Description
What happened:
If I create a Dataset with an attr that is a list of strings with only a single member, write it to netcdf and open it again, the attr is a single string instead of a list of strings.
What you expected to happen:
Writing the Dataset to netcdf and opening it should not change the attrs.
Minimal Complete Verifiable Example:
import xarray as xr
ds = xr.Dataset(attrs={"foo": ["bar"]},)
ds.to_netcdf("ds.nc", engine="netcdf4")
rd = xr.open_dataset("ds.nc", engine="netcdf4")
assert ds.attrs["foo"] == rd.attrs["foo"]
Anything else we need to know?:
This was implemented in PR2045, including the test test_setncattr_string. However, there assert_array_equal
is used to compare the attrs after a roundtrip (line 1454 ), which unfortunately compares element-wise if one of the elements is a scalar. Changing the
test like this:
- assert_array_equal(one_element_list_of_strings, totest.attrs["bar"])
+ assert one_element_list_of_strings == totest.attrs["bar"]
corrects it (the test then fails currently).
Environment:
Output of xr.show_versions()
INSTALLED VERSIONS
commit: f52a95c
python: 3.8.6 (default, Sep 25 2020, 09:36:53)
[GCC 10.2.0]
python-bits: 64
OS: Linux
OS-release: 5.8.0-36-generic
machine: x86_64
processor: x86_64
byteorder: little
LC_ALL: None
LANG: en_US.UTF-8
LOCALE: en_US.UTF-8
libhdf5: 1.12.0
libnetcdf: 4.7.4
xarray: 0.16.3.dev75+gf52a95cb
pandas: 1.2.0
numpy: 1.19.5
scipy: None
netCDF4: 1.5.5.1
pydap: None
h5netcdf: None
h5py: None
Nio: None
zarr: None
cftime: 1.3.0
nc_time_axis: None
PseudoNetCDF: None
rasterio: None
cfgrib: None
iris: None
bottleneck: None
dask: 2020.12.0
distributed: 2020.12.0
matplotlib: None
cartopy: None
seaborn: None
numbagg: None
pint: None
setuptools: 51.1.2
pip: 20.3.3
conda: None
pytest: 6.2.1
IPython: None
sphinx: None