diff of cftime.Datetime

[mathause]

What happened?

A cftime variable returns a timedelta64[ns] when calling diff / + / - and it can then not be added/ subtracted from the original data.

What did you expect to happen?

We can add cftime timedeltas.

Minimal Complete Verifiable Example

import xarray as xr

air = xr.tutorial.open_dataset("air_temperature", use_cftime=True)

air.time + air.time.diff("time") / 2

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.

Relevant log output

air.time.variable.values[1:] - air.time.variable.values[:-1]

returns array([datetime.timedelta(seconds=21600), ...]) but then

xr.Variable(("time",), np.array([datetime.timedelta(0)]))

returns a dtype='timedelta64[ns]' array.

Anything else we need to know?

• See upstream PR: Add handling of cftime bounds xarray-contrib/cf-xarray#441
• Similar to Document cftime arithmetic limitations #7381 (but I don't think it's the same issue, feel free to close if you disagree)
• That might need a special data type for timedeltas of cftime.Datetime objects, or allowing to add 'timedelta64[ns]' to cftime.Datetime objects
• The casting comes from

xarray/xarray/core/variable.py

Line 366 in d8ec3a3

self._data = as_compatible_data(data, fastpath=fastpath)

xarray/xarray/core/variable.py

Line 272 in d8ec3a3

if isinstance(data, timedelta):

Environment

INSTALLED VERSIONS

commit: d8ec3a3
python: 3.10.9 | packaged by conda-forge | (main, Feb 2 2023, 20:20:04) [GCC 11.3.0]
python-bits: 64
OS: Linux
OS-release: 5.14.21-150400.24.63-default
machine: x86_64
processor: x86_64
byteorder: little
LC_ALL: None
LANG: en_GB.UTF-8
LOCALE: ('en_US', 'UTF-8')
libhdf5: 1.12.2
libnetcdf: 4.9.1

xarray: 2023.2.1.dev20+g06a87062
pandas: 1.5.3
numpy: 1.23.5
scipy: 1.10.1
netCDF4: 1.6.2
pydap: installed
h5netcdf: 1.1.0
h5py: 3.8.0
Nio: None
zarr: 2.13.6
cftime: 1.6.2
nc_time_axis: 1.4.1
PseudoNetCDF: 3.2.2
iris: 3.4.1
bottleneck: 1.3.6
dask: 2023.2.1
distributed: 2023.2.1
matplotlib: 3.7.0
cartopy: 0.21.1
seaborn: 0.12.2
numbagg: 0.2.2
fsspec: 2023.1.0
cupy: None
pint: 0.20.1
sparse: 0.14.0
flox: 0.6.8
numpy_groupies: 0.9.20
setuptools: 67.4.0
pip: 23.0.1
conda: None
pytest: 7.2.1
mypy: None
IPython: 8.11.0
sphinx: None

[mathause]

cc @spencerkclark @znicholls

[znichollscr]

cc @spencerkclark @znicholls

Legend thanks

[larsbuntemeyer]

@mathause

That might need a special data type for timedeltas of cftime.Datetime objects, or allowing to add 'timedelta64[ns]' to cftime.Datetime objects

I think this might be related:

• ufuncs on object arrays should call __array_ufunc__ if it exists on the wrapped objects numpy/numpy#15479 (comment)