Inserting data into variable with unlimited dimension with slice(n, n+1, None) is broken

[cpaulik]

Another side effect of #906 it seems

import netCDF4
import numpy as np

with netCDF4.Dataset('test.nc','w') as f:
    f.createDimension('d1',3)
    f.createDimension('d2',None)

    f.createVariable('v1',np.float,('d1','d1',))
    f['v1'][0:1] = np.arange(3)

Running this on 1.5.1.1 gives

1.5.1.1
(0, 3)
Traceback (most recent call last):
  File "01-160948.py", line 23, in <module>
    f['v1'][0:1] = np.arange(3)
  File "netCDF4/_netCDF4.pyx", line 4796, in netCDF4._netCDF4.Variable.__setitem__
  File "/home/cpa/.pyenv/versions/miniconda2-4.3.30/envs/pytesmo/lib/python3.7/site-packages/netCDF4/utils.py", line 374, in _StartCountStride
    datashapenew = datashapenew + (datashape[i],)
IndexError: tuple index out of range

on 1.5.0.1 it works and gives

1.5.0.1
(0, 3)
(1, 3)

The following is also broken so it is not only for the first element:

import netCDF4
import numpy as np

with netCDF4.Dataset('test.nc','w') as f:
    f.createDimension('d1',3)
    f.createDimension('d2',None)

    f.createVariable('v1',np.float,('d2','d1'))
    print(f['v1'].shape)
    f['v1'][0] = np.arange(3)
    print(f['v1'].shape)
    f['v1'][1:2] = np.arange(3)
    print(f['v1'].shape)

[jswhit]

This indexing stuff is a mess, and unlimited dimensions make it even worse. I wish there were someway to use the numpy indexing machinery for this instead of re-inventing our own.

The workaround for this is to use np.arange(3).reshape(1,3).

[jswhit]

BTW - is there a typo in your first example?

f.createVariable('v1',np.float,('d1','d1',)) <-- should this be ('d2','d1')?

[cpaulik]

Yes the first example was copied to soon.

[jswhit]

Fix in pull request #924. Keep trying to break it!

[cpaulik]

Thanks. The fix works for me.