issue_comments

Just to correct something here, missing_value is no longer considered deprecated in the current version of the standard: http://cfconventions.org/Data/cf-conventions/cf-conventions-1.7/cf-conventions.html

Near the bottom is a specific note about removing the deprecation.

After rethinking the issue, I would drop it: one can simply pass ds.fromkeys(ds.data_vars.keys(), {}) as the encoding attribute. Going back to the original problem. The fix above is not enough, the SerializationWarning is still present. An alternative, provided that missing_value attribute is still considered deprecated: http://cfconventions.org/Data/cf-conventions/cf-conventions-1.1/build/cf-conventions.html#missing-data, would be to replace it by _FillValue on decoding: $ diff variables.py variables.py.orig 179,180d178 < if 'FillValue' not in encoding: < encoding['_FillValue'] = encoding.pop('missing_value')``

Indeed it works. Thanks. My quick fix:

$ diff variables.py variables.py.orig 152,155d151 < elif encoding.get('missing_value') is not None: < fill_value = pop_to(encoding, attrs, 'missing_value', name=name) < if not pd.isnull(fill_value): < data = duck_array_ops.fillna(data, fill_value) I also figured out how to write back floating point values: encoding=None means use existing values, so specifying encoding={'tmpk': {}} in to_netcdf() did the trick. Should there be an option for this? What you see on the screen is not what you get in the file.

Yes, this is a challenging design problem and the current state of affairs is definitely not ideal. If you have ideas about how to reveal the presence of an encoding dict associated with a variable without overwhelming the user with information those would be very welcome. The alternative would be to not keep around encoding at all, but I think users do appreciate that reading/writing a netCDF file generally returns the exact same thing.

Indeed it works. Thanks. My quick fix: $ diff variables.py variables.py.orig 152,155d151 < elif encoding.get('missing_value') is not None: < fill_value = pop_to(encoding, attrs, 'missing_value', name=name) < if not pd.isnull(fill_value): < data = duck_array_ops.fillna(data, fill_value) I also figured out how to write back floating point values: encoding=None means use existing values, so specifying encoding={'tmpk': {}} in to_netcdf() did the trick. Should there be an option for this? What you see on the screen is not what you get in the file.

I think this could be fixed by making CFMaskCoder handle missing_value and _FillValue in a symmetric way in encoding/decoding: https://github.com/pydata/xarray/blob/23d54a890e6cfe1de420071b597c911164de4cb8/xarray/coding/variables.py#L142

Right now, it only uses missing_value in the decode() method.