BUG: Fix lost precision with common type of uint64/int64

doc/source/reference/arrays.rst

@@ -667,6 +667,7 @@ Data type introspection
     api.types.is_dtype_equal
     api.types.is_extension_array_dtype
     api.types.is_float_dtype
+    api.types.is_implicit_conversion_to_float64
     api.types.is_int64_dtype
     api.types.is_integer_dtype
     api.types.is_interval_dtype

doc/source/whatsnew/v3.0.0.rst

@@ -404,6 +404,7 @@ Other API changes
 - Index set operations (like union or intersection) will now ignore the dtype of
   an empty ``RangeIndex`` or empty ``Index`` with object dtype when determining
   the dtype of the resulting Index (:issue:`60797`)
+- Added :func:`pandas.api.types.is_implicit_conversion_to_float64` to check if there is a silent conversion to float64 between two dtypes(:issue:`61676`)
 
 .. ---------------------------------------------------------------------------
 .. _whatsnew_300.deprecations:

pandas/core/algorithms.py

@@ -47,15 +47,14 @@
     is_bool_dtype,
     is_complex_dtype,
     is_dict_like,
-    is_dtype_equal,
     is_extension_array_dtype,
     is_float,
     is_float_dtype,
+    is_implicit_conversion_to_float64,
     is_integer,
     is_integer_dtype,
     is_list_like,
     is_object_dtype,
-    is_signed_integer_dtype,
     needs_i8_conversion,
 )
 from pandas.core.dtypes.concat import concat_compat
@@ -511,8 +510,7 @@ def isin(comps: ListLike, values: ListLike) -> npt.NDArray[np.bool_]:
         if (
             len(values) > 0
             and values.dtype.kind in "iufcb"
-            and not is_signed_integer_dtype(comps)
-            and not is_dtype_equal(values, comps)
+            and is_implicit_conversion_to_float64(values, comps)
         ):
             # GH#46485 Use object to avoid upcast to float64 later
             # TODO: Share with _find_common_type_compat

pandas/core/dtypes/api.py

@@ -17,6 +17,7 @@
     is_float,
     is_float_dtype,
     is_hashable,
+    is_implicit_conversion_to_float64,
     is_int64_dtype,
     is_integer,
     is_integer_dtype,
@@ -59,6 +60,7 @@
     "is_float",
     "is_float_dtype",
     "is_hashable",
+    "is_implicit_conversion_to_float64",
     "is_int64_dtype",
     "is_integer",
     "is_integer_dtype",

pandas/core/dtypes/common.py

@@ -713,6 +713,80 @@ def is_dtype_equal(source, target) -> bool:
         return False
 
 
+def is_implicit_conversion_to_float64(source, target) -> bool:
+    """
+    Check if there is an implicit conversion to float64 with both dtypes.
+
+    Parameters
+    ----------
+    source : type or str
+        The first dtype to compare.
+    target : type or str
+        The second dtype to compare.
+
+    Returns
+    -------
+    boolean
+        Whether or not there is an implicit conversion to float64.
+
+    See Also
+    --------
+    api.types.is_categorical_dtype : Check whether the provided array or dtype
+                                            is of the Categorical dtype.
+    api.types.is_string_dtype : Check whether the provided array or dtype
+                                       is of the string dtype.
+    api.types.is_object_dtype : Check whether an array-like or dtype is of the
+                                       object dtype.
+
+    Examples
+    --------
+    >>> from pandas.api.types import is_implicit_conversion_to_float64
+    >>> is_implicit_conversion_to_float64(int, float)
+    False
+    >>> is_implicit_conversion_to_float64("int", int)
+    False
+    >>> is_implicit_conversion_to_float64(int, np.int64)
+    False
+    >>> is_implicit_conversion_to_float64(np.uint64, np.int64)
+    True
+    >>> is_implicit_conversion_to_float64(np.uint64, np.float64)
+    False
+    >>> is_implicit_conversion_to_float64(np.uint64, np.uint64)
+    False
+    >>> is_implicit_conversion_to_float64(np.uint32, np.uint32)
+    False
+    >>> is_implicit_conversion_to_float64(np.uint32, np.int32)
+    False
+    >>> is_implicit_conversion_to_float64(np.int32, np.int32)
+    False
+    >>> is_implicit_conversion_to_float64(object, "category")
+    False
+    >>> is_implicit_conversion_to_float64(np.int64, pd.UInt64Dtype())
+    True
+    >>> from pandas.core.dtypes.dtypes import CategoricalDtype
+    >>> is_implicit_conversion_to_float64(CategoricalDtype(), "category")
+    False
+    """
+    try:
+        src = _get_dtype(source)
+        tar = _get_dtype(target)
+        # check only valid dtypes related to implicit conversion to float64
+        # other data types derived from 64-bit integers such as U/Int64Dtype
+        # should also work
+        if (
+            src.kind in "iu"
+            and src.itemsize == 8  # type: ignore[union-attr]
+            and tar.kind in "iu"
+            and tar.itemsize == 8  # type: ignore[union-attr]
+        ):
+            return src != tar
+        else:
+            return False
+    except (TypeError, AttributeError, ImportError):
+        # invalid comparison
+        return False
+
+
 def is_integer_dtype(arr_or_dtype) -> bool:
     """
     Check whether the provided array or dtype is of an integer dtype.
@@ -1934,6 +2008,7 @@ def is_all_strings(value: ArrayLike) -> bool:
     "is_extension_array_dtype",
     "is_file_like",
     "is_float_dtype",
+    "is_implicit_conversion_to_float64",
     "is_int64_dtype",
     "is_integer_dtype",
     "is_interval_dtype",

pandas/tests/api/test_api.py

@@ -295,6 +295,7 @@ class TestApi(Base):
         "is_float",
         "is_float_dtype",
         "is_hashable",
+        "is_implicit_conversion_to_float64",
         "is_int64_dtype",
         "is_integer",
         "is_integer_dtype",

pandas/tests/api/test_types.py

@@ -20,6 +20,7 @@ class TestTypes(Base):
         "is_dtype_equal",
         "is_float",
         "is_float_dtype",
+        "is_implicit_conversion_to_float64",
         "is_int64_dtype",
         "is_integer",
         "is_integer_dtype",

pandas/tests/test_algos.py

@@ -1197,6 +1197,13 @@ def test_isin_unsigned_dtype(self):
         expected = Series(False)
         tm.assert_series_equal(result, expected)
 
+    def test_isin_unsigned_dtype_other_side(self):
+        # GH#46485
+        ser = Series([1378774140726870442], dtype=np.int64)
+        result = ser.isin([np.uint64(1378774140726870528)])
+        expected = Series(False)
+        tm.assert_series_equal(result, expected)
+
 
 class TestValueCounts:
     def test_value_counts(self):