BUG: regression in DataFrame.combine_first with integer columns (GH14687) (#14886)

Author: jorisvandenbossche

doc/source/whatsnew/v0.19.2.txt

@@ -78,7 +78,7 @@ Bug Fixes
 - Bug in clipboard functions on linux with python2 with unicode and separators (:issue:`13747`)
 - Bug in clipboard functions on Windows 10 and python 3 (:issue:`14362`, :issue:`12807`)
 - Bug in ``.to_clipboard()`` and Excel compat (:issue:`12529`)
-
+- Bug in ``DataFrame.combine_first()`` for integer columns (:issue:`14687`).
 
 - Bug in ``pd.read_csv()`` in which the ``dtype`` parameter was not being respected for empty data (:issue:`14712`)
 - Bug in ``pd.read_csv()`` in which the ``nrows`` parameter was not being respected for large input when using the C engine for parsing (:issue:`7626`)

pandas/core/frame.py

@@ -3665,10 +3665,8 @@ def combine(self, other, func, fill_value=None, overwrite=True):
                 otherSeries[other_mask] = fill_value
 
             # if we have different dtypes, possibily promote
-            if notnull(series).all():
-                new_dtype = this_dtype
-                otherSeries = otherSeries.astype(new_dtype)
-            else:
+            new_dtype = this_dtype
+            if not is_dtype_equal(this_dtype, other_dtype):
                 new_dtype = _find_common_type([this_dtype, other_dtype])
                 if not is_dtype_equal(this_dtype, new_dtype):
                     series = series.astype(new_dtype)

pandas/tests/frame/test_combine_concat.py

@@ -725,3 +725,13 @@ def test_combine_first_period(self):
         exp = pd.DataFrame({'P': exp_dts}, index=[1, 2, 3, 4, 5, 7])
         tm.assert_frame_equal(res, exp)
         self.assertEqual(res['P'].dtype, 'object')
+
+    def test_combine_first_int(self):
+        # GH14687 - integer series that do no align exactly
+
+        df1 = pd.DataFrame({'a': [0, 1, 3, 5]}, dtype='int64')
+        df2 = pd.DataFrame({'a': [1, 4]}, dtype='int64')
+
+        res = df1.combine_first(df2)
+        tm.assert_frame_equal(res, df1)
+        self.assertEqual(res['a'].dtype, 'int64')