BUG: melt losing ea dtype (pandas-dev#50316)

* BUG: melt losing ea dtype

* Add asv

* Fix mypy

Author: phofl

asv_bench/benchmarks/reshape.py

@@ -15,12 +15,17 @@
 
 
 class Melt:
-    def setup(self):
-        self.df = DataFrame(np.random.randn(10000, 3), columns=["A", "B", "C"])
-        self.df["id1"] = np.random.randint(0, 10, 10000)
-        self.df["id2"] = np.random.randint(100, 1000, 10000)
+    params = ["float64", "Float64"]
+    param_names = ["dtype"]
+
+    def setup(self, dtype):
+        self.df = DataFrame(
+            np.random.randn(100_000, 3), columns=["A", "B", "C"], dtype=dtype
+        )
+        self.df["id1"] = pd.Series(np.random.randint(0, 10, 10000))
+        self.df["id2"] = pd.Series(np.random.randint(100, 1000, 10000))
 
-    def time_melt_dataframe(self):
+    def time_melt_dataframe(self, dtype):
         melt(self.df, id_vars=["id1", "id2"])
 
 

doc/source/whatsnew/v2.0.0.rst

@@ -910,6 +910,7 @@ Reshaping
 ^^^^^^^^^
 - Bug in :meth:`DataFrame.pivot_table` raising ``TypeError`` for nullable dtype and ``margins=True`` (:issue:`48681`)
 - Bug in :meth:`DataFrame.unstack` and :meth:`Series.unstack` unstacking wrong level of :class:`MultiIndex` when :class:`MultiIndex` has mixed names (:issue:`48763`)
+- Bug in :meth:`DataFrame.melt` losing extension array dtype (:issue:`41570`)
 - Bug in :meth:`DataFrame.pivot` not respecting ``None`` as column name (:issue:`48293`)
 - Bug in :func:`join` when ``left_on`` or ``right_on`` is or includes a :class:`CategoricalIndex` incorrectly raising ``AttributeError`` (:issue:`48464`)
 - Bug in :meth:`DataFrame.pivot_table` raising ``ValueError`` with parameter ``margins=True`` when result is an empty :class:`DataFrame` (:issue:`49240`)

pandas/core/reshape/melt.py

@@ -33,6 +33,8 @@
 from pandas.core.tools.numeric import to_numeric
 
 if TYPE_CHECKING:
+    from pandas._typing import AnyArrayLike
+
     from pandas import DataFrame
 
 
@@ -124,7 +126,7 @@ def melt(
     N, K = frame.shape
     K -= len(id_vars)
 
-    mdata = {}
+    mdata: dict[Hashable, AnyArrayLike] = {}
     for col in id_vars:
         id_data = frame.pop(col)
         if is_extension_array_dtype(id_data):
@@ -141,17 +143,15 @@ def melt(
 
     mcolumns = id_vars + var_name + [value_name]
 
-    # error: Incompatible types in assignment (expression has type "ndarray",
-    # target has type "Series")
-    mdata[value_name] = frame._values.ravel("F")  # type: ignore[assignment]
+    if frame.shape[1] > 0:
+        mdata[value_name] = concat(
+            [frame.iloc[:, i] for i in range(frame.shape[1])]
+        ).values
+    else:
+        mdata[value_name] = frame._values.ravel("F")
     for i, col in enumerate(var_name):
         # asanyarray will keep the columns as an Index
-
-        # error: Incompatible types in assignment (expression has type "ndarray", target
-        # has type "Series")
-        mdata[col] = np.asanyarray(  # type: ignore[assignment]
-            frame.columns._get_level_values(i)
-        ).repeat(N)
+        mdata[col] = np.asanyarray(frame.columns._get_level_values(i)).repeat(N)
 
     result = frame._constructor(mdata, columns=mcolumns)
 

pandas/tests/reshape/test_melt.py

@@ -420,6 +420,24 @@ def test_melt_with_duplicate_columns(self):
         )
         tm.assert_frame_equal(result, expected)
 
+    @pytest.mark.parametrize("dtype", ["Int8", "Int64"])
+    def test_melt_ea_dtype(self, dtype):
+        # GH#41570
+        df = DataFrame(
+            {
+                "a": pd.Series([1, 2], dtype="Int8"),
+                "b": pd.Series([3, 4], dtype=dtype),
+            }
+        )
+        result = df.melt()
+        expected = DataFrame(
+            {
+                "variable": ["a", "a", "b", "b"],
+                "value": pd.Series([1, 2, 3, 4], dtype=dtype),
+            }
+        )
+        tm.assert_frame_equal(result, expected)
+
 
 class TestLreshape:
     def test_pairs(self):