Added test case to lock in behaviour (pandas-dev#48541)

* Added test case to lock in behaviour

* In previous versions, concatenating to empty EA was resetting type information to np.object

* Update whatsnew

* Addressed code review comments

* whatsnew: made "Loss of dtype" more specific and improved wording (EA -> ExtensionArray, etc)
* testcase: moved test case to test_empty.py and use `tm.assert_frame_equal()` with an expected dataframe

* Fixed style issue

* Use concat/DataFrame as they were directly imported (instead of `pd...`)

* Improved whatsnew entry as per suggestion

Co-authored-by: Sven <[email protected]>

Author: 2 people

doc/source/whatsnew/v1.6.0.rst

@@ -214,6 +214,7 @@ Sparse
 ExtensionArray
 ^^^^^^^^^^^^^^
 - Bug in :meth:`Series.mean` overflowing unnecessarily with nullable integers (:issue:`48378`)
+- Bug when concatenating an empty DataFrame with an ExtensionDtype to another DataFrame with the same ExtensionDtype, the resulting dtype turned into object (:issue:`48510`)
 -
 
 Styler

pandas/tests/reshape/concat/test_empty.py

@@ -284,3 +284,11 @@ def test_concat_empty_dataframe_different_dtypes(self):
         result = concat([df1[:0], df2[:0]])
         assert result["a"].dtype == np.int64
         assert result["b"].dtype == np.object_
+
+    def test_concat_to_empty_ea(self):
+        """48510 `concat` to an empty EA should maintain type EA dtype."""
+        df_empty = DataFrame({"a": pd.array([], dtype=pd.Int64Dtype())})
+        df_new = DataFrame({"a": pd.array([1, 2, 3], dtype=pd.Int64Dtype())})
+        expected = df_new.copy()
+        result = concat([df_empty, df_new])
+        tm.assert_frame_equal(result, expected)