BUG GH22858 When creating empty dataframe, only cast int to float if index given (#22963)

Author: JustinZhengBC

doc/source/whatsnew/v0.24.0.txt

@@ -199,6 +199,7 @@ Other Enhancements
 
 Backwards incompatible API changes
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+- A newly constructed empty :class:`DataFrame` with integer as the ``dtype`` will now only be cast to ``float64`` if ``index`` is specified (:issue:`22858`)
 
 
 .. _whatsnew_0240.api_breaking.interval_values:

pandas/core/dtypes/cast.py

@@ -1220,7 +1220,9 @@ def construct_1d_arraylike_from_scalar(value, length, dtype):
             dtype = dtype.dtype
 
         # coerce if we have nan for an integer dtype
-        if is_integer_dtype(dtype) and isna(value):
+        # GH 22858: only cast to float if an index
+        # (passed here as length) is specified
+        if length and is_integer_dtype(dtype) and isna(value):
             dtype = np.float64
         subarr = np.empty(length, dtype=dtype)
         subarr.fill(value)

pandas/tests/frame/test_constructors.py

@@ -798,25 +798,20 @@ def test_constructor_mrecarray(self):
             result = DataFrame(mrecs, index=[1, 2])
             assert_fr_equal(result, expected)
 
-    def test_constructor_corner(self):
+    def test_constructor_corner_shape(self):
         df = DataFrame(index=[])
         assert df.values.shape == (0, 0)
 
-        # empty but with specified dtype
-        df = DataFrame(index=lrange(10), columns=['a', 'b'], dtype=object)
-        assert df.values.dtype == np.object_
-
-        # does not error but ends up float
-        df = DataFrame(index=lrange(10), columns=['a', 'b'], dtype=int)
-        assert df.values.dtype == np.dtype('float64')
-
-        # #1783 empty dtype object
-        df = DataFrame({}, columns=['foo', 'bar'])
-        assert df.values.dtype == np.object_
-
-        df = DataFrame({'b': 1}, index=lrange(10), columns=list('abc'),
-                       dtype=int)
-        assert df.values.dtype == np.dtype('float64')
+    @pytest.mark.parametrize("data, index, columns, dtype, expected", [
+        (None, lrange(10), ['a', 'b'], object, np.object_),
+        (None, None, ['a', 'b'], 'int64', np.dtype('int64')),
+        (None, lrange(10), ['a', 'b'], int, np.dtype('float64')),
+        ({}, None, ['foo', 'bar'], None, np.object_),
+        ({'b': 1}, lrange(10), list('abc'), int, np.dtype('float64'))
+    ])
+    def test_constructor_dtype(self, data, index, columns, dtype, expected):
+        df = DataFrame(data, index, columns, dtype)
+        assert df.values.dtype == expected
 
     def test_constructor_scalar_inference(self):
         data = {'int': 1, 'bool': True,