patch ExtensionBlock fill_value, closes #27781, closes #29563

Author: scottgigante

doc/source/whatsnew/v1.1.0.rst

@@ -843,7 +843,7 @@ Sparse
 ^^^^^^
 - Creating a :class:`SparseArray` from timezone-aware dtype will issue a warning before dropping timezone information, instead of doing so silently (:issue:`32501`)
 - Bug in :meth:`arrays.SparseArray.from_spmatrix` wrongly read scipy sparse matrix (:issue:`31991`)
--
+- Fixed bug where :class:`DataFrame` containing :class:`SparseArray` filled with ``NaN`` when indexed by a list-like (:issue:`27781`, :issue:`29563`)
 
 ExtensionArray
 ^^^^^^^^^^^^^^
@@ -856,6 +856,7 @@ ExtensionArray
 - Fixed bug where :meth:`StringArray.memory_usage` was not implemented (:issue:`33963`)
 - Fixed bug where :meth:`DataFrameGroupBy` would ignore the ``min_count`` argument for aggregations on nullable boolean dtypes (:issue:`34051`)
 - Fixed bug that `DataFrame(columns=.., dtype='string')` would fail (:issue:`27953`, :issue:`33623`)
+- Fixed bug in :meth:`ExtensionBlock.fill_value` which did not use ``dtype.fill_value`` where available (:issue:`27781`, :issue:`29563`)
 
 Other
 ^^^^^

pandas/core/internals/blocks.py

@@ -1627,7 +1627,10 @@ def _holder(self):
     @property
     def fill_value(self):
         # Used in reindex_indexer
-        return self.values.dtype.na_value
+        try:
+            return self.values.dtype.fill_value
+        except AttributeError:
+            return self.values.dtype.na_value
 
     @property
     def _can_hold_na(self):

pandas/tests/extension/test_sparse.py

@@ -41,6 +41,11 @@ def data_for_twos(request):
     return SparseArray(np.ones(100) * 2)
 
 
+@pytest.fixture(params=[0, np.nan])
+def data_zeros(request):
+    return SparseArray(np.zeros(100, dtype=int), fill_value=request.param)
+
+
 @pytest.fixture(params=[0, np.nan])
 def data_missing(request):
     """Length 2 array with [NA, Valid]"""
@@ -169,6 +174,34 @@ def test_reindex(self, data, na_value):
         self._check_unsupported(data)
         super().test_reindex(data, na_value)
 
+    def test_boolean_mask_frame_fill_value(self, data):
+        # https://github.com/pandas-dev/pandas/issues/27781
+        df = pd.DataFrame({"A": data})
+
+        mask = np.random.choice([True, False], df.shape[0])
+        result = pd.isna(df.iloc[mask]["A"])
+        expected = pd.isna(df["A"].iloc[mask])
+        self.assert_series_equal(result, expected)
+
+        mask = pd.Series(mask, index=df.index)
+        result = pd.isna(df.loc[mask]["A"])
+        expected = pd.isna(df["A"].loc[mask])
+        self.assert_series_equal(result, expected)
+
+    def test_fancy_index_frame_fill_value(self, data):
+        # https://github.com/pandas-dev/pandas/issues/27781
+        df = pd.DataFrame({"A": data})
+
+        mask = np.random.choice(df.shape[0], df.shape[0])
+        result = pd.isna(df.iloc[mask]["A"])
+        expected = pd.isna(df["A"].iloc[mask])
+        self.assert_series_equal(result, expected)
+
+        mask = pd.Series(mask, index=df.index)
+        result = pd.isna(df.loc[mask]["A"])
+        expected = pd.isna(df["A"].loc[mask])
+        self.assert_series_equal(result, expected)
+
 
 # Skipping TestSetitem, since we don't implement it.