Fix to _get_nearest_indexer for pydata/xarray#3751

ci/deps/azure-36-locale.yaml

@@ -15,6 +15,7 @@ dependencies:
 
   # pandas dependencies
   - beautifulsoup4
+  - cftime  # Needed for downstream xarray.CFTimeIndex test
   - gcsfs
   - html5lib
   - ipython

ci/deps/azure-37-locale.yaml

@@ -14,6 +14,7 @@ dependencies:
 
   # pandas dependencies
   - beautifulsoup4
+  - cftime  # Needed for downstream xarray.CFTimeIndex test
   - html5lib
   - ipython
   - jinja2

ci/deps/azure-macos-36.yaml

@@ -14,6 +14,7 @@ dependencies:
   # pandas dependencies
   - beautifulsoup4
   - bottleneck
+  - cftime  # Needed for downstream xarray.CFTimeIndex test
   - html5lib
   - jinja2
   - lxml

ci/deps/travis-36-cov.yaml

@@ -15,6 +15,7 @@ dependencies:
   # pandas dependencies
   - beautifulsoup4
   - botocore>=1.11
+  - cftime  # Needed for downstream xarray.CFTimeIndex test
   - cython>=0.29.13
   - dask
   - fastparquet>=0.3.2

pandas/core/indexes/base.py

@@ -280,7 +280,7 @@ def _outer_indexer(self, left, right):
     # Constructors
 
     def __new__(
-        cls, data=None, dtype=None, copy=False, name=None, tupleize_cols=True, **kwargs,
+        cls, data=None, dtype=None, copy=False, name=None, tupleize_cols=True, **kwargs
     ) -> "Index":
 
         from pandas.core.indexes.range import RangeIndex
@@ -3049,8 +3049,9 @@ def _get_nearest_indexer(self, target: "Index", limit, tolerance) -> np.ndarray:
         left_indexer = self.get_indexer(target, "pad", limit=limit)
         right_indexer = self.get_indexer(target, "backfill", limit=limit)
 
-        left_distances = np.abs(self[left_indexer] - target)
-        right_distances = np.abs(self[right_indexer] - target)
+        target_values = target._values
+        left_distances = np.abs(self._values[left_indexer] - target_values)
+        right_distances = np.abs(self._values[right_indexer] - target_values)
 
         op = operator.lt if self.is_monotonic_increasing else operator.le
         indexer = np.where(
@@ -3059,11 +3060,14 @@ def _get_nearest_indexer(self, target: "Index", limit, tolerance) -> np.ndarray:
             right_indexer,
         )
         if tolerance is not None:
-            indexer = self._filter_indexer_tolerance(target, indexer, tolerance)
+            indexer = self._filter_indexer_tolerance(target_values, indexer, tolerance)
         return indexer
 
     def _filter_indexer_tolerance(
-        self, target: "Index", indexer: np.ndarray, tolerance
+        self,
+        target: Union["Index", np.ndarray, ExtensionArray],
+        indexer: np.ndarray,
+        tolerance,
     ) -> np.ndarray:
         distance = abs(self.values[indexer] - target)
         indexer = np.where(distance <= tolerance, indexer, -1)

pandas/tests/test_downstream.py

@@ -47,6 +47,18 @@ def test_xarray(df):
     assert df.to_xarray() is not None
 
 
+def test_xarray_cftimeindex_nearest():
+    # https://github.com/pydata/xarray/issues/3751
+    cftime = import_module("cftime")
+    xarray = import_module("xarray")
+
+    times = xarray.cftime_range("0001", periods=2)
+    da = xarray.DataArray(range(2), coords=[("time", times)])
+    result = da.sel(time=cftime.DatetimeGregorian(2000, 1, 1), method="nearest")
+    expected = xarray.DataArray(1, coords={"time": cftime.DatetimeGregorian(1, 1, 2)})
+    xarray.testing.assert_identical(result, expected)
+
+
 def test_oo_optimizable():
     # GH 21071
     subprocess.check_call([sys.executable, "-OO", "-c", "import pandas"])