scipy.io.savemat silently discards bad dictionary keys

[orena1]

I have a large (not complicated) dictionary that when I try to save it with sio.savemat I get 400 B file

Reproducing code example:

test.zip

import pickle
import scipy.io as sio
import os
data = pickle.load(open("test.p","r"))
sio.savemat('data',data)
os.path.getsize('data.mat')

Scipy/Numpy/Python version information:

('1.1.0', '1.14.2', sys.version_info(major=2, minor=7, micro=15, releaselevel='final', serial=0))

[matthew-brett]

To give more detail - the test pickle has a dictionary with values that are also dictionaries. These dictionaries have integers as keys. The scipy.io routines discard these, silently, because Matlab structs don't allow integers as fields:

>> struct(1, 'one')
Error using struct
Field names must be non-empty character vectors.

As a result, the thing that gets saved is:

In [23]: d2 = sio.loadmat('data.mat')

In [24]: d2
Out[24]: 
{'__globals__': [],
 '__header__': 'MATLAB 5.0 MAT-file Platform: posix, Created on: Wed May 22 10:57:44 2019',
 '__version__': '1.0',
 'v0': array([[None]], dtype=object),
 'v1': array([[None]], dtype=object),
 'v_post0': array([[None]], dtype=object),
 'v_post1': array([[None]], dtype=object)}

[erheron]

@matthew-brett any ideas for possible fixes here?
I'd be interested in tackling this, but no thoughts for now :|

[matthew-brett]

Nothing immediately springs to mind. Is there any container in Matlab that can deal with integer keys? I see that containers.Map will allow keys that aren't valid variable names, but I don't think it will allow integers.