Repository navigation
BUG: np.fromiter creates corrupted arrays when re-using StringDType instance #32013
Description
Describe the issue:
When re-using the StringDType instance, np.fromiter creates corrupted arrays leading to flaky failures.
Reproduce the code example:
Reproducer:
>>> np.__version__
'2.5.0'
>>> sd = np.dtypes.StringDType()
>>> a = ["a"*18, "b"*18]*10_000
>>> np.fromiter(iter(a), dtype=sd, count=len(a))
array(['aaaaaaaaaaaaaaaaaa', 'bbbbbbbbbbbbbbbbbb', 'aaaaaaaaaaaaaaaaaa',
..., 'bbbbbbbbbbbbbbbbbb', 'aaaaaaaaaaaaaaaaaa',
'bbbbbbbbbbbbbbbbbb'], shape=(20000,), dtype=StringDType())
>>> np.fromiter(iter(a), dtype=sd, count=len(a)) # same snippet ran for the second time segfaults.
Fatal Python error: Segmentation faultPython and NumPy Versions:
import sys, numpy; print(numpy.version); print(sys.version)
2.5.0
3.14.4
Activity
[-]BUG: np.fromiter creates corrupted arrays when re-using StringDType instance[/-][+]BUG: `np.fromiter` creates corrupted arrays when re-using StringDType instance[/+]