What is your issue?

I know that generally, ds2 = ds connects the two objects in memory, and changes in one will also cause changes in the other.

However, I generally assume that certain operations should break this connection, for example: - extracting the underlying np.array from a dataset (changing its type and destroying a lot of the xarray-specific information: index, dimensions, etc.) - using the underlying np.array into a new dataset

In other words, I would expect that using ds['var'].values would be similar to copy.deepcopy(ds['var'].values).

Here's an example that illustrates how in these cases, the objects are still linked in memory:

(apologies for the somewhat hokey example)

``` import xarray as xr import numpy as np

Create a dataset

ds = xr.Dataset(coords = {'lon':(['lon'],np.array([178.2,179.2,-179.8, -178.8,-177.8,-176.8]))}) print('\nds: ') print(ds)

Create a new dataset that uses the values of the first dataset

ds2 = xr.Dataset({'lon1':(['lon'],ds.lon.values)}, coords = {'lon':(['lon'],ds.lon.values)}) print('\nds2: ') print(ds2)

Change ds2's 'lon1' variable

ds2['lon1'][ds2['lon1']<0] = 360 + ds2['lon1'][ds2['lon1']<0]

ds2 is changed as expected

print('\nds2 (should be modified): ') print(ds2)

ds is changed, which is not expected

print('\nds (should not be modified): ') print(ds) ```

The question is - am I right (from a UX perspective) to expect these kinds of operations to disconnect the objects in memory? If so, I might try to update the docs to be a bit clearer on this. (or, alternatively, if these kinds of operations should disconnect the objects in memory, maybe it's better to have .values also call .copy(deep=True).values)

Appreciate y'all's thoughts on this!

What happened?

I'm running ds.interp() using multi-dimensional new coordinates, using xarray's broadcasting to expand the original dataset to new dimensions. In this case, I'm only interpolating on one dimension, but broadcasting out to others.

If the dimensions are all numeric (or, presumably, able to be forced to numeric), then this works without an issue. However, if one of the other dimensions is, e.g., populated with string indices (weather station names, model run ids, etc.), then this process fails, even if the dimension on which the interpolating is conducted is purely numeric.

What did you expect to happen?

Here is an example with only numeric dimensions that works as expected: ``` import xarray as xr import numpy as np

da1 = xr.DataArray(np.reshape(np.arange(0,12),(3,4)), coords = {'dim0':np.arange(0,3), 'dim1':np.arange(0,4)})

da2 = xr.DataArray(np.random.normal(loc=1,size=(2,4),scale=0.5), coords = {'dim2':np.arange(0,2), 'dim1':np.arange(0,4)})

da1.interp(dim0=da2) ``` this produces something like: as expected.

Minimal Complete Verifiable Example

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

da1 = xr.DataArray(np.reshape(np.arange(0,12),(3,4)), coords = {'dim0':np.arange(0,3), 'dim1':np.arange(0,4).astype(str)})

da2 = xr.DataArray(np.random.normal(loc=1,size=(2,4),scale=0.5), coords = {'dim2':np.arange(0,2), 'dim1':np.arange(0,4).astype(str)})

da1.interp(dim0=da2) ```

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

```Python

TypeError Traceback (most recent call last) Cell In[48], line 9 1 da1 = xr.DataArray(np.reshape(np.arange(0,12),(3,4)), 2 coords = {'dim0':np.arange(0,3), 3 'dim1':np.arange(0,4).astype(str)}) 5 da2 = xr.DataArray(np.random.normal(loc=1,size=(2,4),scale=0.5), 6 coords = {'dim2':np.arange(0,2), 7 'dim1':np.arange(0,4).astype(str)}) ----> 9 da1.interp(dim0=da2)

File ~/.conda/envs/climate/lib/python3.10/site-packages/xarray/core/dataarray.py:2204, in DataArray.interp(self, coords, method, assume_sorted, kwargs, coords_kwargs) 2199 if self.dtype.kind not in "uifc": 2200 raise TypeError( 2201 "interp only works for a numeric type array. " 2202 "Given {}.".format(self.dtype) 2203 ) -> 2204 ds = self._to_temp_dataset().interp( 2205 coords, 2206 method=method, 2207 kwargs=kwargs, 2208 assume_sorted=assume_sorted, 2209 coords_kwargs, 2210 ) 2211 return self._from_temp_dataset(ds)

File ~/.conda/envs/climate/lib/python3.10/site-packages/xarray/core/dataset.py:3666, in Dataset.interp(self, coords, method, assume_sorted, kwargs, method_non_numeric, **coords_kwargs) 3664 if method in ["linear", "nearest"]: 3665 for k, v in validated_indexers.items(): -> 3666 obj, newidx = missing._localize(obj, {k: v}) 3667 validated_indexers[k] = newidx[k] 3669 # optimization: create dask coordinate arrays once per Dataset 3670 # rather than once per Variable when dask.array.unify_chunks is called later 3671 # GH4739

File ~/.conda/envs/climate/lib/python3.10/site-packages/xarray/core/missing.py:562, in _localize(var, indexes_coords) 560 indexes = {} 561 for dim, [x, new_x] in indexes_coords.items(): --> 562 minval = np.nanmin(new_x.values) 563 maxval = np.nanmax(new_x.values) 564 index = x.to_index()

File <array_function internals>:5, in nanmin(args, *kwargs)

File ~/.conda/envs/climate/lib/python3.10/site-packages/numpy/lib/nanfunctions.py:319, in nanmin(a, axis, out, keepdims) 315 kwargs['keepdims'] = keepdims 316 if type(a) is np.ndarray and a.dtype != np.object_: 317 # Fast, but not safe for subclasses of ndarray, or object arrays, 318 # which do not implement isnan (gh-9009), or fmin correctly (gh-8975) --> 319 res = np.fmin.reduce(a, axis=axis, out=out, **kwargs) 320 if np.isnan(res).any(): 321 warnings.warn("All-NaN slice encountered", RuntimeWarning, 322 stacklevel=3)

TypeError: cannot perform reduce with flexible type ```

Anything else we need to know?

I'm pretty sure the issue is in this optimization step.

It calls _localize() from missing.py, which calls np.nanmin() and np.nanmax() on all the coordinates, including the ones that aren't used in the interpolation, but only in the broadcasting.

Perhaps a way to fix this would be to have a test in localize for numeric indices, and then only subset the numeric dimensions? (I could see generalizing _localize() to other data types may be more trouble than it's worth, especially for unsorted string dimensions...) Or only subset the dimensions used in the interpolation itself? Or, alternatively, having a way to turn off optimizations like this?

Environment