Diff low precision ints

doc/source/whatsnew/v1.5.0.rst

@@ -302,7 +302,9 @@ Conversion
 - Bug in :meth:`Float64Index.astype` to unsigned integer dtype incorrectly casting to ``np.int64`` dtype (:issue:`45309`)
 - Bug in :meth:`Series.astype` and :meth:`DataFrame.astype` from floating dtype to unsigned integer dtype failing to raise in the presence of negative values (:issue:`45151`)
 - Bug in :func:`array` with ``FloatingDtype`` and values containing float-castable strings incorrectly raising (:issue:`45424`)
+- Bug in :meth:`DataFrame.diff` upcasting ``np.int8`` and ``np.int16`` to ``np.float32`` which is inconsistent with other methods like :meth:`DataFrame.shift` (:issue:`45562`)
 - Bug when comparing string and datetime64ns objects causing ``OverflowError`` exception. (:issue:`45506`)
+-
 
 Strings
 ^^^^^^^

pandas/core/algorithms.py

@@ -1617,9 +1617,9 @@ def diff(arr, n: int, axis: int = 0):
         # int8, int16 are incompatible with float64,
         # see https://github.com/cython/cython/issues/2646
         if arr.dtype.name in ["int8", "int16"]:
-            dtype = np.float32
-        else:
-            dtype = np.float64
+            arr = arr.astype("int32")
+
+        dtype = np.float64
 
     orig_ndim = arr.ndim
     if orig_ndim == 1:

pandas/tests/frame/methods/test_diff.py

@@ -296,8 +296,6 @@ def test_diff_all_int_dtype(self, any_int_numpy_dtype):
         df = DataFrame(range(5))
         df = df.astype(any_int_numpy_dtype)
         result = df.diff()
-        expected_dtype = (
-            "float32" if any_int_numpy_dtype in ("int8", "int16") else "float64"
-        )
+        expected_dtype = "float64"
         expected = DataFrame([np.nan, 1.0, 1.0, 1.0, 1.0], dtype=expected_dtype)
         tm.assert_frame_equal(result, expected)

pandas/tests/groupby/test_groupby_shift_diff.py

@@ -76,7 +76,7 @@ def test_group_diff_real(any_real_numpy_dtype):
     )
     result = df.groupby("a")["b"].diff()
     exp_dtype = "float"
-    if any_real_numpy_dtype in ["int8", "int16", "float32"]:
+    if any_real_numpy_dtype == "float32":
         exp_dtype = "float32"
     expected = Series([np.nan, np.nan, np.nan, 1.0, 3.0], dtype=exp_dtype, name="b")
     tm.assert_series_equal(result, expected)

pandas/tests/test_algos.py

@@ -2447,7 +2447,7 @@ def test_diff_ea_axis(self):
     def test_diff_low_precision_int(self, dtype):
         arr = np.array([0, 1, 1, 0, 0], dtype=dtype)
         result = algos.diff(arr, 1)
-        expected = np.array([np.nan, 1, 0, -1, 0], dtype="float32")
+        expected = np.array([np.nan, 1, 0, -1, 0], dtype="float64")
         tm.assert_numpy_array_equal(result, expected)