`np.bytes_` scalar datasets in NetCDF4 are converted to arrays of bytes

[scottstanie]

What happened?

I've made NetCDF5 files using h5netcdf where some datasets are strings, which get encoded as np.bytes_. This means they show as {SCALAR} when viewed with h5ls.

If I load and save again using engine='h5netcdf', they become arrays of single bytes. I don't see a way to save the same data type as with h5netcdf.

What did you expect to happen?

I expected to be able to save the HDF5 file with to_netcdf(engine='h5netcdf') and get the same result as saving with h5netcdf directly.
Perhaps this is expecting to have an option to load without running np.asarray on the np.bytes_ object.

Minimal Complete Verifiable Example

import xarray as xr
import numpy as np
import h5netcdf

with h5netcdf.File("test-np-bytes.nc", "w") as hf:
    hf.create_variable(name="data", data=np.bytes_("test this string"))

with h5netcdf.File("test-np-bytes.nc") as hf:
    print("Data as originally loaded by h5netcdf")
    data = hf["data"][()]
    print(data.dtype)
    print(data)
    print(repr(data))
    print()

with xr.open_dataset("test-np-bytes.nc", engine="h5netcdf") as ds:
    print("Data as loaded by xarray")
    data = ds.data.values
    print(data.dtype)
    print(data)
    print(repr(data))
    print()

with xr.open_dataset("test-np-bytes.nc", engine="h5netcdf") as ds:
    print("Running to_netcdf...")
    ds.to_netcdf("rewritten.nc", engine="h5netcdf")

with h5netcdf.File("rewritten.nc") as hf:
    print()
    print("New file, loaded by h5netcdf")
    data = hf["data"][()]
    print(data.dtype)
    print(data)
    print(repr(data))

MVCE confirmation

• Minimal example — the example is as focused as reasonably possible to demonstrate the underlying issue in xarray.
• Complete example — the example is self-contained, including all data and the text of any traceback.
• Verifiable example — the example copy & pastes into an IPython prompt or Binder notebook, returning the result.
• New issue — a search of GitHub Issues suggests this is not a duplicate.
• Recent environment — the issue occurs with the latest version of xarray and its dependencies.

Relevant log output

Data as originally loaded by h5netcdf
|S16
b'test this string'
np.bytes_(b'test this string')

Data as loaded by xarray
|S16
np.bytes_(b'test this string')
array(b'test this string', dtype='|S16')

Running to_netcdf...

New file, loaded by h5netcdf
|S1
[b't' b'e' b's' b't' b' ' b't' b'h' b'i' b's' b' ' b's' b't' b'r' b'i'
 b'n' b'g']
array(, dtype='|S1')

Anything else we need to know?

No response

Environment

INSTALLED VERSIONS

commit: None
python: 3.12.8 | packaged by conda-forge | (main, Dec 5 2024, 14:19:53) [Clang 18.1.8 ]
python-bits: 64
OS: Darwin
OS-release: 24.5.0
machine: arm64
processor: arm
byteorder: little
LC_ALL: None
LANG: en_US.UTF-8
LOCALE: ('en_US', 'UTF-8')
libhdf5: 1.14.3
libnetcdf: 4.9.2

xarray: 2025.4.0
pandas: 2.2.3
numpy: 2.0.2
scipy: 1.15.1
netCDF4: 1.7.2
pydap: None
h5netcdf: 1.6.1
h5py: 3.12.1
zarr: 3.0.6
cftime: 1.6.4
nc_time_axis: None
iris: None
bottleneck: None
dask: 2025.4.1
distributed: 2025.4.1
matplotlib: 3.10.0
cartopy: 0.24.0
seaborn: 0.13.2
numbagg: None
fsspec: 2025.2.0
cupy: None
pint: 0.24.4
sparse: None
flox: None
numpy_groupies: None
setuptools: 75.8.0
pip: 24.3.1
conda: 25.1.1
pytest: 8.3.4
mypy: None
IPython: 8.17.2
sphinx: None

[kmuehlbauer]

Hi @scottstanie,

the dataset you've created with h5netcdf is of fixed size scalar string type. NetCDF4 doesn't know fixed length strings. It can only handle arrays of chars (S1) or vlen strings (str).

If opened with engine="h5netcdf" this will be imported as fixed length bytes:

<xarray.Dataset> Size: 16B
Dimensions:  ()
Data variables:
    data     |S16 16B b'test this string'

and with engine="netcdf4" as unicode string (by netcdf-c under the hood):

<xarray.Dataset> Size: 64B
Dimensions:  ()
Data variables:
    data     <U16 64B 'test this string'

When writing the first back to NetCDF4 file (no matter which of the two engines is used) the fixed length string is written as an array of characters with a new dimension the size of the string length.

When writing the second back ( again no matter which of the two engines is used) the unicode string is written as vlen string.

I think the main problem is that there is one way to create a fixed length string dataset in h5netcdf using the data keyword with the new API. You won't be able to do this using either h5netcdf new or legacy API as advertised in the docs. The second issue is, that xarray just returns the fixed length string given by h5netcdf/h5py.

Both issues could be handled in h5netcdf, if this is really necessary.