document limitations of cftime arithmetics

[larsbuntemeyer]

• Closes Document cftime arithmetic limitations #7381
• User visible changes (including notable bug fixes) are documented in whats-new.rst

[larsbuntemeyer]

I have a code example that inentionally should fail, i used the notation

.. code-block:: python

for that part.

[larsbuntemeyer]

I'm not sure why the readthedocs fails, my local build seems to work.

[spencerkclark]

Thanks @larsbuntemeyer! It would be nice to get this working cleanly someday, especially now that xarray supports coarser resolution np.timedelta64 values.

You could also consider including a more verbose, but also more robust, workaround that adds the two DataArrays via apply_ufunc:

def add_cftime_and_timedelta64(cftime_da, timedelta64_da, join="inner"):
    def func(cftime_array, timedelta64_array):
        shape = cftime_array.shape
        cftime_array = cftime_array.ravel()
        timedelta64_array = timedelta64_array.ravel()
        timedelta_array = pd.to_timedelta(timedelta64_array).to_pytimedelta()
        return (cftime_array + timedelta_array).reshape(shape)
    
    return xr.apply_ufunc(
        func, 
        cftime_da,
        timedelta64_da,
        dask="parallelized",
        output_dtypes=[cftime_da.dtype],
        join=join
    )

This function has the advantage that it preserves the coordinate alignment, as well as multi-dimensional and dask compatibility features of xarray.

[larsbuntemeyer]

Thanks @larsbuntemeyer! It would be nice to get this working cleanly someday, especially now that xarray supports coarser resolution np.timedelta64 values.

I agree! That would be really helpful!

You could also consider including a more verbose, but also more robust, workaround that adds the two DataArrays via apply_ufunc:

def add_cftime_and_timedelta64(cftime_da, timedelta64_da, join="inner"):
    def func(cftime_array, timedelta64_array):
        shape = cftime_array.shape
        cftime_array = cftime_array.ravel()
        timedelta64_array = timedelta64_array.ravel()
        timedelta_array = pd.to_timedelta(timedelta64_array).to_pytimedelta()
        return (cftime_array + timedelta_array).reshape(shape)
    
    return xr.apply_ufunc(
        func, 
        cftime_da,
        timedelta64_da,
        dask="parallelized",
        output_dtypes=[cftime_da.dtype],
        join=join
    )

Thanks for this solution! I'll add this to the docs.

[larsbuntemeyer]

@spencerkclark I added your workaround to the docs. Thanks a lot for that!

[spencerkclark]

Thanks @larsbuntemeyer!

[welcome]

Congratulations on completing your first pull request! Welcome to Xarray! We are proud of you, and hope to see you again!