v0.1.18 loadmat() not loading python float/ints from saved mat-file under NumPy 1.23

[jeremypriest]

Note: I wrote all of the below out, and then realized that it appears this issue has been fixed on the main branch. A release of the main branch would be much appreciated.

hdf5storage 0.1.18 is affected by some sort of change in NumPy 1.23. If a python float or int is saved out, it will not be read back in during loadmat(). When it is saved out as a np.float32/64 or np.int32/64, the variable is loaded in properly.

Here is the code example I'm using:

from hdf5storage import loadmat, savemat
import numpy as np

float1 = float(1)
float2 = np.float64(1)

file_mat = 'test.mat'
out_mat = {'float': float1,
           'np_float64': float2}
print(out_mat.keys())

savemat(file_mat, out_mat)

in_mat = loadmat(file_mat)
print(in_mat.keys())

print(np.__version__)

Output under NumPy 1.23.3:

dict_keys(['float', 'np_float64'])
dict_keys(['np_float64'])
1.23.3

Output under NumPy 1.22.4:

dict_keys(['float', 'np_float64'])
dict_keys(['float', 'np_float64'])
1.22.4

Output under NumPy 1.23.3 and hdf5storage main branch (commit ec2e1e34789ad27713b860eb4de5da99817fa315):

dict_keys(['float', 'np_float64'])
dict_keys(['float', 'np_float64'])
1.23.3

[frejanordsiek]

This is fixed in the a06c749b970d012256311ba0b95321ac83cae0ec commit on the 0.1.x branch. It is caused by the numpy.asscalar function being deprecated, which prevented reading the value (it was saved, though). I will be pushing a 0.1.19 bugfix release to PyPI shortly.