Writing np.bool8 data array reads in as int8

[shaunc]

What happened:

I have an dataarray with dtype np.bool_. When I write it using netcdf (engine h5netcdf, or default) and then read in a copy, the copy has dtype int8.

What you expected to happen:

The loaded data array should have dtype bool

Minimal Complete Verifiable Example:

I have had a hard time reducing this to a sample. The data array comes from a larger dataset which exhibits the same problem. I can copy the dataarray using copy() and it still exhibits the problem; however if I build a new data array using the constructor, the new array doesn't exhibit the problem. As far as I can tell, though, the original and the rebuilt dataarray are otherwise identical.

# in a pdb session
(Pdb) ci
<xarray.DataArray 'cut_inclusive' (cut: 15)>
array([False, False, False, False, False, False, False, False, False,
       False, False, False, False, False, False])
Dimensions without coordinates: cut
(Pdb) ci.to_netcdf('foo_ci.nca', engine="h5netcdf")
(Pdb) csi = xr.read_dataarray('foo_ci_nca', engine="h5netcdf"); csi
<xarray.DataArray 'cut_inclusive' (cut: 15)>
array([0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0], dtype=int8)
Dimensions without coordinates: cut
(Pdb) ci2 = xr.DataArray(ci, dims=('cut', ))
(Pdb) ci2.equals(ci)
True
(Pdb) ci2.to_netcdf('foo_ci2.nca', engine="h5netcdf")
(Pdb) csi2 = xr.open_dataarray('foo_ci2.nca', engine="h5netcdf"); csi2
<xarray.DataArray 'cut_inclusive' (cut: 15)>
array([False, False, False, False, False, False, False, False, False, False,
       False, False, False, False, False])
Dimensions without coordinates: cut
(Pdb) ci3 = ci.copy()
(Pdb) ci3.to_netcdf('foo_ci3.nca', engine="h5netcdf")
(Pdb) csi3 = xr.open_dataarray('foo_ci3.nca', engine="h5netcdf"); csi3
<xarray.DataArray 'cut_inclusive' (cut: 15)>
array([0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0], dtype=int8)
Dimensions without coordinates: cut

Anything else we need to know?:

I am at a loss how to investigate why ci and ci3 don't survive round-trip, but ci2 does. Unfortunately, I also have been unable to produce a free-standing example -- whenever I try I get an object that survives round trip intact. I suspect that xarray internals is somewhere/somehow keeping a cache to the original ci (presumably still linked to the overall dataset from which ci came), and this is what is causing the problem, but I don't know where to look. (Suggestions welcome!)

Environment:

Output of xr.show_versions()

xr.show_versions()

INSTALLED VERSIONS

commit: None
python: 3.8.5 (default, Sep 4 2020, 07:30:14)
[GCC 7.3.0]
python-bits: 64
OS: Linux
OS-release: 5.8.0-48-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: None

xarray: 0.17.0
pandas: 1.2.4
numpy: 1.20.2
scipy: 1.6.2
netCDF4: None
pydap: None
h5netcdf: 0.10.0
h5py: 3.2.1
Nio: None
zarr: None
cftime: None
nc_time_axis: None
PseudoNetCDF: None
rasterio: None
cfgrib: None
iris: None
bottleneck: None
dask: 2021.04.0
distributed: 2021.04.0
matplotlib: 3.4.1
cartopy: None
seaborn: None
numbagg: None
pint: None
setuptools: 51.0.0
pip: 20.3.1
conda: None
pytest: 6.2.3
IPython: 7.22.0
sphinx: None

[andersy005]

This is related to #4826

[shaunc]

Aha -- I think that is exactly it. I'll close in favor of that (just noting though that h5netcdf also has the problem).