bpo-32604: Expose the subinterpreters C-API in a "private" stdlib module. (gh-1748)

The module is primarily intended for internal use in the test suite.  Building the module under Windows will come in a follow-up PR.

6 files changed, 3539 insertions, 1 deletions

diff --git a/Include/internal/pystate.h b/Include/internal/pystate.h
index 0b464bcb2e..2b60b25c19 100644
--- a/Include/internal/pystate.h
+++ b/Include/internal/pystate.h
@@ -65,6 +65,79 @@ PyAPI_FUNC(_PyInitError) _PyPathConfig_Calculate(
 PyAPI_FUNC(void) _PyPathConfig_Clear(_PyPathConfig *config);
 
 
+/* interpreter state */
+
+PyAPI_FUNC(PyInterpreterState *) _PyInterpreterState_LookUpID(PY_INT64_T);
+
+
+/* cross-interpreter data */
+
+struct _xid;
+
+// _PyCrossInterpreterData is similar to Py_buffer as an effectively
+// opaque struct that holds data outside the object machinery.  This
+// is necessary to pass between interpreters in the same process.
+typedef struct _xid {
+    // data is the cross-interpreter-safe derivation of a Python object
+    // (see _PyObject_GetCrossInterpreterData).  It will be NULL if the
+    // new_object func (below) encodes the data.
+    void *data;
+    // obj is the Python object from which the data was derived.  This
+    // is non-NULL only if the data remains bound to the object in some
+    // way, such that the object must be "released" (via a decref) when
+    // the data is released.  In that case it is automatically
+    // incref'ed (to match the automatic decref when releaed).
+    PyObject *obj;
+    // interp is the ID of the owning interpreter of the original
+    // object.  It corresponds to the active interpreter when
+    // _PyObject_GetCrossInterpreterData() was called.  This should only
+    // be set by the cross-interpreter machinery.
+    //
+    // We use the ID rather than the PyInterpreterState to avoid issues
+    // with deleted interpreters.
+    int64_t interp;
+    // new_object is a function that returns a new object in the current
+    // interpreter given the data.  The resulting object (a new
+    // reference) will be equivalent to the original object.  This field
+    // is required.
+    PyObject *(*new_object)(struct _xid *);
+    // free is called when the data is released.  If it is NULL then
+    // nothing will be done to free the data.  For some types this is
+    // okay (e.g. bytes) and for those types this field should be set
+    // to NULL.  However, for most the data was allocated just for
+    // cross-interpreter use, so it must be freed when
+    // _PyCrossInterpreterData_Release is called or the memory will
+    // leak.  In that case, at the very least this field should be set
+    // to PyMem_RawFree (the default if not explicitly set to NULL).
+    // The call will happen with the original interpreter activated.
+    void (*free)(void *);
+} _PyCrossInterpreterData;
+
+typedef int (*crossinterpdatafunc)(PyObject *, _PyCrossInterpreterData *);
+PyAPI_FUNC(int) _PyObject_CheckCrossInterpreterData(PyObject *);
+
+PyAPI_FUNC(int) _PyObject_GetCrossInterpreterData(PyObject *, _PyCrossInterpreterData *);
+PyAPI_FUNC(PyObject *) _PyCrossInterpreterData_NewObject(_PyCrossInterpreterData *);
+PyAPI_FUNC(void) _PyCrossInterpreterData_Release(_PyCrossInterpreterData *);
+
+/* cross-interpreter data registry */
+
+/* For now we use a global registry of shareable classes.  An
+   alternative would be to add a tp_* slot for a class's
+   crossinterpdatafunc. It would be simpler and more efficient. */
+
+PyAPI_FUNC(int) _PyCrossInterpreterData_Register_Class(PyTypeObject *, crossinterpdatafunc);
+PyAPI_FUNC(crossinterpdatafunc) _PyCrossInterpreterData_Lookup(PyObject *);
+
+struct _xidregitem;
+
+struct _xidregitem {
+    PyTypeObject *cls;
+    crossinterpdatafunc getdata;
+    struct _xidregitem *next;
+};
+
+
 /* Full Python runtime state */
 
 typedef struct pyruntimestate {
@@ -86,6 +159,11 @@ typedef struct pyruntimestate {
            using a Python int. */
         int64_t next_id;
     } interpreters;
+    // XXX Remove this field once we have a tp_* slot.
+    struct _xidregistry {
+        PyThread_type_lock mutex;
+        struct _xidregitem *head;
+    } xidregistry;
 
 #define NEXITFUNCS 32
     void (*exitfuncs[NEXITFUNCS])(void);
diff --git a/Lib/test/test__xxsubinterpreters.py b/Lib/test/test__xxsubinterpreters.py
new file mode 100644
index 0000000000..2b170443a3
--- /dev/null
+++ b/Lib/test/test__xxsubinterpreters.py
@@ -0,0 +1,1118 @@
+import contextlib
+import os
+import pickle
+from textwrap import dedent, indent
+import threading
+import unittest
+
+from test import support
+from test.support import script_helper
+
+interpreters = support.import_module('_xxsubinterpreters')
+
+
+def _captured_script(script):
+    r, w = os.pipe()
+    indented = script.replace('\n', '\n                ')
+    wrapped = dedent(f"""
+        import contextlib
+        with open({w}, 'w') as chan:
+            with contextlib.redirect_stdout(chan):
+                {indented}
+        """)
+    return wrapped, open(r)
+
+
+def _run_output(interp, request, shared=None):
+    script, chan = _captured_script(request)
+    with chan:
+        interpreters.run_string(interp, script, shared)
+        return chan.read()
+
+
+@contextlib.contextmanager
+def _running(interp):
+    r, w = os.pipe()
+    def run():
+        interpreters.run_string(interp, dedent(f"""
+            # wait for "signal"
+            with open({r}) as chan:
+                chan.read()
+            """))
+
+    t = threading.Thread(target=run)
+    t.start()
+
+    yield
+
+    with open(w, 'w') as chan:
+        chan.write('done')
+    t.join()
+
+
+class IsShareableTests(unittest.TestCase):
+
+    def test_default_shareables(self):
+        shareables = [
+                # singletons
+                None,
+                # builtin objects
+                b'spam',
+                ]
+        for obj in shareables:
+            with self.subTest(obj):
+                self.assertTrue(
+                    interpreters.is_shareable(obj))
+
+    def test_not_shareable(self):
+        class Cheese:
+            def __init__(self, name):
+                self.name = name
+            def __str__(self):
+                return self.name
+
+        class SubBytes(bytes):
+            """A subclass of a shareable type."""
+
+        not_shareables = [
+                # singletons
+                True,
+                False,
+                NotImplemented,
+                ...,
+                # builtin types and objects
+                type,
+                object,
+                object(),
+                Exception(),
+                42,
+                100.0,
+                'spam',
+                # user-defined types and objects
+                Cheese,
+                Cheese('Wensleydale'),
+                SubBytes(b'spam'),
+                ]
+        for obj in not_shareables:
+            with self.subTest(obj):
+                self.assertFalse(
+                    interpreters.is_shareable(obj))
+
+
+class TestBase(unittest.TestCase):
+
+    def tearDown(self):
+        for id in interpreters.list_all():
+            if id == 0:  # main
+                continue
+            try:
+                interpreters.destroy(id)
+            except RuntimeError:
+                pass  # already destroyed
+
+        for cid in interpreters.channel_list_all():
+            try:
+                interpreters.channel_destroy(cid)
+            except interpreters.ChannelNotFoundError:
+                pass  # already destroyed
+
+
+class ListAllTests(TestBase):
+
+    def test_initial(self):
+        main = interpreters.get_main()
+        ids = interpreters.list_all()
+        self.assertEqual(ids, [main])
+
+    def test_after_creating(self):
+        main = interpreters.get_main()
+        first = interpreters.create()
+        second = interpreters.create()
+        ids = interpreters.list_all()
+        self.assertEqual(ids, [main, first, second])
+
+    def test_after_destroying(self):
+        main = interpreters.get_main()
+        first = interpreters.create()
+        second = interpreters.create()
+        interpreters.destroy(first)
+        ids = interpreters.list_all()
+        self.assertEqual(ids, [main, second])
+
+
+class GetCurrentTests(TestBase):
+
+    def test_main(self):
+        main = interpreters.get_main()
+        cur = interpreters.get_current()
+        self.assertEqual(cur, main)
+
+    def test_subinterpreter(self):
+        main = interpreters.get_main()
+        interp = interpreters.create()
+        out = _run_output(interp, dedent("""
+            import _xxsubinterpreters as _interpreters
+            print(_interpreters.get_current())
+            """))
+        cur = int(out.strip())
+        _, expected = interpreters.list_all()
+        self.assertEqual(cur, expected)
+        self.assertNotEqual(cur, main)
+
+
+class GetMainTests(TestBase):
+
+    def test_from_main(self):
+        [expected] = interpreters.list_all()
+        main = interpreters.get_main()
+        self.assertEqual(main, expected)
+
+    def test_from_subinterpreter(self):
+        [expected] = interpreters.list_all()
+        interp = interpreters.create()
+        out = _run_output(interp, dedent("""
+            import _xxsubinterpreters as _interpreters
+            print(_interpreters.get_main())
+            """))
+        main = int(out.strip())
+        self.assertEqual(main, expected)
+
+
+class IsRunningTests(TestBase):
+
+    def test_main(self):
+        main = interpreters.get_main()
+        self.assertTrue(interpreters.is_running(main))
+
+    def test_subinterpreter(self):
+        interp = interpreters.create()
+        self.assertFalse(interpreters.is_running(interp))
+
+        with _running(interp):
+            self.assertTrue(interpreters.is_running(interp))
+        self.assertFalse(interpreters.is_running(interp))
+
+    def test_from_subinterpreter(self):
+        interp = interpreters.create()
+        out = _run_output(interp, dedent(f"""
+            import _xxsubinterpreters as _interpreters
+            if _interpreters.is_running({interp}):
+                print(True)
+            else:
+                print(False)
+            """))
+        self.assertEqual(out.strip(), 'True')
+
+    def test_already_destroyed(self):
+        interp = interpreters.create()
+        interpreters.destroy(interp)
+        with self.assertRaises(RuntimeError):
+            interpreters.is_running(interp)
+
+    def test_does_not_exist(self):
+        with self.assertRaises(RuntimeError):
+            interpreters.is_running(1_000_000)
+
+    def test_bad_id(self):
+        with self.assertRaises(RuntimeError):
+            interpreters.is_running(-1)
+
+
+class CreateTests(TestBase):
+
+    def test_in_main(self):
+        id = interpreters.create()
+
+        self.assertIn(id, interpreters.list_all())
+
+    @unittest.skip('enable this test when working on pystate.c')
+    def test_unique_id(self):
+        seen = set()
+        for _ in range(100):
+            id = interpreters.create()
+            interpreters.destroy(id)
+            seen.add(id)
+
+        self.assertEqual(len(seen), 100)
+
+    def test_in_thread(self):
+        lock = threading.Lock()
+        id = None
+        def f():
+            nonlocal id
+            id = interpreters.create()
+            lock.acquire()
+            lock.release()
+
+        t = threading.Thread(target=f)
+        with lock:
+            t.start()
+        t.join()
+        self.assertIn(id, interpreters.list_all())
+
+    def test_in_subinterpreter(self):
+        main, = interpreters.list_all()
+        id1 = interpreters.create()
+        out = _run_output(id1, dedent("""
+            import _xxsubinterpreters as _interpreters
+            id = _interpreters.create()
+            print(id)
+            """))
+        id2 = int(out.strip())
+
+        self.assertEqual(set(interpreters.list_all()), {main, id1, id2})
+
+    def test_in_threaded_subinterpreter(self):
+        main, = interpreters.list_all()
+        id1 = interpreters.create()
+        id2 = None
+        def f():
+            nonlocal id2
+            out = _run_output(id1, dedent("""
+                import _xxsubinterpreters as _interpreters
+                id = _interpreters.create()
+                print(id)
+                """))
+            id2 = int(out.strip())
+
+        t = threading.Thread(target=f)
+        t.start()
+        t.join()
+
+        self.assertEqual(set(interpreters.list_all()), {main, id1, id2})
+
+    def test_after_destroy_all(self):
+        before = set(interpreters.list_all())
+        # Create 3 subinterpreters.
+        ids = []
+        for _ in range(3):
+            id = interpreters.create()
+            ids.append(id)
+        # Now destroy them.
+        for id in ids:
+            interpreters.destroy(id)
+        # Finally, create another.
+        id = interpreters.create()
+        self.assertEqual(set(interpreters.list_all()), before | {id})
+
+    def test_after_destroy_some(self):
+        before = set(interpreters.list_all())
+        # Create 3 subinterpreters.
+        id1 = interpreters.create()
+        id2 = interpreters.create()
+        id3 = interpreters.create()
+        # Now destroy 2 of them.
+        interpreters.destroy(id1)
+        interpreters.destroy(id3)
+        # Finally, create another.
+        id = interpreters.create()
+        self.assertEqual(set(interpreters.list_all()), before | {id, id2})
+
+
+class DestroyTests(TestBase):
+
+    def test_one(self):
+        id1 = interpreters.create()
+        id2 = interpreters.create()
+        id3 = interpreters.create()
+        self.assertIn(id2, interpreters.list_all())
+        interpreters.destroy(id2)
+        self.assertNotIn(id2, interpreters.list_all())
+        self.assertIn(id1, interpreters.list_all())
+        self.assertIn(id3, interpreters.list_all())
+
+    def test_all(self):
+        before = set(interpreters.list_all())
+        ids = set()
+        for _ in range(3):
+            id = interpreters.create()
+            ids.add(id)
+        self.assertEqual(set(interpreters.list_all()), before | ids)
+        for id in ids:
+            interpreters.destroy(id)
+        self.assertEqual(set(interpreters.list_all()), before)
+
+    def test_main(self):
+        main, = interpreters.list_all()
+        with self.assertRaises(RuntimeError):
+            interpreters.destroy(main)
+
+        def f():
+            with self.assertRaises(RuntimeError):
+                interpreters.destroy(main)
+
+        t = threading.Thread(target=f)
+        t.start()
+        t.join()
+
+    def test_already_destroyed(self):
+        id = interpreters.create()
+        interpreters.destroy(id)
+        with self.assertRaises(RuntimeError):
+            interpreters.destroy(id)
+
+    def test_does_not_exist(self):
+        with self.assertRaises(RuntimeError):
+            interpreters.destroy(1_000_000)
+
+    def test_bad_id(self):
+        with self.assertRaises(RuntimeError):
+            interpreters.destroy(-1)
+
+    def test_from_current(self):
+        main, = interpreters.list_all()
+        id = interpreters.create()
+        script = dedent("""
+            import _xxsubinterpreters as _interpreters
+            _interpreters.destroy({})
+            """).format(id)
+
+        with self.assertRaises(RuntimeError):
+            interpreters.run_string(id, script)
+        self.assertEqual(set(interpreters.list_all()), {main, id})
+
+    def test_from_sibling(self):
+        main, = interpreters.list_all()
+        id1 = interpreters.create()
+        id2 = interpreters.create()
+        script = dedent("""
+            import _xxsubinterpreters as _interpreters
+            _interpreters.destroy({})
+            """).format(id2)
+        interpreters.run_string(id1, script)
+
+        self.assertEqual(set(interpreters.list_all()), {main, id1})
+
+    def test_from_other_thread(self):
+        id = interpreters.create()
+        def f():
+            interpreters.destroy(id)
+
+        t = threading.Thread(target=f)
+        t.start()
+        t.join()
+
+    def test_still_running(self):
+        main, = interpreters.list_all()
+        interp = interpreters.create()
+        with _running(interp):
+            with self.assertRaises(RuntimeError):
+                interpreters.destroy(interp)
+            self.assertTrue(interpreters.is_running(interp))
+
+
+class RunStringTests(TestBase):
+
+    SCRIPT = dedent("""
+        with open('{}', 'w') as out:
+            out.write('{}')
+        """)
+    FILENAME = 'spam'
+
+    def setUp(self):
+        super().setUp()
+        self.id = interpreters.create()
+        self._fs = None
+
+    def tearDown(self):
+        if self._fs is not None:
+            self._fs.close()
+        super().tearDown()
+
+    @property
+    def fs(self):
+        if self._fs is None:
+            self._fs = FSFixture(self)
+        return self._fs
+
+    def test_success(self):
+        script, file = _captured_script('print("it worked!", end="")')
+        with file:
+            interpreters.run_string(self.id, script)
+            out = file.read()
+
+        self.assertEqual(out, 'it worked!')
+
+    def test_in_thread(self):
+        script, file = _captured_script('print("it worked!", end="")')
+        with file:
+            def f():
+                interpreters.run_string(self.id, script)
+
+            t = threading.Thread(target=f)
+            t.start()
+            t.join()
+            out = file.read()
+
+        self.assertEqual(out, 'it worked!')
+
+    def test_create_thread(self):
+        script, file = _captured_script("""
+            import threading
+            def f():
+                print('it worked!', end='')
+
+            t = threading.Thread(target=f)
+            t.start()
+            t.join()
+            """)
+        with file:
+            interpreters.run_string(self.id, script)
+            out = file.read()
+
+        self.assertEqual(out, 'it worked!')
+
+    @unittest.skipUnless(hasattr(os, 'fork'), "test needs os.fork()")
+    def test_fork(self):
+        import tempfile
+        with tempfile.NamedTemporaryFile('w+') as file:
+            file.write('')
+            file.flush()
+
+            expected = 'spam spam spam spam spam'
+            script = dedent(f"""
+                # (inspired by Lib/test/test_fork.py)
+                import os
+                pid = os.fork()
+                if pid == 0:  # child
+                    with open('{file.name}', 'w') as out:
+                        out.write('{expected}')
+                    # Kill the unittest runner in the child process.
+                    os._exit(1)
+                else:
+                    SHORT_SLEEP = 0.1
+                    import time
+                    for _ in range(10):
+                        spid, status = os.waitpid(pid, os.WNOHANG)
+                        if spid == pid:
+                            break
+                        time.sleep(SHORT_SLEEP)
+                    assert(spid == pid)
+                """)
+            interpreters.run_string(self.id, script)
+
+            file.seek(0)
+            content = file.read()
+            self.assertEqual(content, expected)
+
+    def test_already_running(self):
+        with _running(self.id):
+            with self.assertRaises(RuntimeError):
+                interpreters.run_string(self.id, 'print("spam")')
+
+    def test_does_not_exist(self):
+        id = 0
+        while id in interpreters.list_all():
+            id += 1
+        with self.assertRaises(RuntimeError):
+            interpreters.run_string(id, 'print("spam")')
+
+    def test_error_id(self):
+        with self.assertRaises(RuntimeError):
+            interpreters.run_string(-1, 'print("spam")')
+
+    def test_bad_id(self):
+        with self.assertRaises(TypeError):
+            interpreters.run_string('spam', 'print("spam")')
+
+    def test_bad_script(self):
+        with self.assertRaises(TypeError):
+            interpreters.run_string(self.id, 10)
+
+    def test_bytes_for_script(self):
+        with self.assertRaises(TypeError):
+            interpreters.run_string(self.id, b'print("spam")')
+
+    @contextlib.contextmanager
+    def assert_run_failed(self, exctype, msg=None):
+        with self.assertRaises(interpreters.RunFailedError) as caught:
+            yield
+        if msg is None:
+            self.assertEqual(str(caught.exception).split(':')[0],
+                             str(exctype))
+        else:
+            self.assertEqual(str(caught.exception),
+                             "{}: {}".format(exctype, msg))
+
+    def test_invalid_syntax(self):
+        with self.assert_run_failed(SyntaxError):
+            # missing close paren
+            interpreters.run_string(self.id, 'print("spam"')
+
+    def test_failure(self):
+        with self.assert_run_failed(Exception, 'spam'):
+            interpreters.run_string(self.id, 'raise Exception("spam")')
+
+    def test_SystemExit(self):
+        with self.assert_run_failed(SystemExit, '42'):
+            interpreters.run_string(self.id, 'raise SystemExit(42)')
+
+    def test_sys_exit(self):
+        with self.assert_run_failed(SystemExit):
+            interpreters.run_string(self.id, dedent("""
+                import sys
+                sys.exit()
+                """))
+
+        with self.assert_run_failed(SystemExit, '42'):
+            interpreters.run_string(self.id, dedent("""
+                import sys
+                sys.exit(42)
+                """))
+
+    def test_with_shared(self):
+        r, w = os.pipe()
+
+        shared = {
+                'spam': b'ham',
+                'eggs': b'-1',
+                'cheddar': None,
+                }
+        script = dedent(f"""
+            eggs = int(eggs)
+            spam = 42
+            result = spam + eggs
+
+            ns = dict(vars())
+            del ns['__builtins__']
+            import pickle
+            with open({w}, 'wb') as chan:
+                pickle.dump(ns, chan)
+            """)
+        interpreters.run_string(self.id, script, shared)
+        with open(r, 'rb') as chan:
+            ns = pickle.load(chan)
+
+        self.assertEqual(ns['spam'], 42)
+        self.assertEqual(ns['eggs'], -1)
+        self.assertEqual(ns['result'], 41)
+        self.assertIsNone(ns['cheddar'])
+
+    def test_shared_overwrites(self):
+        interpreters.run_string(self.id, dedent("""
+            spam = 'eggs'
+            ns1 = dict(vars())
+            del ns1['__builtins__']
+            """))
+
+        shared = {'spam': b'ham'}
+        script = dedent(f"""
+            ns2 = dict(vars())
+            del ns2['__builtins__']
+        """)
+        interpreters.run_string(self.id, script, shared)
+
+        r, w = os.pipe()
+        script = dedent(f"""
+            ns = dict(vars())
+            del ns['__builtins__']
+            import pickle
+            with open({w}, 'wb') as chan:
+                pickle.dump(ns, chan)
+            """)
+        interpreters.run_string(self.id, script)
+        with open(r, 'rb') as chan:
+            ns = pickle.load(chan)
+
+        self.assertEqual(ns['ns1']['spam'], 'eggs')
+        self.assertEqual(ns['ns2']['spam'], b'ham')
+        self.assertEqual(ns['spam'], b'ham')
+
+    def test_shared_overwrites_default_vars(self):
+        r, w = os.pipe()
+
+        shared = {'__name__': b'not __main__'}
+        script = dedent(f"""
+            spam = 42
+
+            ns = dict(vars())
+            del ns['__builtins__']
+            import pickle
+            with open({w}, 'wb') as chan:
+                pickle.dump(ns, chan)
+            """)
+        interpreters.run_string(self.id, script, shared)
+        with open(r, 'rb') as chan:
+            ns = pickle.load(chan)
+
+        self.assertEqual(ns['__name__'], b'not __main__')
+
+    def test_main_reused(self):
+        r, w = os.pipe()
+        interpreters.run_string(self.id, dedent(f"""
+            spam = True
+
+            ns = dict(vars())
+            del ns['__builtins__']
+            import pickle
+            with open({w}, 'wb') as chan:
+                pickle.dump(ns, chan)
+            del ns, pickle, chan
+            """))
+        with open(r, 'rb') as chan:
+            ns1 = pickle.load(chan)
+
+        r, w = os.pipe()
+        interpreters.run_string(self.id, dedent(f"""
+            eggs = False
+
+            ns = dict(vars())
+            del ns['__builtins__']
+            import pickle
+            with open({w}, 'wb') as chan:
+                pickle.dump(ns, chan)
+            """))
+        with open(r, 'rb') as chan:
+            ns2 = pickle.load(chan)
+
+        self.assertIn('spam', ns1)
+        self.assertNotIn('eggs', ns1)
+        self.assertIn('eggs', ns2)
+        self.assertIn('spam', ns2)
+
+    def test_execution_namespace_is_main(self):
+        r, w = os.pipe()
+
+        script = dedent(f"""
+            spam = 42
+
+            ns = dict(vars())
+            ns['__builtins__'] = str(ns['__builtins__'])
+            import pickle
+            with open({w}, 'wb') as chan:
+                pickle.dump(ns, chan)
+            """)
+        interpreters.run_string(self.id, script)
+        with open(r, 'rb') as chan:
+            ns = pickle.load(chan)
+
+        ns.pop('__builtins__')
+        ns.pop('__loader__')
+        self.assertEqual(ns, {
+            '__name__': '__main__',
+            '__annotations__': {},
+            '__doc__': None,
+            '__package__': None,
+            '__spec__': None,
+            'spam': 42,
+            })
+
+    def test_still_running_at_exit(self):
+        script = dedent(f"""
+        from textwrap import dedent
+        import threading
+        import _xxsubinterpreters as _interpreters
+        def f():
+            _interpreters.run_string(id, dedent('''
+                import time
+                # Give plenty of time for the main interpreter to finish.
+                time.sleep(1_000_000)
+                '''))
+
+        t = threading.Thread(target=f)
+        t.start()
+        """)
+        with support.temp_dir() as dirname:
+            filename = script_helper.make_script(dirname, 'interp', script)
+            with script_helper.spawn_python(filename) as proc:
+                retcode = proc.wait()
+
+        self.assertEqual(retcode, 0)
+
+
+class ChannelIDTests(TestBase):
+
+    def test_default_kwargs(self):
+        cid = interpreters._channel_id(10, force=True)
+
+        self.assertEqual(int(cid), 10)
+        self.assertEqual(cid.end, 'both')
+
+    def test_with_kwargs(self):
+        cid = interpreters._channel_id(10, send=True, force=True)
+        self.assertEqual(cid.end, 'send')
+
+        cid = interpreters._channel_id(10, send=True, recv=False, force=True)
+        self.assertEqual(cid.end, 'send')
+
+        cid = interpreters._channel_id(10, recv=True, force=True)
+        self.assertEqual(cid.end, 'recv')
+
+        cid = interpreters._channel_id(10, recv=True, send=False, force=True)
+        self.assertEqual(cid.end, 'recv')
+
+        cid = interpreters._channel_id(10, send=True, recv=True, force=True)
+        self.assertEqual(cid.end, 'both')
+
+    def test_coerce_id(self):
+        cid = interpreters._channel_id('10', force=True)
+        self.assertEqual(int(cid), 10)
+
+        cid = interpreters._channel_id(10.0, force=True)
+        self.assertEqual(int(cid), 10)
+
+        class Int(str):
+            def __init__(self, value):
+                self._value = value
+            def __int__(self):
+                return self._value
+
+        cid = interpreters._channel_id(Int(10), force=True)
+        self.assertEqual(int(cid), 10)
+
+    def test_bad_id(self):
+        ids = [-1, 2**64, "spam"]
+        for cid in ids:
+            with self.subTest(cid):
+                with self.assertRaises(ValueError):
+                    interpreters._channel_id(cid)
+
+        with self.assertRaises(TypeError):
+            interpreters._channel_id(object())
+
+    def test_bad_kwargs(self):
+        with self.assertRaises(ValueError):
+            interpreters._channel_id(10, send=False, recv=False)
+
+    def test_does_not_exist(self):
+        cid = interpreters.channel_create()
+        with self.assertRaises(interpreters.ChannelNotFoundError):
+            interpreters._channel_id(int(cid) + 1)  # unforced
+
+    def test_repr(self):
+        cid = interpreters._channel_id(10, force=True)
+        self.assertEqual(repr(cid), 'ChannelID(10)')
+
+        cid = interpreters._channel_id(10, send=True, force=True)
+        self.assertEqual(repr(cid), 'ChannelID(10, send=True)')
+
+        cid = interpreters._channel_id(10, recv=True, force=True)
+        self.assertEqual(repr(cid), 'ChannelID(10, recv=True)')
+
+        cid = interpreters._channel_id(10, send=True, recv=True, force=True)
+        self.assertEqual(repr(cid), 'ChannelID(10)')
+
+    def test_equality(self):
+        cid1 = interpreters.channel_create()
+        cid2 = interpreters._channel_id(int(cid1))
+        cid3 = interpreters.channel_create()
+
+        self.assertTrue(cid1 == cid1)
+        self.assertTrue(cid1 == cid2)
+        self.assertTrue(cid1 == int(cid1))
+        self.assertFalse(cid1 == cid3)
+
+        self.assertFalse(cid1 != cid1)
+        self.assertFalse(cid1 != cid2)
+        self.assertTrue(cid1 != cid3)
+
+
+class ChannelTests(TestBase):
+
+    def test_sequential_ids(self):
+        before = interpreters.channel_list_all()
+        id1 = interpreters.channel_create()
+        id2 = interpreters.channel_create()
+        id3 = interpreters.channel_create()
+        after = interpreters.channel_list_all()
+
+        self.assertEqual(id2, int(id1) + 1)
+        self.assertEqual(id3, int(id2) + 1)
+        self.assertEqual(set(after) - set(before), {id1, id2, id3})
+
+    def test_ids_global(self):
+        id1 = interpreters.create()
+        out = _run_output(id1, dedent("""
+            import _xxsubinterpreters as _interpreters
+            cid = _interpreters.channel_create()
+            print(int(cid))
+            """))
+        cid1 = int(out.strip())
+
+        id2 = interpreters.create()
+        out = _run_output(id2, dedent("""
+            import _xxsubinterpreters as _interpreters
+            cid = _interpreters.channel_create()
+            print(int(cid))
+            """))
+        cid2 = int(out.strip())
+
+        self.assertEqual(cid2, int(cid1) + 1)
+
+    ####################
+
+    def test_drop_single_user(self):
+        cid = interpreters.channel_create()
+        interpreters.channel_send(cid, b'spam')
+        interpreters.channel_recv(cid)
+        interpreters.channel_drop_interpreter(cid, send=True, recv=True)
+
+        with self.assertRaises(interpreters.ChannelClosedError):
+            interpreters.channel_send(cid, b'eggs')
+        with self.assertRaises(interpreters.ChannelClosedError):
+            interpreters.channel_recv(cid)
+
+    def test_drop_multiple_users(self):
+        cid = interpreters.channel_create()
+        id1 = interpreters.create()
+        id2 = interpreters.create()
+        interpreters.run_string(id1, dedent(f"""
+            import _xxsubinterpreters as _interpreters
+            _interpreters.channel_send({int(cid)}, b'spam')
+            """))
+        out = _run_output(id2, dedent(f"""
+            import _xxsubinterpreters as _interpreters
+            obj = _interpreters.channel_recv({int(cid)})
+            _interpreters.channel_drop_interpreter({int(cid)})
+            print(repr(obj))
+            """))
+        interpreters.run_string(id1, dedent(f"""
+            _interpreters.channel_drop_interpreter({int(cid)})
+            """))
+
+        self.assertEqual(out.strip(), "b'spam'")
+
+    def test_drop_no_kwargs(self):
+        cid = interpreters.channel_create()
+        interpreters.channel_send(cid, b'spam')
+        interpreters.channel_recv(cid)
+        interpreters.channel_drop_interpreter(cid)
+
+        with self.assertRaises(interpreters.ChannelClosedError):
+            interpreters.channel_send(cid, b'eggs')
+        with self.assertRaises(interpreters.ChannelClosedError):
+            interpreters.channel_recv(cid)
+
+    def test_drop_multiple_times(self):
+        cid = interpreters.channel_create()
+        interpreters.channel_send(cid, b'spam')
+        interpreters.channel_recv(cid)
+        interpreters.channel_drop_interpreter(cid, send=True, recv=True)
+
+        with self.assertRaises(interpreters.ChannelClosedError):
+            interpreters.channel_drop_interpreter(cid, send=True, recv=True)
+
+    def test_drop_with_unused_items(self):
+        cid = interpreters.channel_create()
+        interpreters.channel_send(cid, b'spam')
+        interpreters.channel_send(cid, b'ham')
+        interpreters.channel_drop_interpreter(cid, send=True, recv=True)
+
+        with self.assertRaises(interpreters.ChannelClosedError):
+            interpreters.channel_recv(cid)
+
+    def test_drop_never_used(self):
+        cid = interpreters.channel_create()
+        interpreters.channel_drop_interpreter(cid)
+
+        with self.assertRaises(interpreters.ChannelClosedError):
+            interpreters.channel_send(cid, b'spam')
+        with self.assertRaises(interpreters.ChannelClosedError):
+            interpreters.channel_recv(cid)
+
+    def test_drop_by_unassociated_interp(self):
+        cid = interpreters.channel_create()
+        interpreters.channel_send(cid, b'spam')
+        interp = interpreters.create()
+        interpreters.run_string(interp, dedent(f"""
+            import _xxsubinterpreters as _interpreters
+            _interpreters.channel_drop_interpreter({int(cid)})
+            """))
+        obj = interpreters.channel_recv(cid)
+        interpreters.channel_drop_interpreter(cid)
+
+        with self.assertRaises(interpreters.ChannelClosedError):
+            interpreters.channel_send(cid, b'eggs')
+        self.assertEqual(obj, b'spam')
+
+    def test_drop_close_if_unassociated(self):
+        cid = interpreters.channel_create()
+        interp = interpreters.create()
+        interpreters.run_string(interp, dedent(f"""
+            import _xxsubinterpreters as _interpreters
+            obj = _interpreters.channel_send({int(cid)}, b'spam')
+            _interpreters.channel_drop_interpreter({int(cid)})
+            """))
+
+        with self.assertRaises(interpreters.ChannelClosedError):
+            interpreters.channel_recv(cid)
+
+    def test_drop_partially(self):
+        # XXX Is partial close too wierd/confusing?
+        cid = interpreters.channel_create()
+        interpreters.channel_send(cid, None)
+        interpreters.channel_recv(cid)
+        interpreters.channel_send(cid, b'spam')
+        interpreters.channel_drop_interpreter(cid, send=True)
+        obj = interpreters.channel_recv(cid)
+
+        self.assertEqual(obj, b'spam')
+
+    def test_drop_used_multiple_times_by_single_user(self):
+        cid = interpreters.channel_create()
+        interpreters.channel_send(cid, b'spam')
+        interpreters.channel_send(cid, b'spam')
+        interpreters.channel_send(cid, b'spam')
+        interpreters.channel_recv(cid)
+        interpreters.channel_drop_interpreter(cid, send=True, recv=True)
+
+        with self.assertRaises(interpreters.ChannelClosedError):
+            interpreters.channel_send(cid, b'eggs')
+        with self.assertRaises(interpreters.ChannelClosedError):
+            interpreters.channel_recv(cid)
+
+    ####################
+
+    def test_close_single_user(self):
+        cid = interpreters.channel_create()
+        interpreters.channel_send(cid, b'spam')
+        interpreters.channel_recv(cid)
+        interpreters.channel_close(cid)
+
+        with self.assertRaises(interpreters.ChannelClosedError):
+            interpreters.channel_send(cid, b'eggs')
+        with self.assertRaises(interpreters.ChannelClosedError):
+            interpreters.channel_recv(cid)
+
+    def test_close_multiple_users(self):
+        cid = interpreters.channel_create()
+        id1 = interpreters.create()
+        id2 = interpreters.create()
+        interpreters.run_string(id1, dedent(f"""
+            import _xxsubinterpreters as _interpreters
+            _interpreters.channel_send({int(cid)}, b'spam')
+            """))
+        interpreters.run_string(id2, dedent(f"""
+            import _xxsubinterpreters as _interpreters
+            _interpreters.channel_recv({int(cid)})
+            """))
+        interpreters.channel_close(cid)
+        with self.assertRaises(interpreters.RunFailedError) as cm:
+            interpreters.run_string(id1, dedent(f"""
+                _interpreters.channel_send({int(cid)}, b'spam')
+                """))
+        self.assertIn('ChannelClosedError', str(cm.exception))
+        with self.assertRaises(interpreters.RunFailedError) as cm:
+            interpreters.run_string(id2, dedent(f"""
+                _interpreters.channel_send({int(cid)}, b'spam')
+                """))
+        self.assertIn('ChannelClosedError', str(cm.exception))
+
+    def test_close_multiple_times(self):
+        cid = interpreters.channel_create()
+        interpreters.channel_send(cid, b'spam')
+        interpreters.channel_recv(cid)
+        interpreters.channel_close(cid)
+
+        with self.assertRaises(interpreters.ChannelClosedError):
+            interpreters.channel_close(cid)
+
+    def test_close_with_unused_items(self):
+        cid = interpreters.channel_create()
+        interpreters.channel_send(cid, b'spam')
+        interpreters.channel_send(cid, b'ham')
+        interpreters.channel_close(cid)
+
+        with self.assertRaises(interpreters.ChannelClosedError):
+            interpreters.channel_recv(cid)
+
+    def test_close_never_used(self):
+        cid = interpreters.channel_create()
+        interpreters.channel_close(cid)
+
+        with self.assertRaises(interpreters.ChannelClosedError):
+            interpreters.channel_send(cid, b'spam')
+        with self.assertRaises(interpreters.ChannelClosedError):
+            interpreters.channel_recv(cid)
+
+    def test_close_by_unassociated_interp(self):
+        cid = interpreters.channel_create()
+        interpreters.channel_send(cid, b'spam')
+        interp = interpreters.create()
+        interpreters.run_string(interp, dedent(f"""
+            import _xxsubinterpreters as _interpreters
+            _interpreters.channel_close({int(cid)})
+            """))
+        with self.assertRaises(interpreters.ChannelClosedError):
+            interpreters.channel_recv(cid)
+        with self.assertRaises(interpreters.ChannelClosedError):
+            interpreters.channel_close(cid)
+
+    def test_close_used_multiple_times_by_single_user(self):
+        cid = interpreters.channel_create()
+        interpreters.channel_send(cid, b'spam')
+        interpreters.channel_send(cid, b'spam')
+        interpreters.channel_send(cid, b'spam')
+        interpreters.channel_recv(cid)
+        interpreters.channel_close(cid)
+
+        with self.assertRaises(interpreters.ChannelClosedError):
+            interpreters.channel_send(cid, b'eggs')
+        with self.assertRaises(interpreters.ChannelClosedError):
+            interpreters.channel_recv(cid)
+
+    ####################
+
+    def test_send_recv_main(self):
+        cid = interpreters.channel_create()
+        orig = b'spam'
+        interpreters.channel_send(cid, orig)
+        obj = interpreters.channel_recv(cid)
+
+        self.assertEqual(obj, orig)
+        self.assertIsNot(obj, orig)
+
+    def test_send_recv_same_interpreter(self):
+        id1 = interpreters.create()
+        out = _run_output(id1, dedent("""
+            import _xxsubinterpreters as _interpreters
+            cid = _interpreters.channel_create()
+            orig = b'spam'
+            _interpreters.channel_send(cid, orig)
+            obj = _interpreters.channel_recv(cid)
+            assert obj is not orig
+            assert obj == orig
+            """))
+
+    def test_send_recv_different_interpreters(self):
+        cid = interpreters.channel_create()
+        id1 = interpreters.create()
+        out = _run_output(id1, dedent(f"""
+            import _xxsubinterpreters as _interpreters
+            _interpreters.channel_send({int(cid)}, b'spam')
+            """))
+        obj = interpreters.channel_recv(cid)
+
+        self.assertEqual(obj, b'spam')
+
+    def test_send_not_found(self):
+        with self.assertRaises(interpreters.ChannelNotFoundError):
+            interpreters.channel_send(10, b'spam')
+
+    def test_recv_not_found(self):
+        with self.assertRaises(interpreters.ChannelNotFoundError):
+            interpreters.channel_recv(10)
+
+    def test_recv_empty(self):
+        cid = interpreters.channel_create()
+        with self.assertRaises(interpreters.ChannelEmptyError):
+            interpreters.channel_recv(cid)
+
+    def test_run_string_arg(self):
+        cid = interpreters.channel_create()
+        interp = interpreters.create()
+
+        out = _run_output(interp, dedent("""
+            import _xxsubinterpreters as _interpreters
+            print(cid.end)
+            _interpreters.channel_send(cid, b'spam')
+            """),
+            dict(cid=cid.send))
+        obj = interpreters.channel_recv(cid)
+
+        self.assertEqual(obj, b'spam')
+        self.assertEqual(out.strip(), 'send')
+
+
+if __name__ == '__main__':
+    unittest.main()
diff --git a/Misc/NEWS.d/next/Tests/2018-01-26-21-29-09.bpo-32604.7iazNx.rst b/Misc/NEWS.d/next/Tests/2018-01-26-21-29-09.bpo-32604.7iazNx.rst
new file mode 100644
index 0000000000..f5472f9fe2
--- /dev/null
+++ b/Misc/NEWS.d/next/Tests/2018-01-26-21-29-09.bpo-32604.7iazNx.rst
@@ -0,0 +1,4 @@
+Add a new "_xxsubinterpreters" extension module that exposes the existing
+subinterpreter C-API and a new cross-interpreter data sharing mechanism. The
+module is primarily intended for more thorough testing of the existing
+subinterpreter support.
diff --git a/Modules/_xxsubinterpretersmodule.c b/Modules/_xxsubinterpretersmodule.c
new file mode 100644
index 0000000000..ed79a13d4a
--- /dev/null
+++ b/Modules/_xxsubinterpretersmodule.c
@@ -0,0 +1,2061 @@
+
+/* interpreters module */
+/* low-level access to interpreter primitives */
+
+#include "Python.h"
+#include "frameobject.h"
+#include "internal/pystate.h"
+
+
+static PyInterpreterState *
+_get_current(void)
+{
+    PyThreadState *tstate = PyThreadState_Get();
+    // PyThreadState_Get() aborts if lookup fails, so we don't need
+    // to check the result for NULL.
+    return tstate->interp;
+}
+
+static int64_t
+_coerce_id(PyObject *id)
+{
+    id = PyNumber_Long(id);
+    if (id == NULL) {
+        if (PyErr_ExceptionMatches(PyExc_TypeError)) {
+            PyErr_SetString(PyExc_TypeError,
+                            "'id' must be a non-negative int");
+        }
+        else {
+            PyErr_SetString(PyExc_ValueError,
+                            "'id' must be a non-negative int");
+        }
+        return -1;
+    }
+    long long cid = PyLong_AsLongLong(id);
+    if (cid == -1 && PyErr_Occurred() != NULL) {
+        PyErr_SetString(PyExc_ValueError,
+                        "'id' must be a non-negative int");
+        return -1;
+    }
+    if (cid < 0) {
+        PyErr_SetString(PyExc_ValueError,
+                        "'id' must be a non-negative int");
+        return -1;
+    }
+    if (cid > INT64_MAX) {
+        PyErr_SetString(PyExc_ValueError,
+                        "'id' too large (must be 64-bit int)");
+        return -1;
+    }
+    return cid;
+}
+
+/* data-sharing-specific code ***********************************************/
+
+typedef struct _shareditem {
+    Py_UNICODE *name;
+    Py_ssize_t namelen;
+    _PyCrossInterpreterData data;
+} _shareditem;
+
+void
+_sharedns_clear(_shareditem *shared)
+{
+    for (_shareditem *item=shared; item->name != NULL; item += 1) {
+        _PyCrossInterpreterData_Release(&item->data);
+    }
+}
+
+static _shareditem *
+_get_shared_ns(PyObject *shareable, Py_ssize_t *lenp)
+{
+    if (shareable == NULL || shareable == Py_None) {
+        *lenp = 0;
+        return NULL;
+    }
+    Py_ssize_t len = PyDict_Size(shareable);
+    *lenp = len;
+    if (len == 0) {
+        return NULL;
+    }
+
+    _shareditem *shared = PyMem_NEW(_shareditem, len+1);
+    if (shared == NULL) {
+        return NULL;
+    }
+    for (Py_ssize_t i=0; i < len; i++) {
+        *(shared + i) = (_shareditem){0};
+    }
+    Py_ssize_t pos = 0;
+    for (Py_ssize_t i=0; i < len; i++) {
+        PyObject *key, *value;
+        if (PyDict_Next(shareable, &pos, &key, &value) == 0) {
+            break;
+        }
+        _shareditem *item = shared + i;
+
+        if (_PyObject_GetCrossInterpreterData(value, &item->data) != 0) {
+            break;
+        }
+        item->name = PyUnicode_AsUnicodeAndSize(key, &item->namelen);
+        if (item->name == NULL) {
+            _PyCrossInterpreterData_Release(&item->data);
+            break;
+        }
+        (item + 1)->name = NULL;  // Mark the next one as the last.
+    }
+    if (PyErr_Occurred()) {
+        _sharedns_clear(shared);
+        PyMem_Free(shared);
+        return NULL;
+    }
+    return shared;
+}
+
+static int
+_shareditem_apply(_shareditem *item, PyObject *ns)
+{
+    PyObject *name = PyUnicode_FromUnicode(item->name, item->namelen);
+    if (name == NULL) {
+        return 1;
+    }
+    PyObject *value = _PyCrossInterpreterData_NewObject(&item->data);
+    if (value == NULL) {
+        Py_DECREF(name);
+        return 1;
+    }
+    int res = PyDict_SetItem(ns, name, value);
+    Py_DECREF(name);
+    Py_DECREF(value);
+    return res;
+}
+
+// Ultimately we'd like to preserve enough information about the
+// exception and traceback that we could re-constitute (or at least
+// simulate, a la traceback.TracebackException), and even chain, a copy
+// of the exception in the calling interpreter.
+
+typedef struct _sharedexception {
+    char *msg;
+} _sharedexception;
+
+static _sharedexception *
+_get_shared_exception(void)
+{
+    _sharedexception *err = PyMem_NEW(_sharedexception, 1);
+    if (err == NULL) {
+        return NULL;
+    }
+    PyObject *exc;
+    PyObject *value;
+    PyObject *tb;
+    PyErr_Fetch(&exc, &value, &tb);
+    PyObject *msg;
+    if (value == NULL) {
+        msg = PyUnicode_FromFormat("%S", exc);
+    }
+    else {
+        msg = PyUnicode_FromFormat("%S: %S", exc, value);
+    }
+    if (msg == NULL) {
+        err->msg = "unable to format exception";
+        return err;
+    }
+    err->msg = (char *)PyUnicode_AsUTF8(msg);
+    if (err->msg == NULL) {
+        err->msg = "unable to encode exception";
+    }
+    return err;
+}
+
+static PyObject * RunFailedError;
+
+static int
+interp_exceptions_init(PyObject *ns)
+{
+    // XXX Move the exceptions into per-module memory?
+
+    // An uncaught exception came out of interp_run_string().
+    RunFailedError = PyErr_NewException("_xxsubinterpreters.RunFailedError",
+                                        PyExc_RuntimeError, NULL);
+    if (RunFailedError == NULL) {
+        return -1;
+    }
+    if (PyDict_SetItemString(ns, "RunFailedError", RunFailedError) != 0) {
+        return -1;
+    }
+
+    return 0;
+}
+
+static void
+_apply_shared_exception(_sharedexception *exc)
+{
+    PyErr_SetString(RunFailedError, exc->msg);
+}
+
+/* channel-specific code */
+
+static PyObject *ChannelError;
+static PyObject *ChannelNotFoundError;
+static PyObject *ChannelClosedError;
+static PyObject *ChannelEmptyError;
+
+static int
+channel_exceptions_init(PyObject *ns)
+{
+    // XXX Move the exceptions into per-module memory?
+
+    // A channel-related operation failed.
+    ChannelError = PyErr_NewException("_xxsubinterpreters.ChannelError",
+                                      PyExc_RuntimeError, NULL);
+    if (ChannelError == NULL) {
+        return -1;
+    }
+    if (PyDict_SetItemString(ns, "ChannelError", ChannelError) != 0) {
+        return -1;
+    }
+
+    // An operation tried to use a channel that doesn't exist.
+    ChannelNotFoundError = PyErr_NewException(
+            "_xxsubinterpreters.ChannelNotFoundError", ChannelError, NULL);
+    if (ChannelNotFoundError == NULL) {
+        return -1;
+    }
+    if (PyDict_SetItemString(ns, "ChannelNotFoundError", ChannelNotFoundError) != 0) {
+        return -1;
+    }
+
+    // An operation tried to use a closed channel.
+    ChannelClosedError = PyErr_NewException(
+            "_xxsubinterpreters.ChannelClosedError", ChannelError, NULL);
+    if (ChannelClosedError == NULL) {
+        return -1;
+    }
+    if (PyDict_SetItemString(ns, "ChannelClosedError", ChannelClosedError) != 0) {
+        return -1;
+    }
+
+    // An operation tried to pop from an empty channel.
+    ChannelEmptyError = PyErr_NewException(
+            "_xxsubinterpreters.ChannelEmptyError", ChannelError, NULL);
+    if (ChannelEmptyError == NULL) {
+        return -1;
+    }
+    if (PyDict_SetItemString(ns, "ChannelEmptyError", ChannelEmptyError) != 0) {
+        return -1;
+    }
+
+    return 0;
+}
+
+struct _channelend;
+
+typedef struct _channelend {
+    struct _channelend *next;
+    int64_t interp;
+    int open;
+} _channelend;
+
+static _channelend *
+_channelend_new(int64_t interp)
+{
+    _channelend *end = PyMem_NEW(_channelend, 1);
+    if (end == NULL) {
+        return NULL;
+    }
+
+    end->next = NULL;
+    end->interp = interp;
+
+    end->open = 1;
+
+    return end;
+}
+
+static void
+_channelend_free_all(_channelend *end) {
+    while (end != NULL) {
+        _channelend *last = end;
+        end = end->next;
+        PyMem_Free(last);
+    }
+}
+
+static _channelend *
+_channelend_find(_channelend *first, int64_t interp, _channelend **pprev)
+{
+    _channelend *prev = NULL;
+    _channelend *end = first;
+    while (end != NULL) {
+        if (end->interp == interp) {
+            break;
+        }
+        prev = end;
+        end = end->next;
+    }
+    if (pprev != NULL) {
+        *pprev = prev;
+    }
+    return end;
+}
+
+struct _channelitem;
+
+typedef struct _channelitem {
+    _PyCrossInterpreterData *data;
+    struct _channelitem *next;
+} _channelitem;
+
+struct _channel;
+
+typedef struct _channel {
+    PyThread_type_lock mutex;
+
+    int open;
+
+    int64_t count;
+    _channelitem *first;
+    _channelitem *last;
+
+    // Note that the list entries are never removed for interpreter
+    // for which the channel is closed.  This should be a problem in
+    // practice.  Also, a channel isn't automatically closed when an
+    // interpreter is destroyed.
+    int64_t numsendopen;
+    int64_t numrecvopen;
+    _channelend *send;
+    _channelend *recv;
+} _PyChannelState;
+
+static _PyChannelState *
+_channel_new(void)
+{
+    _PyChannelState *chan = PyMem_NEW(_PyChannelState, 1);
+    if (chan == NULL) {
+        return NULL;
+    }
+    chan->mutex = PyThread_allocate_lock();
+    if (chan->mutex == NULL) {
+        PyMem_Free(chan);
+        PyErr_SetString(ChannelError,
+                        "can't initialize mutex for new channel");
+        return NULL;
+    }
+
+    chan->open = 1;
+
+    chan->count = 0;
+    chan->first = NULL;
+    chan->last = NULL;
+
+    chan->numsendopen = 0;
+    chan->numrecvopen = 0;
+    chan->send = NULL;
+    chan->recv = NULL;
+
+    return chan;
+}
+
+static _channelend *
+_channel_add_end(_PyChannelState *chan, _channelend *prev, int64_t interp,
+                 int send)
+{
+    _channelend *end = _channelend_new(interp);
+    if (end == NULL) {
+        return NULL;
+    }
+
+    if (prev == NULL) {
+        if (send) {
+            chan->send = end;
+        }
+        else {
+            chan->recv = end;
+        }
+    }
+    else {
+        prev->next = end;
+    }
+    if (send) {
+        chan->numsendopen += 1;
+    }
+    else {
+        chan->numrecvopen += 1;
+    }
+    return end;
+}
+
+static _channelend *
+_channel_associate_end(_PyChannelState *chan, int64_t interp, int send)
+{
+    if (!chan->open) {
+        PyErr_SetString(ChannelClosedError, "channel closed");
+        return NULL;
+    }
+
+    _channelend *prev;
+    _channelend *end = _channelend_find(send ? chan->send : chan->recv,
+                                        interp, &prev);
+    if (end != NULL) {
+        if (!end->open) {
+            PyErr_SetString(ChannelClosedError, "channel already closed");
+            return NULL;
+        }
+        // already associated
+        return end;
+    }
+    return _channel_add_end(chan, prev, interp, send);
+}
+
+static void
+_channel_close_channelend(_PyChannelState *chan, _channelend *end, int send)
+{
+    end->open = 0;
+    if (send) {
+        chan->numsendopen -= 1;
+    }
+    else {
+        chan->numrecvopen -= 1;
+    }
+}
+
+static int
+_channel_close_interpreter(_PyChannelState *chan, int64_t interp, int which)
+{
+    PyThread_acquire_lock(chan->mutex, WAIT_LOCK);
+
+    int res = -1;
+    if (!chan->open) {
+        PyErr_SetString(ChannelClosedError, "channel already closed");
+        goto done;
+    }
+
+    _channelend *prev;
+    _channelend *end;
+    if (which >= 0) {  // send/both
+        end = _channelend_find(chan->send, interp, &prev);
+        if (end == NULL) {
+            // never associated so add it
+            end = _channel_add_end(chan, prev, interp, 1);
+            if (end == NULL) {
+                goto done;
+            }
+        }
+        _channel_close_channelend(chan, end, 1);
+    }
+    if (which <= 0) {  // recv/both
+        end = _channelend_find(chan->recv, interp, &prev);
+        if (end == NULL) {
+            // never associated so add it
+            end = _channel_add_end(chan, prev, interp, 0);
+            if (end == NULL) {
+                goto done;
+            }
+        }
+        _channel_close_channelend(chan, end, 0);
+    }
+
+    if (chan->numsendopen == 0 && chan->numrecvopen == 0) {
+        if (chan->send != NULL || chan->recv != NULL) {
+            chan->open = 0;
+        }
+    }
+
+    res = 0;
+done:
+    PyThread_release_lock(chan->mutex);
+    return res;
+}
+
+static int
+_channel_close_all(_PyChannelState *chan)
+{
+    int res = -1;
+    PyThread_acquire_lock(chan->mutex, WAIT_LOCK);
+
+    if (!chan->open) {
+        PyErr_SetString(ChannelClosedError, "channel already closed");
+        goto done;
+    }
+
+    chan->open = 0;
+
+    // We *could* also just leave these in place, since we've marked
+    // the channel as closed already.
+
+    // Ensure all the "send"-associated interpreters are closed.
+    _channelend *end;
+    for (end = chan->send; end != NULL; end = end->next) {
+        _channel_close_channelend(chan, end, 1);
+    }
+
+    // Ensure all the "recv"-associated interpreters are closed.
+    for (end = chan->recv; end != NULL; end = end->next) {
+        _channel_close_channelend(chan, end, 0);
+    }
+
+    res = 0;
+done:
+    PyThread_release_lock(chan->mutex);
+    return res;
+}
+
+static int
+_channel_add(_PyChannelState *chan, int64_t interp,
+             _PyCrossInterpreterData *data)
+{
+    int res = -1;
+
+    PyThread_acquire_lock(chan->mutex, WAIT_LOCK);
+    if (_channel_associate_end(chan, interp, 1) == NULL) {
+        goto done;
+    }
+
+    _channelitem *item = PyMem_NEW(_channelitem, 1);
+    if (item == NULL) {
+        goto done;
+    }
+    item->data = data;
+    item->next = NULL;
+
+    chan->count += 1;
+    if (chan->first == NULL) {
+        chan->first = item;
+    }
+    chan->last = item;
+
+    res = 0;
+done:
+    PyThread_release_lock(chan->mutex);
+    return res;
+}
+
+static _PyCrossInterpreterData *
+_channel_next(_PyChannelState *chan, int64_t interp)
+{
+    _PyCrossInterpreterData *data = NULL;
+    PyThread_acquire_lock(chan->mutex, WAIT_LOCK);
+    if (_channel_associate_end(chan, interp, 0) == NULL) {
+        goto done;
+    }
+
+    _channelitem *item = chan->first;
+    if (item == NULL) {
+        goto done;
+    }
+    chan->first = item->next;
+    if (chan->last == item) {
+        chan->last = NULL;
+    }
+    chan->count -= 1;
+
+    data = item->data;
+    PyMem_Free(item);
+
+done:
+    PyThread_release_lock(chan->mutex);
+    return data;
+}
+
+static void
+_channel_clear(_PyChannelState *chan)
+{
+    _channelitem *item = chan->first;
+    while (item != NULL) {
+        _PyCrossInterpreterData_Release(item->data);
+        PyMem_Free(item->data);
+        _channelitem *last = item;
+        item = item->next;
+        PyMem_Free(last);
+    }
+    chan->first = NULL;
+    chan->last = NULL;
+}
+
+static void
+_channel_free(_PyChannelState *chan)
+{
+    PyThread_acquire_lock(chan->mutex, WAIT_LOCK);
+    _channel_clear(chan);
+    _channelend_free_all(chan->send);
+    _channelend_free_all(chan->recv);
+    PyThread_release_lock(chan->mutex);
+
+    PyThread_free_lock(chan->mutex);
+    PyMem_Free(chan);
+}
+
+struct _channelref;
+
+typedef struct _channelref {
+    int64_t id;
+    _PyChannelState *chan;
+    struct _channelref *next;
+    Py_ssize_t objcount;
+} _channelref;
+
+static _channelref *
+_channelref_new(int64_t id, _PyChannelState *chan)
+{
+    _channelref *ref = PyMem_NEW(_channelref, 1);
+    if (ref == NULL) {
+        return NULL;
+    }
+    ref->id = id;
+    ref->chan = chan;
+    ref->next = NULL;
+    ref->objcount = 0;
+    return ref;
+}
+
+static _channelref *
+_channelref_find(_channelref *first, int64_t id, _channelref **pprev)
+{
+    _channelref *prev = NULL;
+    _channelref *ref = first;
+    while (ref != NULL) {
+        if (ref->id == id) {
+            break;
+        }
+        prev = ref;
+        ref = ref->next;
+    }
+    if (pprev != NULL) {
+        *pprev = prev;
+    }
+    return ref;
+}
+
+typedef struct _channels {
+    PyThread_type_lock mutex;
+    _channelref *head;
+    int64_t numopen;
+    int64_t next_id;
+} _channels;
+
+static int
+_channels_init(_channels *channels)
+{
+    if (channels->mutex == NULL) {
+        channels->mutex = PyThread_allocate_lock();
+        if (channels->mutex == NULL) {
+            PyMem_Free(channels);
+            PyErr_SetString(ChannelError,
+                            "can't initialize mutex for channel management");
+            return -1;
+        }
+    }
+    channels->head = NULL;
+    channels->numopen = 0;
+    channels->next_id = 0;
+    return 0;
+}
+
+static int64_t
+_channels_next_id(_channels *channels)  // needs lock
+{
+    int64_t id = channels->next_id;
+    if (id < 0) {
+        /* overflow */
+        PyErr_SetString(ChannelError,
+                        "failed to get a channel ID");
+        return -1;
+    }
+    channels->next_id += 1;
+    return id;
+}
+
+static _PyChannelState *
+_channels_lookup(_channels *channels, int64_t id, PyThread_type_lock *pmutex)
+{
+    _PyChannelState *chan = NULL;
+    PyThread_acquire_lock(channels->mutex, WAIT_LOCK);
+    if (pmutex != NULL) {
+        *pmutex = NULL;
+    }
+
+    _channelref *ref = _channelref_find(channels->head, id, NULL);
+    if (ref == NULL) {
+        PyErr_Format(ChannelNotFoundError, "channel %d not found", id);
+        goto done;
+    }
+    if (ref->chan == NULL || !ref->chan->open) {
+        PyErr_Format(ChannelClosedError, "channel %d closed", id);
+        goto done;
+    }
+
+    if (pmutex != NULL) {
+        // The mutex will be closed by the caller.
+        *pmutex = channels->mutex;
+    }
+
+    chan = ref->chan;
+done:
+    if (pmutex == NULL || *pmutex == NULL) {
+        PyThread_release_lock(channels->mutex);
+    }
+    return chan;
+}
+
+static int64_t
+_channels_add(_channels *channels, _PyChannelState *chan)
+{
+    int64_t cid = -1;
+    PyThread_acquire_lock(channels->mutex, WAIT_LOCK);
+
+    // Create a new ref.
+    int64_t id = _channels_next_id(channels);
+    if (id < 0) {
+        goto done;
+    }
+    _channelref *ref = _channelref_new(id, chan);
+    if (ref == NULL) {
+        goto done;
+    }
+
+    // Add it to the list.
+    // We assume that the channel is a new one (not already in the list).
+    ref->next = channels->head;
+    channels->head = ref;
+    channels->numopen += 1;
+
+    cid = id;
+done:
+    PyThread_release_lock(channels->mutex);
+    return cid;
+}
+
+static int
+_channels_close(_channels *channels, int64_t cid, _PyChannelState **pchan)
+{
+    int res = -1;
+    PyThread_acquire_lock(channels->mutex, WAIT_LOCK);
+    if (pchan != NULL) {
+        *pchan = NULL;
+    }
+
+    _channelref *ref = _channelref_find(channels->head, cid, NULL);
+    if (ref == NULL) {
+        PyErr_Format(ChannelNotFoundError, "channel %d not found", cid);
+        goto done;
+    }
+
+    if (ref->chan == NULL) {
+        PyErr_Format(ChannelClosedError, "channel %d closed", cid);
+        goto done;
+    }
+    else {
+        if (_channel_close_all(ref->chan) != 0) {
+            goto done;
+        }
+        if (pchan != NULL) {
+            *pchan = ref->chan;
+        }
+        ref->chan = NULL;
+    }
+
+    res = 0;
+done:
+    PyThread_release_lock(channels->mutex);
+    return res;
+}
+
+static void
+_channels_remove_ref(_channels *channels, _channelref *ref, _channelref *prev,
+                     _PyChannelState **pchan)
+{
+    if (ref == channels->head) {
+        channels->head = ref->next;
+    }
+    else {
+        prev->next = ref->next;
+    }
+    channels->numopen -= 1;
+
+    if (pchan != NULL) {
+        *pchan = ref->chan;
+    }
+    PyMem_Free(ref);
+}
+
+static int
+_channels_remove(_channels *channels, int64_t id, _PyChannelState **pchan)
+{
+    int res = -1;
+    PyThread_acquire_lock(channels->mutex, WAIT_LOCK);
+
+    if (pchan != NULL) {
+        *pchan = NULL;
+    }
+
+    _channelref *prev = NULL;
+    _channelref *ref = _channelref_find(channels->head, id, &prev);
+    if (ref == NULL) {
+        PyErr_Format(ChannelNotFoundError, "channel %d not found", id);
+        goto done;
+    }
+
+    _channels_remove_ref(channels, ref, prev, pchan);
+
+    res = 0;
+done:
+    PyThread_release_lock(channels->mutex);
+    return res;
+}
+
+static int
+_channels_add_id_object(_channels *channels, int64_t id)
+{
+    int res = -1;
+    PyThread_acquire_lock(channels->mutex, WAIT_LOCK);
+
+    _channelref *ref = _channelref_find(channels->head, id, NULL);
+    if (ref == NULL) {
+        PyErr_Format(ChannelNotFoundError, "channel %d not found", id);
+        goto done;
+    }
+    ref->objcount += 1;
+
+    res = 0;
+done:
+    PyThread_release_lock(channels->mutex);
+    return res;
+}
+
+static void
+_channels_drop_id_object(_channels *channels, int64_t id)
+{
+    PyThread_acquire_lock(channels->mutex, WAIT_LOCK);
+
+    _channelref *prev = NULL;
+    _channelref *ref = _channelref_find(channels->head, id, &prev);
+    if (ref == NULL) {
+        // Already destroyed.
+        goto done;
+    }
+    ref->objcount -= 1;
+
+    // Destroy if no longer used.
+    if (ref->objcount == 0) {
+        _PyChannelState *chan = NULL;
+        _channels_remove_ref(channels, ref, prev, &chan);
+        if (chan != NULL) {
+            _channel_free(chan);
+        }
+    }
+
+done:
+    PyThread_release_lock(channels->mutex);
+}
+
+int64_t *
+_channels_list_all(_channels *channels, int64_t *count)
+{
+    int64_t *cids = NULL;
+    PyThread_acquire_lock(channels->mutex, WAIT_LOCK);
+    int64_t numopen = channels->numopen;
+    if (numopen >= PY_SSIZE_T_MAX) {
+        PyErr_SetString(PyExc_RuntimeError, "too many channels open");
+        goto done;
+    }
+    int64_t *ids = PyMem_NEW(int64_t, (Py_ssize_t)(channels->numopen));
+    if (ids == NULL) {
+        goto done;
+    }
+    _channelref *ref = channels->head;
+    for (int64_t i=0; ref != NULL; ref = ref->next, i++) {
+        ids[i] = ref->id;
+    }
+    *count = channels->numopen;
+
+    cids = ids;
+done:
+    PyThread_release_lock(channels->mutex);
+    return cids;
+}
+
+/* "high"-level channel-related functions */
+
+static int64_t
+_channel_create(_channels *channels)
+{
+    _PyChannelState *chan = _channel_new();
+    if (chan == NULL) {
+        return -1;
+    }
+    int64_t id = _channels_add(channels, chan);
+    if (id < 0) {
+        _channel_free(chan);
+        return -1;
+    }
+    return id;
+}
+
+static int
+_channel_destroy(_channels *channels, int64_t id)
+{
+    _PyChannelState *chan = NULL;
+    if (_channels_remove(channels, id, &chan) != 0) {
+        return -1;
+    }
+    if (chan != NULL) {
+        _channel_free(chan);
+    }
+    return 0;
+}
+
+static int
+_channel_send(_channels *channels, int64_t id, PyObject *obj)
+{
+    PyInterpreterState *interp = _get_current();
+    if (interp == NULL) {
+        return -1;
+    }
+
+    // Look up the channel.
+    PyThread_type_lock mutex = NULL;
+    _PyChannelState *chan = _channels_lookup(channels, id, &mutex);
+    if (chan == NULL) {
+        return -1;
+    }
+    // Past this point we are responsible for releasing the mutex.
+
+    // Convert the object to cross-interpreter data.
+    _PyCrossInterpreterData *data = PyMem_NEW(_PyCrossInterpreterData, 1);
+    if (data == NULL) {
+        PyThread_release_lock(mutex);
+        return -1;
+    }
+    if (_PyObject_GetCrossInterpreterData(obj, data) != 0) {
+        PyThread_release_lock(mutex);
+        return -1;
+    }
+
+    // Add the data to the channel.
+    int res = _channel_add(chan, interp->id, data);
+    PyThread_release_lock(mutex);
+    if (res != 0) {
+        _PyCrossInterpreterData_Release(data);
+        PyMem_Free(data);
+        return -1;
+    }
+
+    return 0;
+}
+
+static PyObject *
+_channel_recv(_channels *channels, int64_t id)
+{
+    PyInterpreterState *interp = _get_current();
+    if (interp == NULL) {
+        return NULL;
+    }
+
+    // Look up the channel.
+    PyThread_type_lock mutex = NULL;
+    _PyChannelState *chan = _channels_lookup(channels, id, &mutex);
+    if (chan == NULL) {
+        return NULL;
+    }
+    // Past this point we are responsible for releasing the mutex.
+
+    // Pop off the next item from the channel.
+    _PyCrossInterpreterData *data = _channel_next(chan, interp->id);
+    PyThread_release_lock(mutex);
+    if (data == NULL) {
+        PyErr_Format(ChannelEmptyError, "channel %d is empty", id);
+        return NULL;
+    }
+
+    // Convert the data back to an object.
+    PyObject *obj = _PyCrossInterpreterData_NewObject(data);
+    if (obj == NULL) {
+        return NULL;
+    }
+    _PyCrossInterpreterData_Release(data);
+
+    return obj;
+}
+
+static int
+_channel_drop(_channels *channels, int64_t id, int send, int recv)
+{
+    PyInterpreterState *interp = _get_current();
+    if (interp == NULL) {
+        return -1;
+    }
+
+    // Look up the channel.
+    PyThread_type_lock mutex = NULL;
+    _PyChannelState *chan = _channels_lookup(channels, id, &mutex);
+    if (chan == NULL) {
+        return -1;
+    }
+    // Past this point we are responsible for releasing the mutex.
+
+    // Close one or both of the two ends.
+    int res =_channel_close_interpreter(chan, interp->id, send-recv);
+    PyThread_release_lock(mutex);
+    return res;
+}
+
+static int
+_channel_close(_channels *channels, int64_t id)
+{
+    return _channels_close(channels, id, NULL);
+}
+
+/* ChannelID class */
+
+#define CHANNEL_SEND 1
+#define CHANNEL_RECV -1
+
+static PyTypeObject ChannelIDtype;
+
+typedef struct channelid {
+    PyObject_HEAD
+    int64_t id;
+    int end;
+    _channels *channels;
+} channelid;
+
+static channelid *
+newchannelid(PyTypeObject *cls, int64_t cid, int end, _channels *channels,
+             int force)
+{
+    channelid *self = PyObject_New(channelid, cls);
+    if (self == NULL) {
+        return NULL;
+    }
+    self->id = cid;
+    self->end = end;
+    self->channels = channels;
+
+    if (_channels_add_id_object(channels, cid) != 0) {
+        if (force && PyErr_ExceptionMatches(ChannelNotFoundError)) {
+            PyErr_Clear();
+        }
+        else {
+            Py_DECREF((PyObject *)self);
+            return NULL;
+        }
+    }
+
+    return self;
+}
+
+static _channels * _global_channels(void);
+
+static PyObject *
+channelid_new(PyTypeObject *cls, PyObject *args, PyObject *kwds)
+{
+    static char *kwlist[] = {"id", "send", "recv", "force", NULL};
+    PyObject *id;
+    int send = -1;
+    int recv = -1;
+    int force = 0;
+    if (!PyArg_ParseTupleAndKeywords(args, kwds,
+                                     "O|$ppp:ChannelID.__init__", kwlist,
+                                     &id, &send, &recv, &force))
+        return NULL;
+
+    // Coerce and check the ID.
+    int64_t cid;
+    if (PyObject_TypeCheck(id, &ChannelIDtype)) {
+        cid = ((channelid *)id)->id;
+    }
+    else {
+        cid = _coerce_id(id);
+        if (cid < 0) {
+            return NULL;
+        }
+    }
+
+    // Handle "send" and "recv".
+    if (send == 0 && recv == 0) {
+        PyErr_SetString(PyExc_ValueError,
+                        "'send' and 'recv' cannot both be False");
+        return NULL;
+    }
+    int end = 0;
+    if (send == 1) {
+        if (recv == 0 || recv == -1) {
+            end = CHANNEL_SEND;
+        }
+    }
+    else if (recv == 1) {
+        end = CHANNEL_RECV;
+    }
+
+    return (PyObject *)newchannelid(cls, cid, end, _global_channels(), force);
+}
+
+static void
+channelid_dealloc(PyObject *v)
+{
+    int64_t cid = ((channelid *)v)->id;
+    _channels *channels = ((channelid *)v)->channels;
+    Py_TYPE(v)->tp_free(v);
+
+    _channels_drop_id_object(channels, cid);
+}
+
+static PyObject *
+channelid_repr(PyObject *self)
+{
+    PyTypeObject *type = Py_TYPE(self);
+    const char *name = _PyType_Name(type);
+
+    channelid *cid = (channelid *)self;
+    const char *fmt;
+    if (cid->end == CHANNEL_SEND) {
+        fmt = "%s(%d, send=True)";
+    }
+    else if (cid->end == CHANNEL_RECV) {
+        fmt = "%s(%d, recv=True)";
+    }
+    else {
+        fmt = "%s(%d)";
+    }
+    return PyUnicode_FromFormat(fmt, name, cid->id);
+}
+
+PyObject *
+channelid_int(PyObject *self)
+{
+    channelid *cid = (channelid *)self;
+    return PyLong_FromLongLong(cid->id);
+}
+
+static PyNumberMethods channelid_as_number = {
+     0,                        /* nb_add */
+     0,                        /* nb_subtract */
+     0,                        /* nb_multiply */
+     0,                        /* nb_remainder */
+     0,                        /* nb_divmod */
+     0,                        /* nb_power */
+     0,                        /* nb_negative */
+     0,                        /* nb_positive */
+     0,                        /* nb_absolute */
+     0,                        /* nb_bool */
+     0,                        /* nb_invert */
+     0,                        /* nb_lshift */
+     0,                        /* nb_rshift */
+     0,                        /* nb_and */
+     0,                        /* nb_xor */
+     0,                        /* nb_or */
+     (unaryfunc)channelid_int, /* nb_int */
+     0,                        /* nb_reserved */
+     0,                        /* nb_float */
+
+     0,                        /* nb_inplace_add */
+     0,                        /* nb_inplace_subtract */
+     0,                        /* nb_inplace_multiply */
+     0,                        /* nb_inplace_remainder */
+     0,                        /* nb_inplace_power */
+     0,                        /* nb_inplace_lshift */
+     0,                        /* nb_inplace_rshift */
+     0,                        /* nb_inplace_and */
+     0,                        /* nb_inplace_xor */
+     0,                        /* nb_inplace_or */
+
+     0,                        /* nb_floor_divide */
+     0,                        /* nb_true_divide */
+     0,                        /* nb_inplace_floor_divide */
+     0,                        /* nb_inplace_true_divide */
+
+     (unaryfunc)channelid_int, /* nb_index */
+};
+
+static Py_hash_t
+channelid_hash(PyObject *self)
+{
+    channelid *cid = (channelid *)self;
+    PyObject *id = PyLong_FromLongLong(cid->id);
+    if (id == NULL) {
+        return -1;
+    }
+    return PyObject_Hash(id);
+}
+
+static PyObject *
+channelid_richcompare(PyObject *self, PyObject *other, int op)
+{
+    if (op != Py_EQ && op != Py_NE) {
+        Py_RETURN_NOTIMPLEMENTED;
+    }
+
+    if (!PyObject_TypeCheck(self, &ChannelIDtype)) {
+        Py_RETURN_NOTIMPLEMENTED;
+    }
+
+    channelid *cid = (channelid *)self;
+    int equal;
+    if (PyObject_TypeCheck(other, &ChannelIDtype)) {
+        channelid *othercid = (channelid *)other;
+        if (cid->end != othercid->end) {
+            equal = 0;
+        }
+        else {
+            equal = (cid->id == othercid->id);
+        }
+    }
+    else {
+        other = PyNumber_Long(other);
+        if (other == NULL) {
+            PyErr_Clear();
+            Py_RETURN_NOTIMPLEMENTED;
+        }
+        int64_t othercid = PyLong_AsLongLong(other);
+        // XXX decref other here?
+        if (othercid == -1 && PyErr_Occurred() != NULL) {
+            return NULL;
+        }
+        if (othercid < 0 || othercid > INT64_MAX) {
+            equal = 0;
+        }
+        else {
+            equal = (cid->id == othercid);
+        }
+    }
+
+    if ((op == Py_EQ && equal) || (op == Py_NE && !equal)) {
+        Py_RETURN_TRUE;
+    }
+    Py_RETURN_FALSE;
+}
+
+struct _channelid_xid {
+    int64_t id;
+    int end;
+};
+
+static PyObject *
+_channelid_from_xid(_PyCrossInterpreterData *data)
+{
+    struct _channelid_xid *xid = (struct _channelid_xid *)data->data;
+    return (PyObject *)newchannelid(&ChannelIDtype, xid->id, xid->end,
+                                    _global_channels(), 0);
+}
+
+static int
+_channelid_shared(PyObject *obj, _PyCrossInterpreterData *data)
+{
+    struct _channelid_xid *xid = PyMem_NEW(struct _channelid_xid, 1);
+    if (xid == NULL) {
+        return -1;
+    }
+    xid->id = ((channelid *)obj)->id;
+    xid->end = ((channelid *)obj)->end;
+
+    data->data = xid;
+    data->obj = obj;
+    data->new_object = _channelid_from_xid;
+    data->free = PyMem_Free;
+    return 0;
+}
+
+static PyObject *
+channelid_end(PyObject *self, void *end)
+{
+    int force = 1;
+    channelid *cid = (channelid *)self;
+    if (end != NULL) {
+        return (PyObject *)newchannelid(Py_TYPE(self), cid->id, *(int *)end,
+                                        cid->channels, force);
+    }
+
+    if (cid->end == CHANNEL_SEND) {
+        return PyUnicode_InternFromString("send");
+    }
+    if (cid->end == CHANNEL_RECV) {
+        return PyUnicode_InternFromString("recv");
+    }
+    return PyUnicode_InternFromString("both");
+}
+
+static int _channelid_end_send = CHANNEL_SEND;
+static int _channelid_end_recv = CHANNEL_RECV;
+
+static PyGetSetDef channelid_getsets[] = {
+    {"end", (getter)channelid_end, NULL,
+     PyDoc_STR("'send', 'recv', or 'both'")},
+    {"send", (getter)channelid_end, NULL,
+     PyDoc_STR("the 'send' end of the channel"), &_channelid_end_send},
+    {"recv", (getter)channelid_end, NULL,
+     PyDoc_STR("the 'recv' end of the channel"), &_channelid_end_recv},
+    {NULL}
+};
+
+PyDoc_STRVAR(channelid_doc,
+"A channel ID identifies a channel and may be used as an int.");
+
+static PyTypeObject ChannelIDtype = {
+    PyVarObject_HEAD_INIT(&PyType_Type, 0)
+    "_xxsubinterpreters.ChannelID", /* tp_name */
+    sizeof(channelid),              /* tp_size */
+    0,                              /* tp_itemsize */
+    (destructor)channelid_dealloc,  /* tp_dealloc */
+    0,                              /* tp_print */
+    0,                              /* tp_getattr */
+    0,                              /* tp_setattr */
+    0,                              /* tp_as_async */
+    (reprfunc)channelid_repr,       /* tp_repr */
+    &channelid_as_number,           /* tp_as_number */
+    0,                              /* tp_as_sequence */
+    0,                              /* tp_as_mapping */
+    channelid_hash,                 /* tp_hash */
+    0,                              /* tp_call */
+    0,                              /* tp_str */
+    0,                              /* tp_getattro */
+    0,                              /* tp_setattro */
+    0,                              /* tp_as_buffer */
+    Py_TPFLAGS_DEFAULT | Py_TPFLAGS_BASETYPE |
+        Py_TPFLAGS_LONG_SUBCLASS,   /* tp_flags */
+    channelid_doc,                  /* tp_doc */
+    0,                              /* tp_traverse */
+    0,                              /* tp_clear */
+    channelid_richcompare,          /* tp_richcompare */
+    0,                              /* tp_weaklistoffset */
+    0,                              /* tp_iter */
+    0,                              /* tp_iternext */
+    0,                              /* tp_methods */
+    0,                              /* tp_members */
+    channelid_getsets,              /* tp_getset */
+    0,                              /* tp_base */
+    0,                              /* tp_dict */
+    0,                              /* tp_descr_get */
+    0,                              /* tp_descr_set */
+    0,                              /* tp_dictoffset */
+    0,                              /* tp_init */
+    0,                              /* tp_alloc */
+    // Note that we do not set tp_new to channelid_new.  Instead we
+    // set it to NULL, meaning it cannot be instantiated from Python
+    // code.  We do this because there is a strong relationship between
+    // channel IDs and the channel lifecycle, so this limitation avoids
+    // related complications.
+    NULL,                           /* tp_new */
+};
+
+/* interpreter-specific functions *******************************************/
+
+static PyInterpreterState *
+_look_up(PyObject *requested_id)
+{
+    long long id = PyLong_AsLongLong(requested_id);
+    if (id == -1 && PyErr_Occurred() != NULL) {
+        return NULL;
+    }
+    assert(id <= INT64_MAX);
+    return _PyInterpreterState_LookUpID(id);
+}
+
+static PyObject *
+_get_id(PyInterpreterState *interp)
+{
+    PY_INT64_T id = PyInterpreterState_GetID(interp);
+    if (id < 0) {
+        return NULL;
+    }
+    return PyLong_FromLongLong(id);
+}
+
+static int
+_is_running(PyInterpreterState *interp)
+{
+    PyThreadState *tstate = PyInterpreterState_ThreadHead(interp);
+    if (PyThreadState_Next(tstate) != NULL) {
+        PyErr_SetString(PyExc_RuntimeError,
+                        "interpreter has more than one thread");
+        return -1;
+    }
+    PyFrameObject *frame = tstate->frame;
+    if (frame == NULL) {
+        if (PyErr_Occurred() != NULL) {
+            return -1;
+        }
+        return 0;
+    }
+    return (int)(frame->f_executing);
+}
+
+static int
+_ensure_not_running(PyInterpreterState *interp)
+{
+    int is_running = _is_running(interp);
+    if (is_running < 0) {
+        return -1;
+    }
+    if (is_running) {
+        PyErr_Format(PyExc_RuntimeError, "interpreter already running");
+        return -1;
+    }
+    return 0;
+}
+
+static int
+_run_script(PyInterpreterState *interp, const char *codestr,
+            _shareditem *shared, Py_ssize_t num_shared,
+            _sharedexception **exc)
+{
+    assert(num_shared >= 0);
+    PyObject *main_mod = PyMapping_GetItemString(interp->modules, "__main__");
+    if (main_mod == NULL) {
+        goto error;
+    }
+    PyObject *ns = PyModule_GetDict(main_mod);  // borrowed
+    Py_DECREF(main_mod);
+    if (ns == NULL) {
+        goto error;
+    }
+    Py_INCREF(ns);
+
+    // Apply the cross-interpreter data.
+    if (shared != NULL) {
+        for (Py_ssize_t i=0; i < num_shared; i++) {
+            _shareditem *item = &shared[i];
+            if (_shareditem_apply(item, ns) != 0) {
+                Py_DECREF(ns);
+                goto error;
+            }
+        }
+    }
+
+    // Run the string (see PyRun_SimpleStringFlags).
+    PyObject *result = PyRun_StringFlags(codestr, Py_file_input, ns, ns, NULL);
+    Py_DECREF(ns);
+    if (result == NULL) {
+        goto error;
+    }
+    else {
+        Py_DECREF(result);  // We throw away the result.
+    }
+
+    return 0;
+
+error:
+    *exc = _get_shared_exception();
+    PyErr_Clear();
+    return -1;
+}
+
+static int
+_run_script_in_interpreter(PyInterpreterState *interp, const char *codestr,
+                           PyObject *shareables)
+{
+    if (_ensure_not_running(interp) < 0) {
+        return -1;
+    }
+
+    Py_ssize_t num_shared = -1;
+    _shareditem *shared = _get_shared_ns(shareables, &num_shared);
+    if (shared == NULL && PyErr_Occurred()) {
+        return -1;
+    }
+
+    // Switch to interpreter.
+    PyThreadState *tstate = PyInterpreterState_ThreadHead(interp);
+    PyThreadState *save_tstate = PyThreadState_Swap(tstate);
+
+    // Run the script.
+    _sharedexception *exc = NULL;
+    int result = _run_script(interp, codestr, shared, num_shared, &exc);
+
+    // Switch back.
+    if (save_tstate != NULL) {
+        PyThreadState_Swap(save_tstate);
+    }
+
+    // Propagate any exception out to the caller.
+    if (exc != NULL) {
+        _apply_shared_exception(exc);
+        PyMem_Free(exc);
+    }
+    else if (result != 0) {
+        // We were unable to allocate a shared exception.
+        PyErr_NoMemory();
+    }
+
+    if (shared != NULL) {
+        _sharedns_clear(shared);
+        PyMem_Free(shared);
+    }
+
+    return result;
+}
+
+
+/* module level code ********************************************************/
+
+/* globals is the process-global state for the module.  It holds all
+   the data that we need to share between interpreters, so it cannot
+   hold PyObject values. */
+static struct globals {
+    _channels channels;
+} _globals = {{0}};
+
+static int
+_init_globals(void)
+{
+    if (_channels_init(&_globals.channels) != 0) {
+        return -1;
+    }
+    return 0;
+}
+
+static _channels *
+_global_channels(void) {
+    return &_globals.channels;
+}
+
+static PyObject *
+interp_create(PyObject *self, PyObject *args)
+{
+    if (!PyArg_UnpackTuple(args, "create", 0, 0)) {
+        return NULL;
+    }
+
+    // Create and initialize the new interpreter.
+    PyThreadState *tstate, *save_tstate;
+    save_tstate = PyThreadState_Swap(NULL);
+    tstate = Py_NewInterpreter();
+    PyThreadState_Swap(save_tstate);
+    if (tstate == NULL) {
+        /* Since no new thread state was created, there is no exception to
+           propagate; raise a fresh one after swapping in the old thread
+           state. */
+        PyErr_SetString(PyExc_RuntimeError, "interpreter creation failed");
+        return NULL;
+    }
+    return _get_id(tstate->interp);
+}
+
+PyDoc_STRVAR(create_doc,
+"create() -> ID\n\
+\n\
+Create a new interpreter and return a unique generated ID.");
+
+
+static PyObject *
+interp_destroy(PyObject *self, PyObject *args)
+{
+    PyObject *id;
+    if (!PyArg_UnpackTuple(args, "destroy", 1, 1, &id)) {
+        return NULL;
+    }
+    if (!PyLong_Check(id)) {
+        PyErr_SetString(PyExc_TypeError, "ID must be an int");
+        return NULL;
+    }
+
+    // Look up the interpreter.
+    PyInterpreterState *interp = _look_up(id);
+    if (interp == NULL) {
+        return NULL;
+    }
+
+    // Ensure we don't try to destroy the current interpreter.
+    PyInterpreterState *current = _get_current();
+    if (current == NULL) {
+        return NULL;
+    }
+    if (interp == current) {
+        PyErr_SetString(PyExc_RuntimeError,
+                        "cannot destroy the current interpreter");
+        return NULL;
+    }
+
+    // Ensure the interpreter isn't running.
+    /* XXX We *could* support destroying a running interpreter but
+       aren't going to worry about it for now. */
+    if (_ensure_not_running(interp) < 0) {
+        return NULL;
+    }
+
+    // Destroy the interpreter.
+    //PyInterpreterState_Delete(interp);
+    PyThreadState *tstate, *save_tstate;
+    tstate = PyInterpreterState_ThreadHead(interp);
+    save_tstate = PyThreadState_Swap(tstate);
+    Py_EndInterpreter(tstate);
+    PyThreadState_Swap(save_tstate);
+
+    Py_RETURN_NONE;
+}
+
+PyDoc_STRVAR(destroy_doc,
+"destroy(ID)\n\
+\n\
+Destroy the identified interpreter.\n\
+\n\
+Attempting to destroy the current interpreter results in a RuntimeError.\n\
+So does an unrecognized ID.");
+
+
+static PyObject *
+interp_list_all(PyObject *self)
+{
+    PyObject *ids, *id;
+    PyInterpreterState *interp;
+
+    ids = PyList_New(0);
+    if (ids == NULL) {
+        return NULL;
+    }
+
+    interp = PyInterpreterState_Head();
+    while (interp != NULL) {
+        id = _get_id(interp);
+        if (id == NULL) {
+            Py_DECREF(ids);
+            return NULL;
+        }
+        // insert at front of list
+        if (PyList_Insert(ids, 0, id) < 0) {
+            Py_DECREF(ids);
+            return NULL;
+        }
+
+        interp = PyInterpreterState_Next(interp);
+    }
+
+    return ids;
+}
+
+PyDoc_STRVAR(list_all_doc,
+"list_all() -> [ID]\n\
+\n\
+Return a list containing the ID of every existing interpreter.");
+
+
+static PyObject *
+interp_get_current(PyObject *self)
+{
+    PyInterpreterState *interp =_get_current();
+    if (interp == NULL) {
+        return NULL;
+    }
+    return _get_id(interp);
+}
+
+PyDoc_STRVAR(get_current_doc,
+"get_current() -> ID\n\
+\n\
+Return the ID of current interpreter.");
+
+
+static PyObject *
+interp_get_main(PyObject *self)
+{
+    // Currently, 0 is always the main interpreter.
+    return PyLong_FromLongLong(0);
+}
+
+PyDoc_STRVAR(get_main_doc,
+"get_main() -> ID\n\
+\n\
+Return the ID of main interpreter.");
+
+
+static PyObject *
+interp_run_string(PyObject *self, PyObject *args)
+{
+    PyObject *id, *code;
+    PyObject *shared = NULL;
+    if (!PyArg_UnpackTuple(args, "run_string", 2, 3, &id, &code, &shared)) {
+        return NULL;
+    }
+    if (!PyLong_Check(id)) {
+        PyErr_SetString(PyExc_TypeError, "first arg (ID) must be an int");
+        return NULL;
+    }
+    if (!PyUnicode_Check(code)) {
+        PyErr_SetString(PyExc_TypeError,
+                        "second arg (code) must be a string");
+        return NULL;
+    }
+
+    // Look up the interpreter.
+    PyInterpreterState *interp = _look_up(id);
+    if (interp == NULL) {
+        return NULL;
+    }
+
+    // Extract code.
+    Py_ssize_t size;
+    const char *codestr = PyUnicode_AsUTF8AndSize(code, &size);
+    if (codestr == NULL) {
+        return NULL;
+    }
+    if (strlen(codestr) != (size_t)size) {
+        PyErr_SetString(PyExc_ValueError,
+                        "source code string cannot contain null bytes");
+        return NULL;
+    }
+
+    // Run the code in the interpreter.
+    if (_run_script_in_interpreter(interp, codestr, shared) != 0) {
+        return NULL;
+    }
+    Py_RETURN_NONE;
+}
+
+PyDoc_STRVAR(run_string_doc,
+"run_string(ID, sourcetext)\n\
+\n\
+Execute the provided string in the identified interpreter.\n\
+\n\
+See PyRun_SimpleStrings.");
+
+
+static PyObject *
+object_is_shareable(PyObject *self, PyObject *args)
+{
+    PyObject *obj;
+    if (!PyArg_UnpackTuple(args, "is_shareable", 1, 1, &obj)) {
+        return NULL;
+    }
+    if (_PyObject_CheckCrossInterpreterData(obj) == 0) {
+        Py_RETURN_TRUE;
+    }
+    PyErr_Clear();
+    Py_RETURN_FALSE;
+}
+
+PyDoc_STRVAR(is_shareable_doc,
+"is_shareable(obj) -> bool\n\
+\n\
+Return True if the object's data may be shared between interpreters and\n\
+False otherwise.");
+
+
+static PyObject *
+interp_is_running(PyObject *self, PyObject *args)
+{
+    PyObject *id;
+    if (!PyArg_UnpackTuple(args, "is_running", 1, 1, &id)) {
+        return NULL;
+    }
+    if (!PyLong_Check(id)) {
+        PyErr_SetString(PyExc_TypeError, "ID must be an int");
+        return NULL;
+    }
+
+    PyInterpreterState *interp = _look_up(id);
+    if (interp == NULL) {
+        return NULL;
+    }
+    int is_running = _is_running(interp);
+    if (is_running < 0) {
+        return NULL;
+    }
+    if (is_running) {
+        Py_RETURN_TRUE;
+    }
+    Py_RETURN_FALSE;
+}
+
+PyDoc_STRVAR(is_running_doc,
+"is_running(id) -> bool\n\
+\n\
+Return whether or not the identified interpreter is running.");
+
+static PyObject *
+channel_create(PyObject *self)
+{
+    int64_t cid = _channel_create(&_globals.channels);
+    if (cid < 0) {
+        return NULL;
+    }
+    PyObject *id = (PyObject *)newchannelid(&ChannelIDtype, cid, 0,
+                                            &_globals.channels, 0);
+    if (id == NULL) {
+        if (_channel_destroy(&_globals.channels, cid) != 0) {
+            // XXX issue a warning?
+        }
+        return NULL;
+    }
+    assert(((channelid *)id)->channels != NULL);
+    return id;
+}
+
+PyDoc_STRVAR(channel_create_doc,
+"channel_create() -> ID\n\
+\n\
+Create a new cross-interpreter channel and return a unique generated ID.");
+
+static PyObject *
+channel_destroy(PyObject *self, PyObject *args)
+{
+    PyObject *id;
+    if (!PyArg_UnpackTuple(args, "channel_destroy", 1, 1, &id)) {
+        return NULL;
+    }
+    int64_t cid = _coerce_id(id);
+    if (cid < 0) {
+        return NULL;
+    }
+
+    if (_channel_destroy(&_globals.channels, cid) != 0) {
+        return NULL;
+    }
+    Py_RETURN_NONE;
+}
+
+PyDoc_STRVAR(channel_destroy_doc,
+"channel_destroy(ID)\n\
+\n\
+Close and finalize the channel.  Afterward attempts to use the channel\n\
+will behave as though it never existed.");
+
+static PyObject *
+channel_list_all(PyObject *self)
+{
+    int64_t count = 0;
+    int64_t *cids = _channels_list_all(&_globals.channels, &count);
+    if (cids == NULL) {
+        if (count == 0) {
+            return PyList_New(0);
+        }
+        return NULL;
+    }
+    PyObject *ids = PyList_New((Py_ssize_t)count);
+    if (ids == NULL) {
+        // XXX free cids
+        return NULL;
+    }
+    for (int64_t i=0; i < count; cids++, i++) {
+        PyObject *id = (PyObject *)newchannelid(&ChannelIDtype, *cids, 0,
+                                                &_globals.channels, 0);
+        if (id == NULL) {
+            Py_DECREF(ids);
+            ids = NULL;
+            break;
+        }
+        PyList_SET_ITEM(ids, i, id);
+    }
+    // XXX free cids
+    return ids;
+}
+
+PyDoc_STRVAR(channel_list_all_doc,
+"channel_list_all() -> [ID]\n\
+\n\
+Return the list of all IDs for active channels.");
+
+static PyObject *
+channel_send(PyObject *self, PyObject *args)
+{
+    PyObject *id;
+    PyObject *obj;
+    if (!PyArg_UnpackTuple(args, "channel_send", 2, 2, &id, &obj)) {
+        return NULL;
+    }
+    int64_t cid = _coerce_id(id);
+    if (cid < 0) {
+        return NULL;
+    }
+
+    if (_channel_send(&_globals.channels, cid, obj) != 0) {
+        return NULL;
+    }
+    Py_RETURN_NONE;
+}
+
+PyDoc_STRVAR(channel_send_doc,
+"channel_send(ID, obj)\n\
+\n\
+Add the object's data to the channel's queue.");
+
+static PyObject *
+channel_recv(PyObject *self, PyObject *args)
+{
+    PyObject *id;
+    if (!PyArg_UnpackTuple(args, "channel_recv", 1, 1, &id)) {
+        return NULL;
+    }
+    int64_t cid = _coerce_id(id);
+    if (cid < 0) {
+        return NULL;
+    }
+
+    return _channel_recv(&_globals.channels, cid);
+}
+
+PyDoc_STRVAR(channel_recv_doc,
+"channel_recv(ID) -> obj\n\
+\n\
+Return a new object from the data at the from of the channel's queue.");
+
+static PyObject *
+channel_close(PyObject *self, PyObject *args, PyObject *kwds)
+{
+    PyObject *id;
+    if (!PyArg_UnpackTuple(args, "channel_recv", 1, 1, &id)) {
+        return NULL;
+    }
+    int64_t cid = _coerce_id(id);
+    if (cid < 0) {
+        return NULL;
+    }
+
+    if (_channel_close(&_globals.channels, cid) != 0) {
+        return NULL;
+    }
+    Py_RETURN_NONE;
+}
+
+PyDoc_STRVAR(channel_close_doc,
+"channel_close(ID)\n\
+\n\
+Close the channel for all interpreters.  Once the channel's ID has\n\
+no more ref counts the channel will be destroyed.");
+
+static PyObject *
+channel_drop_interpreter(PyObject *self, PyObject *args, PyObject *kwds)
+{
+    // Note that only the current interpreter is affected.
+    static char *kwlist[] = {"id", "send", "recv"};
+    PyObject *id;
+    int send = -1;
+    int recv = -1;
+    if (!PyArg_ParseTupleAndKeywords(args, kwds,
+                                     "O|$pp:channel_drop_interpreter", kwlist,
+                                     &id, &send, &recv))
+        return NULL;
+
+    int64_t cid = _coerce_id(id);
+    if (cid < 0) {
+        return NULL;
+    }
+    if (send < 0 && recv < 0) {
+        send = 1;
+        recv = 1;
+    }
+    else {
+        if (send < 0) {
+            send = 0;
+        }
+        if (recv < 0) {
+            recv = 0;
+        }
+    }
+    if (_channel_drop(&_globals.channels, cid, send, recv) != 0) {
+        return NULL;
+    }
+    Py_RETURN_NONE;
+}
+
+PyDoc_STRVAR(channel_drop_interpreter_doc,
+"channel_drop_interpreter(ID, *, send=None, recv=None)\n\
+\n\
+Close the channel for the current interpreter.  'send' and 'recv'\n\
+(bool) may be used to indicate the ends to close.  By default both\n\
+ends are closed.  Closing an already closed end is a noop.");
+
+static PyObject *
+channel__channel_id(PyObject *self, PyObject *args, PyObject *kwds)
+{
+    return channelid_new(&ChannelIDtype, args, kwds);
+}
+
+static PyMethodDef module_functions[] = {
+    {"create",                    (PyCFunction)interp_create,
+     METH_VARARGS, create_doc},
+    {"destroy",                   (PyCFunction)interp_destroy,
+     METH_VARARGS, destroy_doc},
+    {"list_all",                  (PyCFunction)interp_list_all,
+     METH_NOARGS, list_all_doc},
+    {"get_current",               (PyCFunction)interp_get_current,
+     METH_NOARGS, get_current_doc},
+    {"get_main",                  (PyCFunction)interp_get_main,
+     METH_NOARGS, get_main_doc},
+    {"is_running",                (PyCFunction)interp_is_running,
+     METH_VARARGS, is_running_doc},
+    {"run_string",                (PyCFunction)interp_run_string,
+     METH_VARARGS, run_string_doc},
+
+    {"is_shareable",              (PyCFunction)object_is_shareable,
+     METH_VARARGS, is_shareable_doc},
+
+    {"channel_create",            (PyCFunction)channel_create,
+     METH_NOARGS, channel_create_doc},
+    {"channel_destroy",           (PyCFunction)channel_destroy,
+     METH_VARARGS, channel_destroy_doc},
+    {"channel_list_all",          (PyCFunction)channel_list_all,
+     METH_NOARGS, channel_list_all_doc},
+    {"channel_send",              (PyCFunction)channel_send,
+     METH_VARARGS, channel_send_doc},
+    {"channel_recv",              (PyCFunction)channel_recv,
+     METH_VARARGS, channel_recv_doc},
+    {"channel_close",             (PyCFunction)channel_close,
+     METH_VARARGS, channel_close_doc},
+    {"channel_drop_interpreter",  (PyCFunction)channel_drop_interpreter,
+     METH_VARARGS | METH_KEYWORDS, channel_drop_interpreter_doc},
+    {"_channel_id",               (PyCFunction)channel__channel_id,
+     METH_VARARGS | METH_KEYWORDS, NULL},
+
+    {NULL,                        NULL}           /* sentinel */
+};
+
+
+/* initialization function */
+
+PyDoc_STRVAR(module_doc,
+"This module provides primitive operations to manage Python interpreters.\n\
+The 'interpreters' module provides a more convenient interface.");
+
+static struct PyModuleDef interpretersmodule = {
+    PyModuleDef_HEAD_INIT,
+    "_xxsubinterpreters",  /* m_name */
+    module_doc,            /* m_doc */
+    -1,                    /* m_size */
+    module_functions,      /* m_methods */
+    NULL,                  /* m_slots */
+    NULL,                  /* m_traverse */
+    NULL,                  /* m_clear */
+    NULL                   /* m_free */
+};
+
+
+PyMODINIT_FUNC
+PyInit__xxsubinterpreters(void)
+{
+    if (_init_globals() != 0) {
+        return NULL;
+    }
+
+    /* Initialize types */
+    ChannelIDtype.tp_base = &PyLong_Type;
+    if (PyType_Ready(&ChannelIDtype) != 0) {
+        return NULL;
+    }
+
+    /* Create the module */
+    PyObject *module = PyModule_Create(&interpretersmodule);
+    if (module == NULL) {
+        return NULL;
+    }
+
+    /* Add exception types */
+    PyObject *ns = PyModule_GetDict(module);  // borrowed
+    if (interp_exceptions_init(ns) != 0) {
+        return NULL;
+    }
+    if (channel_exceptions_init(ns) != 0) {
+        return NULL;
+    }
+
+    /* Add other types */
+    Py_INCREF(&ChannelIDtype);
+    if (PyDict_SetItemString(ns, "ChannelID", (PyObject *)&ChannelIDtype) != 0) {
+        return NULL;
+    }
+
+    if (_PyCrossInterpreterData_Register_Class(&ChannelIDtype, _channelid_shared)) {
+        return NULL;
+    }
+
+    return module;
+}
diff --git a/Python/pystate.c b/Python/pystate.c
index 909d831465..a474549a8c 100644
--- a/Python/pystate.c
+++ b/Python/pystate.c
@@ -54,8 +54,13 @@ _PyRuntimeState_Init_impl(_PyRuntimeState *runtime)
     if (runtime->interpreters.mutex == NULL) {
         return _Py_INIT_ERR("Can't initialize threads for interpreter");
     }
-
     runtime->interpreters.next_id = -1;
+
+    runtime->xidregistry.mutex = PyThread_allocate_lock();
+    if (runtime->xidregistry.mutex == NULL) {
+        return _Py_INIT_ERR("Can't initialize threads for cross-interpreter data registry");
+    }
+
     return _Py_INIT_OK();
 }
 
@@ -166,6 +171,7 @@ PyInterpreterState_New(void)
         /* overflow or Py_Initialize() not called! */
         PyErr_SetString(PyExc_RuntimeError,
                         "failed to get an interpreter ID");
+        /* XXX deallocate! */
         interp = NULL;
     } else {
         interp->id = _PyRuntime.interpreters.next_id;
@@ -256,6 +262,28 @@ PyInterpreterState_GetID(PyInterpreterState *interp)
 }
 
 
+PyInterpreterState *
+_PyInterpreterState_LookUpID(PY_INT64_T requested_id)
+{
+    if (requested_id < 0)
+        goto error;
+
+    PyInterpreterState *interp = PyInterpreterState_Head();
+    while (interp != NULL) {
+        PY_INT64_T id = PyInterpreterState_GetID(interp);
+        if (id < 0)
+            return NULL;
+        if (requested_id == id)
+            return interp;
+        interp = PyInterpreterState_Next(interp);
+    }
+
+error:
+    PyErr_Format(PyExc_RuntimeError,
+                 "unrecognized interpreter ID %lld", requested_id);
+    return NULL;
+}
+
 /* Default implementation for _PyThreadState_GetFrame */
 static struct _frame *
 threadstate_getframe(PyThreadState *self)
@@ -1024,6 +1052,251 @@ PyGILState_Release(PyGILState_STATE oldstate)
 }
 
 
+/**************************/
+/* cross-interpreter data */
+/**************************/
+
+/* cross-interpreter data */
+
+crossinterpdatafunc _PyCrossInterpreterData_Lookup(PyObject *);
+
+/* This is a separate func from _PyCrossInterpreterData_Lookup in order
+   to keep the registry code separate. */
+static crossinterpdatafunc
+_lookup_getdata(PyObject *obj)
+{
+    crossinterpdatafunc getdata = _PyCrossInterpreterData_Lookup(obj);
+    if (getdata == NULL && PyErr_Occurred() == 0)
+        PyErr_Format(PyExc_ValueError,
+                     "%S does not support cross-interpreter data", obj);
+    return getdata;
+}
+
+int
+_PyObject_CheckCrossInterpreterData(PyObject *obj)
+{
+    crossinterpdatafunc getdata = _lookup_getdata(obj);
+    if (getdata == NULL) {
+        return -1;
+    }
+    return 0;
+}
+
+static int
+_check_xidata(_PyCrossInterpreterData *data)
+{
+    // data->data can be anything, including NULL, so we don't check it.
+
+    // data->obj may be NULL, so we don't check it.
+
+    if (data->interp < 0) {
+        PyErr_SetString(PyExc_SystemError, "missing interp");
+        return -1;
+    }
+
+    if (data->new_object == NULL) {
+        PyErr_SetString(PyExc_SystemError, "missing new_object func");
+        return -1;
+    }
+
+    // data->free may be NULL, so we don't check it.
+
+    return 0;
+}
+
+int
+_PyObject_GetCrossInterpreterData(PyObject *obj, _PyCrossInterpreterData *data)
+{
+    PyThreadState *tstate = PyThreadState_Get();
+    // PyThreadState_Get() aborts if lookup fails, so we don't need
+    // to check the result for NULL.
+    PyInterpreterState *interp = tstate->interp;
+
+    // Reset data before re-populating.
+    *data = (_PyCrossInterpreterData){0};
+    data->free = PyMem_RawFree;  // Set a default that may be overridden.
+
+    // Call the "getdata" func for the object.
+    Py_INCREF(obj);
+    crossinterpdatafunc getdata = _lookup_getdata(obj);
+    if (getdata == NULL) {
+        Py_DECREF(obj);
+        return -1;
+    }
+    int res = getdata(obj, data);
+    Py_DECREF(obj);
+    if (res != 0) {
+        return -1;
+    }
+
+    // Fill in the blanks and validate the result.
+    Py_XINCREF(data->obj);
+    data->interp = interp->id;
+    if (_check_xidata(data) != 0) {
+        _PyCrossInterpreterData_Release(data);
+        return -1;
+    }
+
+    return 0;
+}
+
+void
+_PyCrossInterpreterData_Release(_PyCrossInterpreterData *data)
+{
+    if (data->data == NULL && data->obj == NULL) {
+        // Nothing to release!
+        return;
+    }
+
+    // Switch to the original interpreter.
+    PyInterpreterState *interp = _PyInterpreterState_LookUpID(data->interp);
+    if (interp == NULL) {
+        // The intepreter was already destroyed.
+        if (data->free != NULL) {
+            // XXX Someone leaked some memory...
+        }
+        return;
+    }
+    PyThreadState *tstate = PyInterpreterState_ThreadHead(interp);
+    PyThreadState *save_tstate = PyThreadState_Swap(tstate);
+
+    // "Release" the data and/or the object.
+    if (data->free != NULL) {
+        data->free(data->data);
+    }
+    Py_XDECREF(data->obj);
+
+    // Switch back.
+    if (save_tstate != NULL)
+        PyThreadState_Swap(save_tstate);
+}
+
+PyObject *
+_PyCrossInterpreterData_NewObject(_PyCrossInterpreterData *data)
+{
+    return data->new_object(data);
+}
+
+/* registry of {type -> crossinterpdatafunc} */
+
+/* For now we use a global registry of shareable classes.  An
+   alternative would be to add a tp_* slot for a class's
+   crossinterpdatafunc. It would be simpler and more efficient. */
+
+static int
+_register_xidata(PyTypeObject *cls, crossinterpdatafunc getdata)
+{
+    // Note that we effectively replace already registered classes
+    // rather than failing.
+    struct _xidregitem *newhead = PyMem_RawMalloc(sizeof(struct _xidregitem));
+    if (newhead == NULL)
+        return -1;
+    newhead->cls = cls;
+    newhead->getdata = getdata;
+    newhead->next = _PyRuntime.xidregistry.head;
+    _PyRuntime.xidregistry.head = newhead;
+    return 0;
+}
+
+static void _register_builtins_for_crossinterpreter_data(void);
+
+int
+_PyCrossInterpreterData_Register_Class(PyTypeObject *cls,
+                                       crossinterpdatafunc getdata)
+{
+    if (!PyType_Check(cls)) {
+        PyErr_Format(PyExc_ValueError, "only classes may be registered");
+        return -1;
+    }
+    if (getdata == NULL) {
+        PyErr_Format(PyExc_ValueError, "missing 'getdata' func");
+        return -1;
+    }
+
+    // Make sure the class isn't ever deallocated.
+    Py_INCREF((PyObject *)cls);
+
+    PyThread_acquire_lock(_PyRuntime.xidregistry.mutex, WAIT_LOCK);
+    if (_PyRuntime.xidregistry.head == NULL) {
+        _register_builtins_for_crossinterpreter_data();
+    }
+    int res = _register_xidata(cls, getdata);
+    PyThread_release_lock(_PyRuntime.xidregistry.mutex);
+    return res;
+}
+
+crossinterpdatafunc
+_PyCrossInterpreterData_Lookup(PyObject *obj)
+{
+    PyObject *cls = PyObject_Type(obj);
+    crossinterpdatafunc getdata = NULL;
+    PyThread_acquire_lock(_PyRuntime.xidregistry.mutex, WAIT_LOCK);
+    struct _xidregitem *cur = _PyRuntime.xidregistry.head;
+    if (cur == NULL) {
+        _register_builtins_for_crossinterpreter_data();
+        cur = _PyRuntime.xidregistry.head;
+    }
+    for(; cur != NULL; cur = cur->next) {
+        if (cur->cls == (PyTypeObject *)cls) {
+            getdata = cur->getdata;
+            break;
+        }
+    }
+    PyThread_release_lock(_PyRuntime.xidregistry.mutex);
+    return getdata;
+}
+
+/* cross-interpreter data for builtin types */
+
+static PyObject *
+_new_bytes_object(_PyCrossInterpreterData *data)
+{
+    return PyBytes_FromString((char *)(data->data));
+}
+
+static int
+_bytes_shared(PyObject *obj, _PyCrossInterpreterData *data)
+{
+    data->data = (void *)(PyBytes_AS_STRING(obj));
+    data->obj = obj;  // Will be "released" (decref'ed) when data released.
+    data->new_object = _new_bytes_object;
+    data->free = NULL;  // Do not free the data (it belongs to the object).
+    return 0;
+}
+
+static PyObject *
+_new_none_object(_PyCrossInterpreterData *data)
+{
+    // XXX Singleton refcounts are problematic across interpreters...
+    Py_INCREF(Py_None);
+    return Py_None;
+}
+
+static int
+_none_shared(PyObject *obj, _PyCrossInterpreterData *data)
+{
+    data->data = NULL;
+    // data->obj remains NULL
+    data->new_object = _new_none_object;
+    data->free = NULL;  // There is nothing to free.
+    return 0;
+}
+
+static void
+_register_builtins_for_crossinterpreter_data(void)
+{
+    // None
+    if (_register_xidata((PyTypeObject *)PyObject_Type(Py_None), _none_shared) != 0) {
+        Py_FatalError("could not register None for cross-interpreter sharing");
+    }
+
+    // bytes
+    if (_register_xidata(&PyBytes_Type, _bytes_shared) != 0) {
+        Py_FatalError("could not register bytes for cross-interpreter sharing");
+    }
+}
+
+
 #ifdef __cplusplus
 }
 #endif
diff --git a/setup.py b/setup.py
index ba0a7624cf..6f9a39719b 100644
--- a/setup.py
+++ b/setup.py
@@ -755,6 +755,10 @@ class PyBuildExt(build_ext):
             ['_xxtestfuzz/_xxtestfuzz.c', '_xxtestfuzz/fuzzer.c'])
         )
 
+        # Python interface to subinterpreter C-API.
+        exts.append(Extension('_xxsubinterpreters', ['_xxsubinterpretersmodule.c'],
+                              define_macros=[('Py_BUILD_CORE', '')]))
+
         #
         # Here ends the simple stuff.  From here on, modules need certain
         # libraries, are platform-specific, or present other surprises.