Using numpy=2 I am seeing problems in test_image.py because of _addValue in image.py. Specifically, the usage arr[y, x] += value now apparently does strict type/bounds checking such that trying to add a negative value to an unsigned integer array is an OverflowError. (With numpy=1 arr += value would fail in the same way, but doing the specific index would not for ... reasons.).
A possible workaround is to force unsafe casting such that the RHS can be np.array(value).astype(self._array.dtype, casting="unsafe"). I don't know if this has performance implications or if somebody has a cleverer idea.
rmjarvis
commented
2 months ago
Using numpy=2 I am seeing problems in
test_image.pybecause of_addValueinimage.py. Specifically, the usagearr[y, x] += valuenow apparently does strict type/bounds checking such that trying to add a negative value to an unsigned integer array is anOverflowError. (With numpy=1arr += valuewould fail in the same way, but doing the specific index would not for ... reasons.).A possible workaround is to force unsafe casting such that the RHS can be
np.array(value).astype(self._array.dtype, casting="unsafe"). I don't know if this has performance implications or if somebody has a cleverer idea.