BUG/PERF: MaskedArray.searchsorted(np.nan)

doc/source/whatsnew/v1.5.0.rst

@@ -196,7 +196,7 @@ Sparse
 
 ExtensionArray
 ^^^^^^^^^^^^^^
--
+- Bug in :meth:`IntegerArray.searchsorted` and :meth:`FloatingArray.searchsorted` returning inconsistent results when acting on ``np.nan`` (:issue:`45255`)
 -
 
 Styler

pandas/core/arrays/masked.py

@@ -3,6 +3,7 @@
 from typing import (
     TYPE_CHECKING,
     Any,
+    Literal,
     Sequence,
     TypeVar,
     overload,
@@ -78,6 +79,10 @@
 if TYPE_CHECKING:
     from pandas import Series
     from pandas.core.arrays import BooleanArray
+    from pandas._typing import (
+        NumpySorter,
+        NumpyValueArrayLike,
+    )
 
 from pandas.compat.numpy import function as nv
 
@@ -681,6 +686,23 @@ def copy(self: BaseMaskedArrayT) -> BaseMaskedArrayT:
         mask = mask.copy()
         return type(self)(data, mask, copy=False)
 
+    @doc(ExtensionArray.searchsorted)
+    def searchsorted(
+        self,
+        value: NumpyValueArrayLike | ExtensionArray,
+        side: Literal["left", "right"] = "left",
+        sorter: NumpySorter = None,
+    ) -> npt.NDArray[np.intp] | np.intp:
+        if self._hasna:
+            raise ValueError(
+                "searchsorted requires array to be sorted, which is impossible "
+                "with NAs present."
+            )
+        if isinstance(value, ExtensionArray):
+            value = value.astype(object)
+        # Base class searchsorted would cast to object, which is *much* slower.
+        return self._data.searchsorted(value, side=side, sorter=sorter)
+
     @doc(ExtensionArray.factorize)
     def factorize(self, na_sentinel: int = -1) -> tuple[np.ndarray, ExtensionArray]:
         arr = self._data

pandas/tests/arrays/masked_shared.py

@@ -55,6 +55,14 @@ def test_scalar(self, other, comparison_op, dtype):
 class NumericOps:
     # Shared by IntegerArray and FloatingArray, not BooleanArray
 
+    def test_searchsorted_nan(self, dtype):
+        # The base class casts to object dtype, for which searchsorted returns
+        #  0 from the left and 10 from the right.
+        arr = pd.array(range(10), dtype=dtype)
+
+        assert arr.searchsorted(np.nan, side="left") == 10
+        assert arr.searchsorted(np.nan, side="right") == 10
+
     def test_no_shared_mask(self, data):
         result = data + 1
         assert not tm.shares_memory(result, data)