You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Describe the bug
A number of cudf tests create numpy or cupy arrays with data that is out of bounds for the specified dtype. Here is one easy example where 453 is larger than the largest representable uint8 (2^8=256) and so the result overflows. We should not be doing this. Not only is it bad practice, but numpy 1.24 has started warning about this behavior, so it will eventually be removed. Note that this is an important issue for us because we treat warnings as errors in our test suite, so fixing this bug is effectively a blocker for updating to a newer numpy version.
Steps/Code to reproduce bug
Installing numpy 1.24 into a conda environment with numba 0.56 is not possible because numba constrains the numpy version, and at present cudf is constrained to numba<0.57. However, practical support was added in #13271, so a reasonable environment may be created by removing the numba and numpy dependencies (https://github.com/rapidsai/cudf/blob/branch-23.06/conda/environments/all_cuda-118_arch-x86_64.yaml#L50) and then pip installing numba>=0.57 and numpy==1.24.3 into that environment. Building cudf in that environment will reproduce the issue.
Expected behavior
Test suite should pass
The text was updated successfully, but these errors were encountered:
We introduced an upper bound to the numpy pinnings in #13289 in order to temporarily avoid this problem. However, we are working to resolve it ASAP and remove the pinning.
Describe the bug
A number of cudf tests create numpy or cupy arrays with data that is out of bounds for the specified dtype. Here is one easy example where 453 is larger than the largest representable uint8 (2^8=256) and so the result overflows. We should not be doing this. Not only is it bad practice, but numpy 1.24 has started warning about this behavior, so it will eventually be removed. Note that this is an important issue for us because we treat warnings as errors in our test suite, so fixing this bug is effectively a blocker for updating to a newer numpy version.
Steps/Code to reproduce bug
Installing numpy 1.24 into a conda environment with numba 0.56 is not possible because numba constrains the numpy version, and at present cudf is constrained to numba<0.57. However, practical support was added in #13271, so a reasonable environment may be created by removing the numba and numpy dependencies (https://github.com/rapidsai/cudf/blob/branch-23.06/conda/environments/all_cuda-118_arch-x86_64.yaml#L50) and then pip installing numba>=0.57 and numpy==1.24.3 into that environment. Building cudf in that environment will reproduce the issue.
Expected behavior
Test suite should pass
The text was updated successfully, but these errors were encountered: