/tests/test_validation.py')
0 files changed, 0 insertions, 0 deletions