issue_comments: 482880911

As you've noticed, these arrays are "read only" because otherwise indexing can modify more than the original values, e.g., consider: python original = xr.DataArray(np.zeros(3), dims='x') result = original.expand_dims(y=2) result.data.flags.writeable = True result[0, 0] = 1 print(result) Both "y" values were set to 1! <xarray.DataArray (y: 2, x: 3)> array([[1., 0., 0.], [1., 0., 0.]]) Dimensions without coordinates: y, x

The work around is to call .copy() on the array after calling expand_dims(), e.g., original = xr.DataArray(np.zeros(3), dims='x') result = original.expand_dims(y=2).copy() result[0, 0] = 1 print(result) Now the correct result is printed: <xarray.DataArray (y: 2, x: 3)> array([[1., 0., 0.], [0., 0., 0.]]) Dimensions without coordinates: y, x

This is indeed intended behavior: by making the result read-only, we can expand dimensions without copying the original data, and without needing to worry about indexing modifying the wrong values.

That said, we could certainly improve the usability of this feature in xarray. Some options: - Mention the work-around of using .copy() in the error message xarray prints when an array is read-only. - Add a copy argument to expand_dims, so users can write copy=True if they want a writeable result. - Consider changing the behavior when a dimension is inserted with size 1 to make the result writeable -- in this case, individual elements of result can be modified (which will also modify the input array). But maybe it would be surprising to users if the result of expand_dims() is sometimes but not always writeable?