Can't read zero-size netCDF timedelta array

[deepers]

What happened?

When I write a zero-sized timedelta DataArray to a netCDF file, I can't read it back again.

What did you expect to happen?

I should be able to read back the array that was serialized to disk.

Minimal Complete Verifiable Example

import numpy as np
import xarray as xr
da = xr.DataArray(np.array([], dtype='timedelta64[ns]'))
da.to_netcdf('/tmp/foo.nc')
xr.load_dataarray('/tmp/foo.nc', decode_timedelta=True)

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

Traceback (most recent call last):
  File "/tmp/xarray-bug.py", line 5, in <module>
    xr.load_dataarray('/tmp/foo.nc', decode_timedelta=True)
    ~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/tmp/.venv/xarray-bug/lib64/python3.13/site-packages/xarray/backends/api.py", line 336, in load_dataarray
    return da.load()
           ~~~~~~~^^
  File "/tmp/.venv/xarray-bug/lib64/python3.13/site-packages/xarray/core/dataarray.py", line 1175, in load
    ds = self._to_temp_dataset().load(**kwargs)
  File "/tmp/.venv/xarray-bug/lib64/python3.13/site-packages/xarray/core/dataset.py", line 551, in load
    v.load()
    ~~~~~~^^
  File "/tmp/.venv/xarray-bug/lib64/python3.13/site-packages/xarray/core/variable.py", line 956, in load
    self._data = to_duck_array(self._data, **kwargs)
                 ~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^
  File "/tmp/.venv/xarray-bug/lib64/python3.13/site-packages/xarray/namedarray/pycompat.py", line 143, in to_duck_array
    return data.get_duck_array()  # type: ignore[no-untyped-call, no-any-return]
           ~~~~~~~~~~~~~~~~~~~^^
  File "/tmp/.venv/xarray-bug/lib64/python3.13/site-packages/xarray/core/indexing.py", line 837, in get_duck_array
    self._ensure_cached()
    ~~~~~~~~~~~~~~~~~~~^^
  File "/tmp/.venv/xarray-bug/lib64/python3.13/site-packages/xarray/core/indexing.py", line 834, in _ensure_cached
    self.array = as_indexable(self.array.get_duck_array())
                              ~~~~~~~~~~~~~~~~~~~~~~~~~^^
  File "/tmp/.venv/xarray-bug/lib64/python3.13/site-packages/xarray/core/indexing.py", line 791, in get_duck_array
    return self.array.get_duck_array()
           ~~~~~~~~~~~~~~~~~~~~~~~~~^^
  File "/tmp/.venv/xarray-bug/lib64/python3.13/site-packages/xarray/core/indexing.py", line 661, in get_duck_array
    array = array.get_duck_array()
  File "/tmp/.venv/xarray-bug/lib64/python3.13/site-packages/xarray/coding/common.py", line 76, in get_duck_array
    return self.func(self.array.get_duck_array())
           ~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/tmp/.venv/xarray-bug/lib64/python3.13/site-packages/xarray/coding/times.py", line 664, in decode_cf_timedelta
    _check_timedelta_range(np.nanmin(num_timedeltas), unit, time_unit)
                           ~~~~~~~~~^^^^^^^^^^^^^^^^
  File "/tmp/.venv/xarray-bug/lib64/python3.13/site-packages/numpy/lib/_nanfunctions_impl.py", line 357, in nanmin
    res = np.fmin.reduce(a, axis=axis, out=out, **kwargs)
ValueError: zero-size array to reduction operation fmin which has no identity

Anything else we need to know?

No response

Environment

INSTALLED VERSIONS

commit: None
python: 3.13.3 (main, Apr 22 2025, 00:00:00) [GCC 15.0.1 20250418 (Red Hat 15.0.1-0)]
python-bits: 64
OS: Linux
OS-release: 6.14.4-300.fc42.x86_64
machine: x86_64
processor:
byteorder: little
LC_ALL: None
LANG: en_CA.UTF-8
LOCALE: ('en_CA', 'UTF-8')
libhdf5: None
libnetcdf: None

xarray: 2025.4.0
pandas: 2.2.3
numpy: 2.2.5
scipy: 1.15.3
netCDF4: None
pydap: None
h5netcdf: None
h5py: None
zarr: None
cftime: None
nc_time_axis: None
iris: None
bottleneck: None
dask: None
distributed: None
matplotlib: None
cartopy: None
seaborn: None
numbagg: None
fsspec: None
cupy: None
pint: None
sparse: None
flox: None
numpy_groupies: None
setuptools: None
pip: 24.3.1
conda: None
pytest: None
mypy: None
IPython: 9.2.0
sphinx: None

[kmuehlbauer]

@deepers Thanks for raising. Could you please add the output of ncdump /tmp/foo.nc?

[deepers]

netcdf foo {
dimensions:
	dim_0 = UNLIMITED ; // (0 currently)
variables:
	int __xarray_dataarray_variable__(dim_0) ;
		__xarray_dataarray_variable__:units = "days" ;
data:
}

[kmuehlbauer]

@deepers Thanks! Does it trigger the same error if you do an xr.open_dataset(). It does look like a bug to me because it checks for data inside an empty array. We should take care of that and probably check for empty array.

[deepers]

@kmuehlbauer It actually works with xr.open_dataarray() and xr.open_dataset().

xr.open_dataarray('/tmp/foo.nc', decode_timedelta=True)

Output

<xarray.DataArray (dim_0: 0)> Size: 0B
[0 values with dtype=timedelta64[ns]]
Dimensions without coordinates: dim_0

[deepers]

@kmuehlbauer Actually, I think the above only works because of the lazy loading.

xr.open_dataarray('/tmp/foo.nc', decode_timedelta=True).values

has output

---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
Cell In[11], line 1
----> 1 xr.open_dataarray('/tmp/foo.nc', decode_timedelta=True).values

File ~/.venv/xarray-bug/lib64/python3.13/site-packages/xarray/core/dataarray.py:815, in DataArray.values(self)
    802 @property
    803 def values(self) -> np.ndarray:
    804     """
    805     The array's data converted to numpy.ndarray.
    806 
   (...)    813     to this array may be reflected in the DataArray as well.
    814     """
--> 815     return self.variable.values

File ~/.venv/xarray-bug/lib64/python3.13/site-packages/xarray/core/variable.py:516, in Variable.values(self)
    513 @property
    514 def values(self) -> np.ndarray:
    515     """The variable's data as a numpy.ndarray"""
--> 516     return _as_array_or_item(self._data)

File ~/.venv/xarray-bug/lib64/python3.13/site-packages/xarray/core/variable.py:302, in _as_array_or_item(data)
    288 def _as_array_or_item(data):
    289     """Return the given values as a numpy array, or as an individual item if
    290     it's a 0d datetime64 or timedelta64 array.
    291 
   (...)    300     TODO: remove this (replace with np.asarray) once these issues are fixed
    301     """
--> 302     data = np.asarray(data)
    303     if data.ndim == 0:
    304         kind = data.dtype.kind

File ~/.venv/xarray-bug/lib64/python3.13/site-packages/xarray/core/indexing.py:511, in ExplicitlyIndexed.__array__(self, dtype, copy)
    506 def __array__(
    507     self, dtype: np.typing.DTypeLike = None, /, *, copy: bool | None = None
    508 ) -> np.ndarray:
    509     # Leave casting to an array up to the underlying array type.
    510     if Version(np.__version__) >= Version("2.0.0"):
--> 511         return np.asarray(self.get_duck_array(), dtype=dtype, copy=copy)
    512     else:
    513         return np.asarray(self.get_duck_array(), dtype=dtype)

File ~/.venv/xarray-bug/lib64/python3.13/site-packages/xarray/core/indexing.py:837, in MemoryCachedArray.get_duck_array(self)
    836 def get_duck_array(self):
--> 837     self._ensure_cached()
    838     return self.array.get_duck_array()

File ~/.venv/xarray-bug/lib64/python3.13/site-packages/xarray/core/indexing.py:834, in MemoryCachedArray._ensure_cached(self)
    833 def _ensure_cached(self):
--> 834     self.array = as_indexable(self.array.get_duck_array())

File ~/.venv/xarray-bug/lib64/python3.13/site-packages/xarray/core/indexing.py:791, in CopyOnWriteArray.get_duck_array(self)
    790 def get_duck_array(self):
--> 791     return self.array.get_duck_array()

File ~/.venv/xarray-bug/lib64/python3.13/site-packages/xarray/core/indexing.py:661, in LazilyIndexedArray.get_duck_array(self)
    656 # self.array[self.key] is now a numpy array when
    657 # self.array is a BackendArray subclass
    658 # and self.key is BasicIndexer((slice(None, None, None),))
    659 # so we need the explicit check for ExplicitlyIndexed
    660 if isinstance(array, ExplicitlyIndexed):
--> 661     array = array.get_duck_array()
    662 return _wrap_numpy_scalars(array)

File ~/.venv/xarray-bug/lib64/python3.13/site-packages/xarray/coding/common.py:76, in _ElementwiseFunctionArray.get_duck_array(self)
     75 def get_duck_array(self):
---> 76     return self.func(self.array.get_duck_array())

File ~/.venv/xarray-bug/lib64/python3.13/site-packages/xarray/coding/times.py:664, in decode_cf_timedelta(num_timedeltas, units, time_unit)
    662 with warnings.catch_warnings():
    663     warnings.filterwarnings("ignore", "All-NaN slice encountered", RuntimeWarning)
--> 664     _check_timedelta_range(np.nanmin(num_timedeltas), unit, time_unit)
    665     _check_timedelta_range(np.nanmax(num_timedeltas), unit, time_unit)
    667 timedeltas = _numbers_to_timedelta(
    668     num_timedeltas, unit, "s", "timedeltas", target_unit=time_unit
    669 )

File ~/.venv/xarray-bug/lib64/python3.13/site-packages/numpy/lib/_nanfunctions_impl.py:357, in nanmin(a, axis, out, keepdims, initial, where)
    352     kwargs['where'] = where
    354 if type(a) is np.ndarray and a.dtype != np.object_:
    355     # Fast, but not safe for subclasses of ndarray, or object arrays,
    356     # which do not implement isnan (gh-9009), or fmin correctly (gh-8975)
--> 357     res = np.fmin.reduce(a, axis=axis, out=out, **kwargs)
    358     if np.isnan(res).any():
    359         warnings.warn("All-NaN slice encountered", RuntimeWarning,
    360                       stacklevel=2)

ValueError: zero-size array to reduction operation fmin which has no identity

[kmuehlbauer]

@deepers Thanks for testing. The issue is present in both directions, encoding and decoding. I've a possible fix in #10313. Would you mind checking your workflow against this PR?

[deepers]

@kmuehlbauer My original example runs fine with your fix, but a slightly different case fails with a different error. Note the shape of the array below.

It seems like this particular error is specific to the SciPy netCDF backend. If I install netCDF4, then this runs fine.

import numpy as np
import xarray as xr

da = xr.DataArray(np.array([[]], dtype='timedelta64[ns]'))
da.to_netcdf('/tmp/foo.nc')
xr.load_dataarray('/tmp/foo.nc', decode_timedelta=True)

Output

---------------------------------------------------------------------------
KeyError                                  Traceback (most recent call last)
File ~/.venv/xarray-bug-fix/lib64/python3.13/site-packages/xarray/backends/file_manager.py:211, in CachingFileManager._acquire_with_cache_info(self, needs_lock)
    210 try:
--> 211     file = self._cache[self._key]
    212 except KeyError:

File ~/.venv/xarray-bug-fix/lib64/python3.13/site-packages/xarray/backends/lru_cache.py:56, in LRUCache.__getitem__(self, key)
     55 with self._lock:
---> 56     value = self._cache[key]
     57     self._cache.move_to_end(key)

KeyError: [<function _open_scipy_netcdf at 0x7f133492a7a0>, ('/tmp/foo.nc',), 'r', (('mmap', None), ('version', 2)), 'bc5300d8-acbf-4f4e-bc0d-a878d7accb71']

During handling of the above exception, another exception occurred:

TypeError                                 Traceback (most recent call last)
File ~/xarray-bug.py:6
      4 da = xr.DataArray(np.array([[]], dtype='timedelta64[ns]'))
      5 da.to_netcdf('/tmp/foo.nc')
----> 6 xr.load_dataarray('/tmp/foo.nc', decode_timedelta=True)

File ~/.venv/xarray-bug-fix/lib64/python3.13/site-packages/xarray/backends/api.py:335, in load_dataarray(filename_or_obj, **kwargs)
    332 if "cache" in kwargs:
    333     raise TypeError("cache has no effect in this context")
--> 335 with open_dataarray(filename_or_obj, **kwargs) as da:
    336     return da.load()

File ~/.venv/xarray-bug-fix/lib64/python3.13/site-packages/xarray/backends/api.py:881, in open_dataarray(filename_or_obj, engine, chunks, cache, decode_cf, mask_and_scale, decode_times, decode_timedelta, use_cftime, concat_characters, decode_coords, drop_variables, inline_array, chunked_array_type, from_array_kwargs, backend_kwargs, **kwargs)
    710 def open_dataarray(
    711     filename_or_obj: str | os.PathLike[Any] | ReadBuffer | AbstractDataStore,
    712     *,
   (...)    731     **kwargs,
    732 ) -> DataArray:
    733     """Open an DataArray from a file or file-like object containing a single
    734     data variable.
    735 
   (...)    878     open_dataset
    879     """
--> 881     dataset = open_dataset(
    882         filename_or_obj,
    883         decode_cf=decode_cf,
    884         mask_and_scale=mask_and_scale,
    885         decode_times=decode_times,
    886         concat_characters=concat_characters,
    887         decode_coords=decode_coords,
    888         engine=engine,
    889         chunks=chunks,
    890         cache=cache,
    891         drop_variables=drop_variables,
    892         inline_array=inline_array,
    893         chunked_array_type=chunked_array_type,
    894         from_array_kwargs=from_array_kwargs,
    895         backend_kwargs=backend_kwargs,
    896         use_cftime=use_cftime,
    897         decode_timedelta=decode_timedelta,
    898         **kwargs,
    899     )
    901     if len(dataset.data_vars) != 1:
    902         if len(dataset.data_vars) == 0:

File ~/.venv/xarray-bug-fix/lib64/python3.13/site-packages/xarray/backends/api.py:687, in open_dataset(filename_or_obj, engine, chunks, cache, decode_cf, mask_and_scale, decode_times, decode_timedelta, use_cftime, concat_characters, decode_coords, drop_variables, inline_array, chunked_array_type, from_array_kwargs, backend_kwargs, **kwargs)
    675 decoders = _resolve_decoders_kwargs(
    676     decode_cf,
    677     open_backend_dataset_parameters=backend.open_dataset_parameters,
   (...)    683     decode_coords=decode_coords,
    684 )
    686 overwrite_encoded_chunks = kwargs.pop("overwrite_encoded_chunks", None)
--> 687 backend_ds = backend.open_dataset(
    688     filename_or_obj,
    689     drop_variables=drop_variables,
    690     **decoders,
    691     **kwargs,
    692 )
    693 ds = _dataset_from_backend_dataset(
    694     backend_ds,
    695     filename_or_obj,
   (...)    705     **kwargs,
    706 )
    707 return ds

File ~/.venv/xarray-bug-fix/lib64/python3.13/site-packages/xarray/backends/scipy_.py:333, in ScipyBackendEntrypoint.open_dataset(self, filename_or_obj, mask_and_scale, decode_times, concat_characters, decode_coords, drop_variables, use_cftime, decode_timedelta, mode, format, group, mmap, lock)
    331 store_entrypoint = StoreBackendEntrypoint()
    332 with close_on_error(store):
--> 333     ds = store_entrypoint.open_dataset(
    334         store,
    335         mask_and_scale=mask_and_scale,
    336         decode_times=decode_times,
    337         concat_characters=concat_characters,
    338         decode_coords=decode_coords,
    339         drop_variables=drop_variables,
    340         use_cftime=use_cftime,
    341         decode_timedelta=decode_timedelta,
    342     )
    343 return ds

File ~/.venv/xarray-bug-fix/lib64/python3.13/site-packages/xarray/backends/store.py:44, in StoreBackendEntrypoint.open_dataset(self, filename_or_obj, mask_and_scale, decode_times, concat_characters, decode_coords, drop_variables, use_cftime, decode_timedelta)
     30 def open_dataset(
     31     self,
     32     filename_or_obj: str | os.PathLike[Any] | ReadBuffer | AbstractDataStore,
   (...)     40     decode_timedelta=None,
     41 ) -> Dataset:
     42     assert isinstance(filename_or_obj, AbstractDataStore)
---> 44     vars, attrs = filename_or_obj.load()
     45     encoding = filename_or_obj.get_encoding()
     47     vars, attrs, coord_names = conventions.decode_cf_variables(
     48         vars,
     49         attrs,
   (...)     56         decode_timedelta=decode_timedelta,
     57     )

File ~/.venv/xarray-bug-fix/lib64/python3.13/site-packages/xarray/backends/common.py:312, in AbstractDataStore.load(self)
    293 def load(self):
    294     """
    295     This loads the variables and attributes simultaneously.
    296     A centralized loading function makes it easier to create
   (...)    309     are requested, so care should be taken to make sure its fast.
    310     """
    311     variables = FrozenDict(
--> 312         (_decode_variable_name(k), v) for k, v in self.get_variables().items()
    313     )
    314     attributes = FrozenDict(self.get_attrs())
    315     return variables, attributes

File ~/.venv/xarray-bug-fix/lib64/python3.13/site-packages/xarray/backends/scipy_.py:199, in ScipyDataStore.get_variables(self)
    197 def get_variables(self):
    198     return FrozenDict(
--> 199         (k, self.open_store_variable(k, v)) for k, v in self.ds.variables.items()
    200     )

File ~/.venv/xarray-bug-fix/lib64/python3.13/site-packages/xarray/backends/scipy_.py:188, in ScipyDataStore.ds(self)
    186 @property
    187 def ds(self):
--> 188     return self._manager.acquire()

File ~/.venv/xarray-bug-fix/lib64/python3.13/site-packages/xarray/backends/file_manager.py:193, in CachingFileManager.acquire(self, needs_lock)
    178 def acquire(self, needs_lock=True):
    179     """Acquire a file object from the manager.
    180 
    181     A new file is only opened if it has expired from the
   (...)    191         An open file object, as returned by ``opener(*args, **kwargs)``.
    192     """
--> 193     file, _ = self._acquire_with_cache_info(needs_lock)
    194     return file

File ~/.venv/xarray-bug-fix/lib64/python3.13/site-packages/xarray/backends/file_manager.py:217, in CachingFileManager._acquire_with_cache_info(self, needs_lock)
    215     kwargs = kwargs.copy()
    216     kwargs["mode"] = self._mode
--> 217 file = self._opener(*self._args, **kwargs)
    218 if self._mode == "w":
    219     # ensure file doesn't get overridden when opened again
    220     self._mode = "a"

File ~/.venv/xarray-bug-fix/lib64/python3.13/site-packages/xarray/backends/scipy_.py:127, in _open_scipy_netcdf(filename, mode, mmap, version)
    124     filename = io.BytesIO(filename)
    126 try:
--> 127     return scipy.io.netcdf_file(filename, mode=mode, mmap=mmap, version=version)
    128 except TypeError as e:  # netcdf3 message is obscure in this case
    129     errmsg = e.args[0]

File ~/.venv/xarray-bug-fix/lib64/python3.13/site-packages/scipy/io/_netcdf.py:279, in netcdf_file.__init__(self, filename, mode, mmap, version, maskandscale)
    276 self._attributes = {}
    278 if mode in 'ra':
--> 279     self._read()

File ~/.venv/xarray-bug-fix/lib64/python3.13/site-packages/scipy/io/_netcdf.py:611, in netcdf_file._read(self)
    609 self._read_dim_array()
    610 self._read_gatt_array()
--> 611 self._read_var_array()

File ~/.venv/xarray-bug-fix/lib64/python3.13/site-packages/scipy/io/_netcdf.py:692, in netcdf_file._read_var_array(self)
    689     data = None
    690 else:  # not a record variable
    691     # Calculate size to avoid problems with vsize (above)
--> 692     a_size = reduce(mul, shape, 1) * size
    693     if self.use_mmap:
    694         data = self._mm_buf[begin_:begin_+a_size].view(dtype=dtype_)

TypeError: unsupported operand type(s) for *: 'int' and 'NoneType'

[kmuehlbauer]

It seems like this particular error is specific to the SciPy netCDF backend. If I install netCDF4, then this runs fine.

Please open a new issue for that. That's a problem with scipy and netCDF3 Classic.

[deepers]

Thanks. For the record, the following tests now pass with the fix.

import numpy as np
import xarray as xr

da1 = xr.DataArray(np.array([], dtype='timedelta64[ns]'))
da2 = xr.DataArray(np.array([[]], dtype='timedelta64[ns]'))
da3 = xr.DataArray(np.empty((0, 0), dtype='timedelta64[ns]'))

filename = '/tmp/foo.nc'

for da in da1, da2, da3:
    for engine in 'netcdf4', 'h5netcdf':
        da.to_netcdf(filename, engine=engine)
        xr.load_dataarray(filename, decode_timedelta=True)
        xr.open_dataarray(filename, decode_timedelta=True).values