issues

What happened?

This is my example, that doesn't work since v2023.09.0:

```python import numpy as np import pandas as pd import xarray as xr

time = pd.date_range("1970-01-01", "1970-01-31", freq="6h") ds = xr.Dataset(coords=dict(time=time))

units = "days since 1960-01-01 00:00:00" calendar = "gregorian" encoding = dict(time=dict(units=units, calendar=calendar, dtype=np.dtype("float64")))

ds.to_netcdf("test.nc", encoding=encoding)

! ncdump -v time test.nc ```

This gives the following output:

``` netcdf test { dimensions: time = 121 ; variables: double time(time) ; time:_FillValue = NaN ; time:units = "days since 1960-01-01" ; time:calendar = "gregorian" ; data:

time = 3653, 3653, 3653, 3653, 3654, 3654, 3654, 3654, 3655, 3655, 3655, 3655, 3656, 3656, 3656, 3656, 3657, 3657, 3657, 3657, 3658, 3658, 3658, 3658, 3659, 3659, 3659, 3659, 3660, 3660, 3660, 3660, 3661, 3661, 3661, 3661, 3662, 3662, 3662, 3662, 3663, 3663, 3663, 3663, 3664, 3664, 3664, 3664, 3665, 3665, 3665, 3665, 3666, 3666, 3666, 3666, 3667, 3667, 3667, 3667, 3668, 3668, 3668, 3668, 3669, 3669, 3669, 3669, 3670, 3670, 3670, 3670, 3671, 3671, 3671, 3671, 3672, 3672, 3672, 3672, 3673, 3673, 3673, 3673, 3674, 3674, 3674, 3674, 3675, 3675, 3675, 3675, 3676, 3676, 3676, 3676, 3677, 3677, 3677, 3677, 3678, 3678, 3678, 3678, 3679, 3679, 3679, 3679, 3680, 3680, 3680, 3680, 3681, 3681, 3681, 3681, 3682, 3682, 3682, 3682, 3683 ; } ```

It seems like the subdaily fraction is truncated. Note, that this does not happend, if i set the units to the start of the time range, e.g., units = "days since 1970-01-01 00:00:00". This results correctly in

``` netcdf test { dimensions: time = 121 ; variables: double time(time) ; time:_FillValue = NaN ; time:units = "days since 1970-01-01" ; time:calendar = "gregorian" ; data:

time = 0, 0.25, 0.5, 0.75, 1, 1.25, 1.5, 1.75, 2, 2.25, 2.5, 2.75, 3, 3.25, 3.5, 3.75, 4, 4.25, 4.5, 4.75, 5, 5.25, 5.5, 5.75, 6, 6.25, 6.5, 6.75, 7, 7.25, 7.5, 7.75, 8, 8.25, 8.5, 8.75, 9, 9.25, 9.5, 9.75, 10, 10.25, 10.5, 10.75, 11, 11.25, 11.5, 11.75, 12, 12.25, 12.5, 12.75, 13, 13.25, 13.5, 13.75, 14, 14.25, 14.5, 14.75, 15, 15.25, 15.5, 15.75, 16, 16.25, 16.5, 16.75, 17, 17.25, 17.5, 17.75, 18, 18.25, 18.5, 18.75, 19, 19.25, 19.5, 19.75, 20, 20.25, 20.5, 20.75, 21, 21.25, 21.5, 21.75, 22, 22.25, 22.5, 22.75, 23, 23.25, 23.5, 23.75, 24, 24.25, 24.5, 24.75, 25, 25.25, 25.5, 25.75, 26, 26.25, 26.5, 26.75, 27, 27.25, 27.5, 27.75, 28, 28.25, 28.5, 28.75, 29, 29.25, 29.5, 29.75, 30 ; } ```

What did you expect to happen?

I expect subdaily frequencies to be encoded correctly also if the units startdate is different from the startdate of the time axis, e.g. v2023.08.0 correctly preserves fractions:

``` netcdf test { dimensions: time = 121 ; variables: double time(time) ; time:_FillValue = NaN ; time:units = "days since 1960-01-01" ; time:calendar = "gregorian" ; data:

time = 3653, 3653.25, 3653.5, 3653.75, 3654, 3654.25, 3654.5, 3654.75, 3655, 3655.25, 3655.5, 3655.75, 3656, 3656.25, 3656.5, 3656.75, 3657, 3657.25, 3657.5, 3657.75, 3658, 3658.25, 3658.5, 3658.75, 3659, 3659.25, 3659.5, 3659.75, 3660, 3660.25, 3660.5, 3660.75, 3661, 3661.25, 3661.5, 3661.75, 3662, 3662.25, 3662.5, 3662.75, 3663, 3663.25, 3663.5, 3663.75, 3664, 3664.25, 3664.5, 3664.75, 3665, 3665.25, 3665.5, 3665.75, 3666, 3666.25, 3666.5, 3666.75, 3667, 3667.25, 3667.5, 3667.75, 3668, 3668.25, 3668.5, 3668.75, 3669, 3669.25, 3669.5, 3669.75, 3670, 3670.25, 3670.5, 3670.75, 3671, 3671.25, 3671.5, 3671.75, 3672, 3672.25, 3672.5, 3672.75, 3673, 3673.25, 3673.5, 3673.75, 3674, 3674.25, 3674.5, 3674.75, 3675, 3675.25, 3675.5, 3675.75, 3676, 3676.25, 3676.5, 3676.75, 3677, 3677.25, 3677.5, 3677.75, 3678, 3678.25, 3678.5, 3678.75, 3679, 3679.25, 3679.5, 3679.75, 3680, 3680.25, 3680.5, 3680.75, 3681, 3681.25, 3681.5, 3681.75, 3682, 3682.25, 3682.5, 3682.75, 3683 ; } ```

Minimal Complete Verifiable Example

```Python import numpy as np import pandas as pd import xarray as xr

time = pd.date_range("1970-01-01", "1970-01-31", freq="6h") ds = xr.Dataset(coords=dict(time=time))

units = "days since 1960-01-01 00:00:00" calendar = "gregorian" encoding = dict(time=dict(units=units, calendar=calendar, dtype=np.dtype("float64")))

ds.to_netcdf("test.nc", encoding=encoding)

! ncdump -v time test.nc ```

MVCE confirmation

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

Relevant log output

No response

Anything else we need to know?

This is still an issue in the current main.

Environment

• [x] Closes #7398
• [x] Tests added
• [x] User visible changes (including notable bug fixes) are documented in whats-new.rst

This PR implements the workaround suggested in #6879.

What is your issue?

Hey all, i have quick use case where i am unsure if that's related to xarray. I just want to do simple arithmetics with a cftime time axis, e.g.

python import xarray as xr import numpy as np time = xr.DataArray(xr.cftime_range(start="2000", periods=6, freq="1MS", calendar="noleap"), dims='time') and then, e.g., compute mid of those timesteps: python time[:-1] + 0.5 * time.diff('time') which results in ```

UFuncTypeError Traceback (most recent call last) Input In [72], in <cell line: 1>() ----> 1 time[1:] + 0.5 * time.diff('time')

File /work/ch0636/g300046/conda_envs/pyremo-dev/lib/python3.9/site-packages/xarray/core/_typed_ops.py:206, in DataArrayOpsMixin.add(self, other) 205 def add(self, other): --> 206 return self._binary_op(other, operator.add)

File /work/ch0636/g300046/conda_envs/pyremo-dev/lib/python3.9/site-packages/xarray/core/dataarray.py:3530, in DataArray._binary_op(self, other, f, reflexive) 3526 other_variable = getattr(other, "variable", other) 3527 other_coords = getattr(other, "coords", None) 3529 variable = ( -> 3530 f(self.variable, other_variable) 3531 if not reflexive 3532 else f(other_variable, self.variable) 3533 ) 3534 coords, indexes = self.coords._merge_raw(other_coords, reflexive) 3535 name = self._result_name(other)

File /work/ch0636/g300046/conda_envs/pyremo-dev/lib/python3.9/site-packages/xarray/core/_typed_ops.py:396, in VariableOpsMixin.add(self, other) 395 def add(self, other): --> 396 return self._binary_op(other, operator.add)

File /work/ch0636/g300046/conda_envs/pyremo-dev/lib/python3.9/site-packages/xarray/core/variable.py:2519, in Variable._binary_op(self, other, f, reflexive) 2516 attrs = self._attrs if keep_attrs else None 2517 with np.errstate(all="ignore"): 2518 new_data = ( -> 2519 f(self_data, other_data) if not reflexive else f(other_data, self_data) 2520 ) 2521 result = Variable(dims, new_data, attrs=attrs) 2522 return result

UFuncTypeError: ufunc 'add' cannot use operands with types dtype('O') and dtype('<m8[ns]') The problem is the differentiation results in `np.timedelta64` objects:python time.diff('time') ![grafik](https://user-images.githubusercontent.com/5659125/207844626-76adf571-92f7-4645-9e9d-5ba01d08afa3.png) with which the arithmetics are not implemented in `cftime`. However, differentiation with `datetime.timdelta` (which is wrapped by `np.timedelta64`) works in this case since it's implemented in cftime, e.g., when i do `np.diff`, it works:python time[:-1] + 0.5 * np.diff(time) ```

i get the expected result. I am not sure what to expect from xarray and if this might be an issue of cftime (e.g., implement differentiation with np.timedelta64). I was wondering why the differentiation with cftime objects results in np.timedelta64 and not datetime.timedelta objects (like in the pure numpy differentation).

What is your issue?

I just recognized that convert_calendar adds a time coordinate to variables that have been static before, e.g.,

```python import numpy as np import pandas as pd import xarray as xr

np.random.seed(0) temperature = 15 + 8 * np.random.randn(2, 2, 3) precipitation = 10 * np.random.rand(2, 2, 3) orog = 0 + 100 * np.random.randn(2, 2) lon = [[-99.83, -99.32], [-99.79, -99.23]] lat = [[42.25, 42.21], [42.63, 42.59]] time = pd.date_range("2014-09-06", periods=3) reference_time = pd.Timestamp("2014-09-05")

ds = xr.Dataset( data_vars=dict( temperature=(["x", "y", "time"], temperature), precipitation=(["x", "y", "time"], precipitation), orog = (["x", "y"], orog) ), coords=dict( lon=(["x", "y"], lon), lat=(["x", "y"], lat), time=time, reference_time=reference_time, ), attrs=dict(description="Weather related data."), ) ds ![grafik](https://user-images.githubusercontent.com/5659125/209124247-4d6bb64f-8396-4a90-bcce-c420b022b0cf.png) I would like to convert the time axis to cftime objects usingpython ds.convert_calendar(ds.time.dt.calendar, use_cftime=True) ds `` ![grafik](https://user-images.githubusercontent.com/5659125/209124475-2ca60508-2d39-432b-816d-0f2718cda5b4.png) However, this results in an unexpected time coordinate in theorog` variable...