[Python-checkins] bpo-41428: Implementation for PEP 604 (GH-21515)

Maggie Moss webhook-mailer at python.org
Tue Sep 15 09:45:18 EDT 2020


https://github.com/python/cpython/commit/1b4552c5e8e925f24c15f707050f22c977171125
commit: 1b4552c5e8e925f24c15f707050f22c977171125
branch: master
author: Maggie Moss <MaggieMoss at users.noreply.github.com>
committer: GitHub <noreply at github.com>
date: 2020-09-09T21:23:24+01:00
summary:

bpo-41428: Implementation for PEP 604 (GH-21515)

See https://www.python.org/dev/peps/pep-0604/ for more information.

Co-authored-by: Pablo Galindo <pablogsal at gmail.com>

files:
A Include/internal/pycore_unionobject.h
A Misc/NEWS.d/next/Core and Builtins/2020-07-28-22-43-27.bpo-41428.FM6xsI.rst
A Objects/unionobject.c
M Lib/test/test_isinstance.py
M Lib/test/test_types.py
M Lib/test/test_typing.py
M Lib/types.py
M Lib/typing.py
M Makefile.pre.in
M Objects/abstract.c
M Objects/typeobject.c
M PCbuild/pythoncore.vcxproj
M PCbuild/pythoncore.vcxproj.filters

diff --git a/Include/internal/pycore_unionobject.h b/Include/internal/pycore_unionobject.h
new file mode 100644
index 0000000000000..fa8ba6ed944c1
--- /dev/null
+++ b/Include/internal/pycore_unionobject.h
@@ -0,0 +1,17 @@
+#ifndef Py_INTERNAL_UNIONOBJECT_H
+#define Py_INTERNAL_UNIONOBJECT_H
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#ifndef Py_BUILD_CORE
+#  error "this header requires Py_BUILD_CORE define"
+#endif
+
+PyAPI_FUNC(PyObject *) _Py_Union(PyObject *args);
+PyAPI_DATA(PyTypeObject) _Py_UnionType;
+
+#ifdef __cplusplus
+}
+#endif
+#endif /* !Py_INTERNAL_UNIONOBJECT_H */
diff --git a/Lib/test/test_isinstance.py b/Lib/test/test_isinstance.py
index 53639e984e48a..91e79c295481d 100644
--- a/Lib/test/test_isinstance.py
+++ b/Lib/test/test_isinstance.py
@@ -4,6 +4,7 @@
 
 import unittest
 import sys
+import typing
 
 
 

@@ -208,6 +209,25 @@ def test_isinstance_abstract(self):
         self.assertEqual(False, isinstance(AbstractChild(), Super))
         self.assertEqual(False, isinstance(AbstractChild(), Child))
 
+    def test_isinstance_with_or_union(self):
+        self.assertTrue(isinstance(Super(), Super | int))
+        self.assertFalse(isinstance(None, str | int))
+        self.assertTrue(isinstance(3, str | int))
+        self.assertTrue(isinstance("", str | int))
+        self.assertTrue(isinstance([], typing.List | typing.Tuple))
+        self.assertTrue(isinstance(2, typing.List | int))
+        self.assertFalse(isinstance(2, typing.List | typing.Tuple))
+        self.assertTrue(isinstance(None, int | None))
+        self.assertFalse(isinstance(3.14, int | str))
+        with self.assertRaises(TypeError):
+            isinstance(2, list[int])
+        with self.assertRaises(TypeError):
+            isinstance(2, list[int] | int)
+        with self.assertRaises(TypeError):
+            isinstance(2, int | str | list[int] | float)
+
+
+
     def test_subclass_normal(self):
         # normal classes
         self.assertEqual(True, issubclass(Super, Super))
@@ -217,6 +237,8 @@ def test_subclass_normal(self):
         self.assertEqual(True, issubclass(Child, Child))
         self.assertEqual(True, issubclass(Child, Super))
         self.assertEqual(False, issubclass(Child, AbstractSuper))
+        self.assertTrue(issubclass(typing.List, typing.List|typing.Tuple))
+        self.assertFalse(issubclass(int, typing.List|typing.Tuple))
 
     def test_subclass_abstract(self):
         # abstract classes
@@ -251,6 +273,16 @@ def test_isinstance_recursion_limit(self):
         # blown
         self.assertRaises(RecursionError, blowstack, isinstance, '', str)
 
+    def test_subclass_with_union(self):
+        self.assertTrue(issubclass(int, int | float | int))
+        self.assertTrue(issubclass(str, str | Child | str))
+        self.assertFalse(issubclass(dict, float|str))
+        self.assertFalse(issubclass(object, float|str))
+        with self.assertRaises(TypeError):
+            issubclass(2, Child | Super)
+        with self.assertRaises(TypeError):
+            issubclass(int, list[int] | Child)
+
     def test_issubclass_refcount_handling(self):
         # bpo-39382: abstract_issubclass() didn't hold item reference while
         # peeking in the bases tuple, in the single inheritance case.
diff --git a/Lib/test/test_types.py b/Lib/test/test_types.py
index 49dc5bf40e3ed..f499fb9c8c51a 100644
--- a/Lib/test/test_types.py
+++ b/Lib/test/test_types.py
@@ -2,6 +2,7 @@
 
 from test.support import run_with_locale
 import collections.abc
+from collections import namedtuple
 import inspect
 import pickle
 import locale
@@ -9,6 +10,12 @@
 import types
 import unittest.mock
 import weakref
+import typing
+
+class Example:
+    pass
+
+class Forward: ...
 
 class TypesTests(unittest.TestCase):
 
@@ -598,6 +605,113 @@ def test_method_descriptor_types(self):
         self.assertIsInstance(int.from_bytes, types.BuiltinMethodType)
         self.assertIsInstance(int.__new__, types.BuiltinMethodType)
 
+    def test_or_types_operator(self):
+        self.assertEqual(int | str, typing.Union[int, str])
+        self.assertNotEqual(int | list, typing.Union[int, str])
+        self.assertEqual(str | int, typing.Union[int, str])
+        self.assertEqual(int | None, typing.Union[int, None])
+        self.assertEqual(None | int, typing.Union[int, None])
+        self.assertEqual(int | str | list, typing.Union[int, str, list])
+        self.assertEqual(int | (str | list), typing.Union[int, str, list])
+        self.assertEqual(str | (int | list), typing.Union[int, str, list])
+        self.assertEqual(typing.List | typing.Tuple, typing.Union[typing.List, typing.Tuple])
+        self.assertEqual(typing.List[int] | typing.Tuple[int], typing.Union[typing.List[int], typing.Tuple[int]])
+        self.assertEqual(typing.List[int] | None, typing.Union[typing.List[int], None])
+        self.assertEqual(None | typing.List[int], typing.Union[None, typing.List[int]])
+        self.assertEqual(str | float | int | complex | int, (int | str) | (float | complex))
+        self.assertEqual(typing.Union[str, int, typing.List[int]], str | int | typing.List[int])
+        self.assertEqual(int | int, int)
+        self.assertEqual(
+            BaseException |
+            bool |
+            bytes |
+            complex |
+            float |
+            int |
+            list |
+            map |
+            set,
+            typing.Union[
+                BaseException,
+                bool,
+                bytes,
+                complex,
+                float,
+                int,
+                list,
+                map,
+                set,
+            ])
+        with self.assertRaises(TypeError):
+            int | 3
+        with self.assertRaises(TypeError):
+            3 | int
+        with self.assertRaises(TypeError):
+            Example() | int
+        with self.assertRaises(TypeError):
+            (int | str) < typing.Union[str, int]
+        with self.assertRaises(TypeError):
+            (int | str) < (int | bool)
+        with self.assertRaises(TypeError):
+            (int | str) <= (int | str)
+        with self.assertRaises(TypeError):
+            # Check that we don't crash if typing.Union does not have a tuple in __args__
+            x = typing.Union[str, int]
+            x.__args__ = [str, int]
+            (int | str ) == x
+
+    def test_or_type_operator_with_TypeVar(self):
+        TV = typing.TypeVar('T')
+        assert TV | str == typing.Union[TV, str]
+        assert str | TV == typing.Union[str, TV]
+
+    def test_or_type_operator_with_forward(self):
+        T = typing.TypeVar('T')
+        ForwardAfter = T | 'Forward'
+        ForwardBefore = 'Forward' | T
+        def forward_after(x: ForwardAfter[int]) -> None: ...
+        def forward_before(x: ForwardBefore[int]) -> None: ...
+        assert typing.get_args(typing.get_type_hints(forward_after)['x']) == (int, Forward)
+        assert typing.get_args(typing.get_type_hints(forward_before)['x']) == (int, Forward)
+
+    def test_or_type_operator_with_Protocol(self):
+        class Proto(typing.Protocol):
+            def meth(self) -> int:
+                ...
+        assert Proto | str == typing.Union[Proto, str]
+
+    def test_or_type_operator_with_Alias(self):
+        assert list | str == typing.Union[list, str]
+        assert typing.List | str == typing.Union[typing.List, str]
+
+    def test_or_type_operator_with_NamedTuple(self):
+        NT=namedtuple('A', ['B', 'C', 'D'])
+        assert NT | str == typing.Union[NT,str]
+
+    def test_or_type_operator_with_TypedDict(self):
+        class Point2D(typing.TypedDict):
+            x: int
+            y: int
+            label: str
+        assert Point2D | str == typing.Union[Point2D, str]
+
+    def test_or_type_operator_with_NewType(self):
+        UserId = typing.NewType('UserId', int)
+        assert UserId | str == typing.Union[UserId, str]
+
+    def test_or_type_operator_with_IO(self):
+        assert typing.IO | str == typing.Union[typing.IO, str]
+
+    def test_or_type_operator_with_SpecialForm(self):
+        assert typing.Any | str == typing.Union[typing.Any, str]
+        assert typing.NoReturn | str == typing.Union[typing.NoReturn, str]
+        assert typing.Optional[int] | str == typing.Union[typing.Optional[int], str]
+        assert typing.Optional[int] | str == typing.Union[int, str, None]
+        assert typing.Union[int, bool] | str == typing.Union[int, bool, str]
+
+    def test_or_type_repr(self):
+        assert repr(int | None) == "int | None"
+        assert repr(int | typing.GenericAlias(list, int)) == "int | list[int]"
 
 class MappingProxyTests(unittest.TestCase):
     mappingproxy = types.MappingProxyType
diff --git a/Lib/test/test_typing.py b/Lib/test/test_typing.py
index b3be99141afca..05140fc61b9bf 100644
--- a/Lib/test/test_typing.py
+++ b/Lib/test/test_typing.py
@@ -244,8 +244,6 @@ def test_subclass_error(self):
             issubclass(int, Union)
         with self.assertRaises(TypeError):
             issubclass(Union, int)
-        with self.assertRaises(TypeError):
-            issubclass(int, Union[int, str])
         with self.assertRaises(TypeError):
             issubclass(Union[int, str], int)
 
@@ -347,10 +345,6 @@ def test_empty(self):
         with self.assertRaises(TypeError):
             Union[()]
 
-    def test_union_instance_type_error(self):
-        with self.assertRaises(TypeError):
-            isinstance(42, Union[int, str])
-
     def test_no_eval_union(self):
         u = Union[int, str]
         def f(x: u): ...
diff --git a/Lib/types.py b/Lib/types.py
index ad2020ec69b63..9642e7212caac 100644
--- a/Lib/types.py
+++ b/Lib/types.py
@@ -294,6 +294,7 @@ def wrapped(*args, **kwargs):
 
 
 GenericAlias = type(list[int])
+Union = type(int | str)
 
 
 __all__ = [n for n in globals() if n[:1] != '_']
diff --git a/Lib/typing.py b/Lib/typing.py
index 2899a0213d434..2aedbeb852a71 100644
--- a/Lib/typing.py
+++ b/Lib/typing.py
@@ -117,7 +117,6 @@
 # namespace, but excluded from __all__ because they might stomp on
 # legitimate imports of those modules.
 
-
 def _type_check(arg, msg, is_argument=True):
     """Check that the argument is a type, and return it (internal helper).
 
@@ -145,7 +144,7 @@ def _type_check(arg, msg, is_argument=True):
         return arg
     if isinstance(arg, _SpecialForm) or arg in (Generic, Protocol):
         raise TypeError(f"Plain {arg} is not valid as type argument")
-    if isinstance(arg, (type, TypeVar, ForwardRef)):
+    if isinstance(arg, (type, TypeVar, ForwardRef, types.Union)):
         return arg
     if not callable(arg):
         raise TypeError(f"{msg} Got {arg!r:.100}.")
@@ -205,7 +204,7 @@ def _remove_dups_flatten(parameters):
     # Flatten out Union[Union[...], ...].
     params = []
     for p in parameters:
-        if isinstance(p, _UnionGenericAlias):
+        if isinstance(p, (_UnionGenericAlias, types.Union)):
             params.extend(p.__args__)
         elif isinstance(p, tuple) and len(p) > 0 and p[0] is Union:
             params.extend(p[1:])
@@ -586,6 +585,12 @@ def __init__(self, name, *constraints, bound=None,
         if def_mod != 'typing':
             self.__module__ = def_mod
 
+    def __or__(self, right):
+        return Union[self, right]
+
+    def __ror__(self, right):
+        return Union[self, right]
+
     def __repr__(self):
         if self.__covariant__:
             prefix = '+'
@@ -693,6 +698,12 @@ def __eq__(self, other):
     def __hash__(self):
         return hash((self.__origin__, self.__args__))
 
+    def __or__(self, right):
+        return Union[self, right]
+
+    def __ror__(self, right):
+        return Union[self, right]
+
     @_tp_cache
     def __getitem__(self, params):
         if self.__origin__ in (Generic, Protocol):
@@ -792,6 +803,11 @@ def __subclasscheck__(self, cls):
     def __reduce__(self):
         return self._name
 
+    def __or__(self, right):
+        return Union[self, right]
+
+    def __ror__(self, right):
+        return Union[self, right]
 
 class _CallableGenericAlias(_GenericAlias, _root=True):
     def __repr__(self):
@@ -878,6 +894,15 @@ def __repr__(self):
                 return f'typing.Optional[{_type_repr(args[0])}]'
         return super().__repr__()
 
+    def __instancecheck__(self, obj):
+        return self.__subclasscheck__(type(obj))
+
+    def __subclasscheck__(self, cls):
+        for arg in self.__args__:
+            if issubclass(cls, arg):
+                return True
+
+
 
 class Generic:
     """Abstract base class for generic types.
diff --git a/Makefile.pre.in b/Makefile.pre.in
index 5d3ac705a3625..921bd08ea505d 100644
--- a/Makefile.pre.in
+++ b/Makefile.pre.in
@@ -432,6 +432,7 @@ OBJECT_OBJS=	\
 		Objects/typeobject.o \
 		Objects/unicodeobject.o \
 		Objects/unicodectype.o \
+		Objects/unionobject.o \
 		Objects/weakrefobject.o
 
 ##########################################################################
@@ -1128,6 +1129,7 @@ PYTHON_HEADERS= \
 		$(srcdir)/Include/internal/pycore_sysmodule.h \
 		$(srcdir)/Include/internal/pycore_traceback.h \
 		$(srcdir)/Include/internal/pycore_tuple.h \
+		$(srcdir)/Include/internal/pycore_unionobject.h \
 		$(srcdir)/Include/internal/pycore_warnings.h \
 		$(DTRACE_HEADERS)
 
diff --git a/Misc/NEWS.d/next/Core and Builtins/2020-07-28-22-43-27.bpo-41428.FM6xsI.rst b/Misc/NEWS.d/next/Core and Builtins/2020-07-28-22-43-27.bpo-41428.FM6xsI.rst
new file mode 100644
index 0000000000000..a6652de927511
--- /dev/null
+++ b/Misc/NEWS.d/next/Core and Builtins/2020-07-28-22-43-27.bpo-41428.FM6xsI.rst	
@@ -0,0 +1 @@
+Implement PEP 604. This supports (int | str) etc. in place of Union[str, int].
\ No newline at end of file
diff --git a/Objects/abstract.c b/Objects/abstract.c
index 7bd72c9b5dcc2..c471f184f6c84 100644
--- a/Objects/abstract.c
+++ b/Objects/abstract.c
@@ -1,6 +1,7 @@
 /* Abstract Object Interface (many thanks to Jim Fulton) */
 
 #include "Python.h"
+#include "pycore_unionobject.h"      // _Py_UnionType && _Py_Union()
 #include "pycore_abstract.h"      // _PyIndex_Check()
 #include "pycore_ceval.h"         // _Py_EnterRecursiveCall()
 #include "pycore_pyerrors.h"      // _PyErr_Occurred()
@@ -839,7 +840,6 @@ binary_op(PyObject *v, PyObject *w, const int op_slot, const char *op_name)
                 Py_TYPE(w)->tp_name);
             return NULL;
         }
-
         return binop_type_error(v, w, op_name);
     }
     return result;
@@ -2412,7 +2412,6 @@ object_isinstance(PyObject *inst, PyObject *cls)
     PyObject *icls;
     int retval;
     _Py_IDENTIFIER(__class__);
-
     if (PyType_Check(cls)) {
         retval = PyObject_TypeCheck(inst, (PyTypeObject *)cls);
         if (retval == 0) {
@@ -2432,7 +2431,7 @@ object_isinstance(PyObject *inst, PyObject *cls)
     }
     else {
         if (!check_class(cls,
-            "isinstance() arg 2 must be a type or tuple of types"))
+            "isinstance() arg 2 must be a type, a tuple of types or a union"))
             return -1;
         retval = _PyObject_LookupAttrId(inst, &PyId___class__, &icls);
         if (icls != NULL) {
@@ -2525,10 +2524,14 @@ recursive_issubclass(PyObject *derived, PyObject *cls)
     if (!check_class(derived,
                      "issubclass() arg 1 must be a class"))
         return -1;
-    if (!check_class(cls,
-                    "issubclass() arg 2 must be a class"
-                    " or tuple of classes"))
+
+    PyTypeObject *type = Py_TYPE(cls);
+    int is_union = (PyType_Check(type) && type == &_Py_UnionType);
+    if (!is_union && !check_class(cls,
+                            "issubclass() arg 2 must be a class,"
+                            " a tuple of classes, or a union.")) {
         return -1;
+    }
 
     return abstract_issubclass(derived, cls);
 }
diff --git a/Objects/typeobject.c b/Objects/typeobject.c
index 74040757a0702..3bb2c338fe0b5 100644
--- a/Objects/typeobject.c
+++ b/Objects/typeobject.c
@@ -6,6 +6,7 @@
 #include "pycore_object.h"
 #include "pycore_pyerrors.h"
 #include "pycore_pystate.h"       // _PyThreadState_GET()
+#include "pycore_unionobject.h"   // _Py_Union()
 #include "frameobject.h"
 #include "structmember.h"         // PyMemberDef
 
@@ -3753,6 +3754,21 @@ type_is_gc(PyTypeObject *type)
     return type->tp_flags & Py_TPFLAGS_HEAPTYPE;
 }
 
+static PyObject *
+type_or(PyTypeObject* self, PyObject* param) {
+    PyObject *tuple = PyTuple_Pack(2, self, param);
+    if (tuple == NULL) {
+        return NULL;
+    }
+    PyObject *new_union = _Py_Union(tuple);
+    Py_DECREF(tuple);
+    return new_union;
+}
+
+static PyNumberMethods type_as_number = {
+        .nb_or = (binaryfunc)type_or, // Add __or__ function
+};
+
 PyTypeObject PyType_Type = {
     PyVarObject_HEAD_INIT(&PyType_Type, 0)
     "type",                                     /* tp_name */
@@ -3764,7 +3780,7 @@ PyTypeObject PyType_Type = {
     0,                                          /* tp_setattr */
     0,                                          /* tp_as_async */
     (reprfunc)type_repr,                        /* tp_repr */
-    0,                                          /* tp_as_number */
+    &type_as_number,                            /* tp_as_number */
     0,                                          /* tp_as_sequence */
     0,                                          /* tp_as_mapping */
     0,                                          /* tp_hash */
@@ -5598,7 +5614,6 @@ PyType_Ready(PyTypeObject *type)
             add_subclass((PyTypeObject *)b, type) < 0)
             goto error;
     }
-
     /* All done -- set the ready flag */
     type->tp_flags =
         (type->tp_flags & ~Py_TPFLAGS_READYING) | Py_TPFLAGS_READY;
diff --git a/Objects/unionobject.c b/Objects/unionobject.c
new file mode 100644
index 0000000000000..0ef7abb4c55c2
--- /dev/null
+++ b/Objects/unionobject.c
@@ -0,0 +1,464 @@
+// types.Union -- used to represent e.g. Union[int, str], int | str
+#include "Python.h"
+#include "pycore_unionobject.h"
+#include "structmember.h"
+
+
+typedef struct {
+    PyObject_HEAD
+    PyObject *args;
+} unionobject;
+
+static void
+unionobject_dealloc(PyObject *self)
+{
+    unionobject *alias = (unionobject *)self;
+
+    Py_XDECREF(alias->args);
+    self->ob_type->tp_free(self);
+}
+
+static Py_hash_t
+union_hash(PyObject *self)
+{
+    unionobject *alias = (unionobject *)self;
+    Py_hash_t h1 = PyObject_Hash(alias->args);
+    if (h1 == -1) {
+        return -1;
+    }
+    return h1;
+}
+
+static int
+is_generic_alias_in_args(PyObject *args) {
+    Py_ssize_t nargs = PyTuple_GET_SIZE(args);
+    for (Py_ssize_t iarg = 0; iarg < nargs; iarg++) {
+        PyObject *arg = PyTuple_GET_ITEM(args, iarg);
+        if (Py_TYPE(arg) == &Py_GenericAliasType) {
+            return 0;
+        }
+    }
+    return 1;
+}
+
+static PyObject *
+union_instancecheck(PyObject *self, PyObject *instance)
+{
+    unionobject *alias = (unionobject *) self;
+    Py_ssize_t nargs = PyTuple_GET_SIZE(alias->args);
+    if (!is_generic_alias_in_args(alias->args)) {
+        PyErr_SetString(PyExc_TypeError,
+            "isinstance() argument 2 cannot contain a parameterized generic");
+        return NULL;
+    }
+    for (Py_ssize_t iarg = 0; iarg < nargs; iarg++) {
+        PyObject *arg = PyTuple_GET_ITEM(alias->args, iarg);
+        if (arg == Py_None) {
+            arg = (PyObject *)&_PyNone_Type;
+        }
+        if (PyType_Check(arg) && PyObject_IsInstance(instance, arg) != 0) {
+            Py_RETURN_TRUE;
+        }
+    }
+    Py_RETURN_FALSE;
+}
+
+static PyObject *
+union_subclasscheck(PyObject *self, PyObject *instance)
+{
+    if (!PyType_Check(instance)) {
+        PyErr_SetString(PyExc_TypeError, "issubclass() arg 1 must be a class");
+        return NULL;
+    }
+    unionobject *alias = (unionobject *)self;
+    if (!is_generic_alias_in_args(alias->args)) {
+        PyErr_SetString(PyExc_TypeError,
+            "issubclass() argument 2 cannot contain a parameterized generic");
+        return NULL;
+    }
+    Py_ssize_t nargs = PyTuple_GET_SIZE(alias->args);
+    for (Py_ssize_t iarg = 0; iarg < nargs; iarg++) {
+        PyObject *arg = PyTuple_GET_ITEM(alias->args, iarg);
+        if (PyType_Check(arg) && (PyType_IsSubtype((PyTypeObject *)instance, (PyTypeObject *)arg) != 0)) {
+            Py_RETURN_TRUE;
+        }
+    }
+   Py_RETURN_FALSE;
+}
+
+static int
+is_typing_module(PyObject *obj) {
+    PyObject *module = PyObject_GetAttrString(obj, "__module__");
+    if (module == NULL) {
+        return -1;
+    }
+    int is_typing = PyUnicode_Check(module) && _PyUnicode_EqualToASCIIString(module, "typing");
+    Py_DECREF(module);
+    return is_typing;
+}
+
+static int
+is_typing_name(PyObject *obj, char *name)
+{
+    PyTypeObject *type = Py_TYPE(obj);
+    if (strcmp(type->tp_name, name) != 0) {
+        return 0;
+    }
+    return is_typing_module(obj);
+}
+
+static PyObject *
+union_richcompare(PyObject *a, PyObject *b, int op)
+{
+    PyObject *result = NULL;
+    if (op != Py_EQ && op != Py_NE) {
+        result = Py_NotImplemented;
+        Py_INCREF(result);
+        return result;
+    }
+
+    PyTypeObject *type = Py_TYPE(b);
+
+    PyObject* a_set = PySet_New(((unionobject*)a)->args);
+    if (a_set == NULL) {
+        return NULL;
+    }
+    PyObject* b_set = PySet_New(NULL);
+    if (b_set == NULL) {
+        goto exit;
+    }
+
+    // Populate b_set with the data from the right object
+    int is_typing_union = is_typing_name(b, "_UnionGenericAlias");
+    if (is_typing_union < 0) {
+        goto exit;
+    }
+    if (is_typing_union) {
+        PyObject *b_args = PyObject_GetAttrString(b, "__args__");
+        if (b_args == NULL) {
+            goto exit;
+        }
+        if (!PyTuple_CheckExact(b_args)) {
+            Py_DECREF(b_args);
+            PyErr_SetString(PyExc_TypeError, "__args__ argument of typing.Union object is not a tuple");
+            goto exit;
+        }
+        Py_ssize_t b_arg_length = PyTuple_GET_SIZE(b_args);
+        for (Py_ssize_t i = 0; i < b_arg_length; i++) {
+            PyObject* arg = PyTuple_GET_ITEM(b_args, i);
+            if (arg == (PyObject *)&_PyNone_Type) {
+                arg = Py_None;
+            }
+            if (PySet_Add(b_set, arg) == -1) {
+                Py_DECREF(b_args);
+                goto exit;
+            }
+        }
+        Py_DECREF(b_args);
+    } else if (type == &_Py_UnionType) {
+        PyObject* args = ((unionobject*) b)->args;
+        Py_ssize_t arg_length = PyTuple_GET_SIZE(args);
+        for (Py_ssize_t i = 0; i < arg_length; i++) {
+            PyObject* arg = PyTuple_GET_ITEM(args, i);
+            if (PySet_Add(b_set, arg) == -1) {
+                goto exit;
+            }
+        }
+    } else {
+        if (PySet_Add(b_set, b) == -1) {
+            goto exit;
+        }
+    }
+    result = PyObject_RichCompare(a_set, b_set, op);
+exit:
+    Py_XDECREF(a_set);
+    Py_XDECREF(b_set);
+    return result;
+}
+
+static PyObject*
+flatten_args(PyObject* args)
+{
+    int arg_length = PyTuple_GET_SIZE(args);
+    int total_args = 0;
+    // Get number of total args once it's flattened.
+    for (Py_ssize_t i = 0; i < arg_length; i++) {
+        PyObject *arg = PyTuple_GET_ITEM(args, i);
+        PyTypeObject* arg_type = Py_TYPE(arg);
+        if (arg_type == &_Py_UnionType) {
+            total_args += PyTuple_GET_SIZE(((unionobject*) arg)->args);
+        } else {
+            total_args++;
+        }
+    }
+    // Create new tuple of flattened args.
+    PyObject *flattened_args = PyTuple_New(total_args);
+    if (flattened_args == NULL) {
+        return NULL;
+    }
+    Py_ssize_t pos = 0;
+    for (Py_ssize_t i = 0; i < arg_length; i++) {
+        PyObject *arg = PyTuple_GET_ITEM(args, i);
+        PyTypeObject* arg_type = Py_TYPE(arg);
+        if (arg_type == &_Py_UnionType) {
+            PyObject* nested_args = ((unionobject*)arg)->args;
+            int nested_arg_length = PyTuple_GET_SIZE(nested_args);
+            for (int j = 0; j < nested_arg_length; j++) {
+                PyObject* nested_arg = PyTuple_GET_ITEM(nested_args, j);
+                Py_INCREF(nested_arg);
+                PyTuple_SET_ITEM(flattened_args, pos, nested_arg);
+                pos++;
+            }
+        } else {
+            Py_INCREF(arg);
+            PyTuple_SET_ITEM(flattened_args, pos, arg);
+            pos++;
+        }
+    }
+    return flattened_args;
+}
+
+static PyObject*
+dedup_and_flatten_args(PyObject* args)
+{
+    args = flatten_args(args);
+    if (args == NULL) {
+        return NULL;
+    }
+    Py_ssize_t arg_length = PyTuple_GET_SIZE(args);
+    PyObject *new_args = PyTuple_New(arg_length);
+    if (new_args == NULL) {
+        return NULL;
+    }
+    // Add unique elements to an array.
+    int added_items = 0;
+    for (Py_ssize_t i = 0; i < arg_length; i++) {
+        int is_duplicate = 0;
+        PyObject* i_element = PyTuple_GET_ITEM(args, i);
+        for (Py_ssize_t j = i + 1; j < arg_length; j++) {
+            PyObject* j_element = PyTuple_GET_ITEM(args, j);
+            if (i_element == j_element) {
+                is_duplicate = 1;
+            }
+        }
+        if (!is_duplicate) {
+            Py_INCREF(i_element);
+            PyTuple_SET_ITEM(new_args, added_items, i_element);
+            added_items++;
+        }
+    }
+    Py_DECREF(args);
+    _PyTuple_Resize(&new_args, added_items);
+    return new_args;
+}
+
+static int
+is_typevar(PyObject *obj)
+{
+    return is_typing_name(obj, "TypeVar");
+}
+
+static int
+is_special_form(PyObject *obj)
+{
+    return is_typing_name(obj, "_SpecialForm");
+}
+
+static int
+is_new_type(PyObject *obj)
+{
+    PyTypeObject *type = Py_TYPE(obj);
+    if (type != &PyFunction_Type) {
+        return 0;
+    }
+    return is_typing_module(obj);
+}
+
+static int
+is_unionable(PyObject *obj)
+{
+    if (obj == Py_None) {
+        return 1;
+    }
+    PyTypeObject *type = Py_TYPE(obj);
+    return (
+        is_typevar(obj) ||
+        is_new_type(obj) ||
+        is_special_form(obj) ||
+        PyType_Check(obj) ||
+        type == &Py_GenericAliasType ||
+        type == &_Py_UnionType);
+}
+
+static PyObject *
+type_or(PyTypeObject* self, PyObject* param)
+{
+    PyObject *tuple = PyTuple_Pack(2, self, param);
+    if (tuple == NULL) {
+        return NULL;
+    }
+    PyObject *new_union = _Py_Union(tuple);
+    Py_DECREF(tuple);
+    return new_union;
+}
+
+static int
+union_repr_item(_PyUnicodeWriter *writer, PyObject *p)
+{
+    _Py_IDENTIFIER(__module__);
+    _Py_IDENTIFIER(__qualname__);
+    _Py_IDENTIFIER(__origin__);
+    _Py_IDENTIFIER(__args__);
+    PyObject *qualname = NULL;
+    PyObject *module = NULL;
+    PyObject *r = NULL;
+    int err;
+
+    int has_origin = _PyObject_HasAttrId(p, &PyId___origin__);
+    if (has_origin < 0) {
+        goto exit;
+    }
+
+    if (has_origin) {
+        int has_args = _PyObject_HasAttrId(p, &PyId___args__);
+        if (has_args < 0) {
+            goto exit;
+        }
+        if (has_args) {
+            // It looks like a GenericAlias
+            goto use_repr;
+        }
+    }
+
+    if (_PyObject_LookupAttrId(p, &PyId___qualname__, &qualname) < 0) {
+        goto exit;
+    }
+    if (qualname == NULL) {
+        goto use_repr;
+    }
+    if (_PyObject_LookupAttrId(p, &PyId___module__, &module) < 0) {
+        goto exit;
+    }
+    if (module == NULL || module == Py_None) {
+        goto use_repr;
+    }
+
+    // Looks like a class
+    if (PyUnicode_Check(module) &&
+        _PyUnicode_EqualToASCIIString(module, "builtins"))
+    {
+        // builtins don't need a module name
+        r = PyObject_Str(qualname);
+        goto exit;
+    }
+    else {
+        r = PyUnicode_FromFormat("%S.%S", module, qualname);
+        goto exit;
+    }
+
+use_repr:
+    r = PyObject_Repr(p);
+exit:
+    Py_XDECREF(qualname);
+    Py_XDECREF(module);
+    if (r == NULL) {
+        return -1;
+    }
+    err = _PyUnicodeWriter_WriteStr(writer, r);
+    Py_DECREF(r);
+    return err;
+}
+
+static PyObject *
+union_repr(PyObject *self)
+{
+    unionobject *alias = (unionobject *)self;
+    Py_ssize_t len = PyTuple_GET_SIZE(alias->args);
+
+    _PyUnicodeWriter writer;
+    _PyUnicodeWriter_Init(&writer);
+     for (Py_ssize_t i = 0; i < len; i++) {
+        if (i > 0 && _PyUnicodeWriter_WriteASCIIString(&writer, " | ", 3) < 0) {
+            goto error;
+        }
+        PyObject *p = PyTuple_GET_ITEM(alias->args, i);
+        if (union_repr_item(&writer, p) < 0) {
+            goto error;
+        }
+    }
+    return _PyUnicodeWriter_Finish(&writer);
+error:
+    _PyUnicodeWriter_Dealloc(&writer);
+    return NULL;
+}
+
+static PyMemberDef union_members[] = {
+        {"__args__", T_OBJECT, offsetof(unionobject, args), READONLY},
+        {0}
+};
+
+static PyMethodDef union_methods[] = {
+        {"__instancecheck__", union_instancecheck, METH_O},
+        {"__subclasscheck__", union_subclasscheck, METH_O},
+        {0}};
+
+static PyNumberMethods union_as_number = {
+        .nb_or = (binaryfunc)type_or, // Add __or__ function
+};
+
+PyTypeObject _Py_UnionType = {
+    PyVarObject_HEAD_INIT(&PyType_Type, 0)
+    .tp_name = "types.Union",
+    .tp_doc = "Represent a PEP 604 union type\n"
+              "\n"
+              "E.g. for int | str",
+    .tp_basicsize = sizeof(unionobject),
+    .tp_dealloc = unionobject_dealloc,
+    .tp_alloc = PyType_GenericAlloc,
+    .tp_free = PyObject_Del,
+    .tp_flags = Py_TPFLAGS_DEFAULT,
+    .tp_hash = union_hash,
+    .tp_getattro = PyObject_GenericGetAttr,
+    .tp_members = union_members,
+    .tp_methods = union_methods,
+    .tp_richcompare = union_richcompare,
+    .tp_as_number = &union_as_number,
+    .tp_repr = union_repr,
+};
+
+PyObject *
+_Py_Union(PyObject *args)
+{
+    assert(PyTuple_CheckExact(args));
+
+    unionobject* result = NULL;
+
+    // Check arguments are unionable.
+    int nargs = PyTuple_GET_SIZE(args);
+    for (Py_ssize_t iarg = 0; iarg < nargs; iarg++) {
+        PyObject *arg = PyTuple_GET_ITEM(args, iarg);
+        if (arg == NULL) {
+            return NULL;
+        }
+        int is_arg_unionable = is_unionable(arg);
+        if (is_arg_unionable < 0) {
+            return NULL;
+        }
+        if (!is_arg_unionable) {
+            Py_INCREF(Py_NotImplemented);
+            return Py_NotImplemented;
+        }
+    }
+
+    result = PyObject_New(unionobject, &_Py_UnionType);
+    if (result == NULL) {
+        return NULL;
+    }
+
+    result->args = dedup_and_flatten_args(args);
+    if (result->args == NULL) {
+        Py_DECREF(result);
+        return NULL;
+    }
+    return (PyObject*)result;
+}
diff --git a/PCbuild/pythoncore.vcxproj b/PCbuild/pythoncore.vcxproj
index 295b5e21eba5c..266a193c1e86a 100644
--- a/PCbuild/pythoncore.vcxproj
+++ b/PCbuild/pythoncore.vcxproj
@@ -196,6 +196,7 @@
     <ClInclude Include="..\Include\internal\pycore_sysmodule.h" />
     <ClInclude Include="..\Include\internal\pycore_traceback.h" />
     <ClInclude Include="..\Include\internal\pycore_tuple.h" />
+    <ClInclude Include="..\Include\internal\pycore_unionobject.h" />
     <ClInclude Include="..\Include\internal\pycore_warnings.h" />
     <ClInclude Include="..\Include\interpreteridobject.h" />
     <ClInclude Include="..\Include\intrcheck.h" />
@@ -412,6 +413,7 @@
     <ClCompile Include="..\Objects\typeobject.c" />
     <ClCompile Include="..\Objects\unicodectype.c" />
     <ClCompile Include="..\Objects\unicodeobject.c" />
+    <ClCompile Include="..\Objects\unionobject.c" />
     <ClCompile Include="..\Objects\weakrefobject.c" />
     <ClCompile Include="..\Parser\myreadline.c" />
     <ClCompile Include="..\Parser\tokenizer.c" />
diff --git a/PCbuild/pythoncore.vcxproj.filters b/PCbuild/pythoncore.vcxproj.filters
index ec82e42f61e8b..22d9b79157698 100644
--- a/PCbuild/pythoncore.vcxproj.filters
+++ b/PCbuild/pythoncore.vcxproj.filters
@@ -573,6 +573,9 @@
     <ClInclude Include="..\Include\internal\pycore_tuple.h">
       <Filter>Include\internal</Filter>
     </ClInclude>
+    <ClInclude Include="..\Include\internal\pycore_unionobject.h">
+      <Filter>Include\internal</Filter>
+    </ClInclude>
     <ClInclude Include="$(zlibDir)\crc32.h">
       <Filter>Modules\zlib</Filter>
     </ClInclude>
@@ -1175,6 +1178,9 @@
     <ClCompile Include="..\Objects\genericaliasobject.c">
       <Filter>Objects</Filter>
     </ClCompile>
+    <ClCompile Include="..\Objects\unionobject.c">
+      <Filter>Objects</Filter>
+    </ClCompile>
   </ItemGroup>
   <ItemGroup>
     <ResourceCompile Include="..\PC\python_nt.rc">



More information about the Python-checkins mailing list