Suggested changes:

* removed excess test case from `test_threading` * changed NEWS entry to not mention internal function * allowed `subprocess.Popen` at shutdown if `preexec_fn` is `None`
python · gpshead · Jun 4, 2023 · May 24, 2023 · May 24, 2023 · May 24, 2023
commit bbc8ca4ae24dc5ba296a2923a355b77aeaac7626
diff --git a/Lib/test/test_threading.py b/Lib/test/test_threading.py
@@ -531,34 +531,6 @@ def test_daemon_param(self):
         t = threading.Thread(daemon=True)
         self.assertTrue(t.daemon)
 
-    @support.requires_fork()
-    def test_fork_at_exit(self):
-        # bpo-42350: Calling os.fork() after threading._shutdown() must
-        # not log an error.
-        code = textwrap.dedent("""
-            import atexit
-            import os
-            import sys
-            from test.support import wait_process
-
-            # Import the threading module to register its "at fork" callback
-            import threading
-
-            def exit_handler():
-                pid = os.fork()
-                if not pid:
-                    print("child process ok", file=sys.stderr, flush=True)
-                    # child process
-                else:
-                    wait_process(pid, exitcode=0)
-
-            # exit_handler() will be called after threading._shutdown()
-            atexit.register(exit_handler)
-        """)
-        _, out, err = assert_python_ok("-c", code)
-        self.assertEqual(out, b'')
-        self.assertIn(b"can't fork at interpreter shutdown", err)
-
     @support.requires_fork()
     def test_dummy_thread_after_fork(self):
         # Issue #14308: a dummy thread in the active list doesn't mess up

diff --git a/Misc/NEWS.d/next/Core and Builtins/2023-05-24-12-10-54.gh-issue-104690.HX3Jou.rst b/Misc/NEWS.d/next/Core and Builtins/2023-05-24-12-10-54.gh-issue-104690.HX3Jou.rst
@@ -1 +1 @@
-Starting new threads and process creation through fork/vfork at interpreter exit is no longer supported, as it can lead to race condition between main Python thread freeing existing thread states and :func:`_thread.start_new_thread` trying to allocate and use state of just created thread or fork/vfork trying to use main thread state in child process.
+Starting new threads and process creation through fork/vfork at interpreter exit is no longer supported, as it can lead to race condition between main Python thread freeing existing thread states and internal :mod:`threading` routines trying to allocate and use state of just created thread or fork/vfork trying to use main thread state in child process.
@@ -944,9 +944,9 @@ subprocess_fork_exec_impl(PyObject *module, PyObject *process_args,
     Py_ssize_t fds_to_keep_len = PyTuple_GET_SIZE(py_fds_to_keep);
 
     PyInterpreterState *interp = PyInterpreterState_Get();
-    if (interp->finalizing) {
+    if ((preexec_fn != Py_None) && interp->finalizing) {
         PyErr_SetString(PyExc_RuntimeError,
-                        "can't create child process at interpreter shutdown");
+                        "preexec_fn not supported at interpreter shutdown");
         return NULL;
     }
     if ((preexec_fn != Py_None) && (interp != PyInterpreterState_Main())) {
Original file line number	Diff line number	Diff line change
		@@ -1 +1 @@
		Starting new threads and process creation through fork/vfork at interpreter exit is no longer supported, as it can lead to race condition between main Python thread freeing existing thread states and :func:`_thread.start_new_thread` trying to allocate and use state of just created thread or fork/vfork trying to use main thread state in child process.
		Starting new threads and process creation through fork/vfork at interpreter exit is no longer supported, as it can lead to race condition between main Python thread freeing existing thread states and internal :mod:`threading` routines trying to allocate and use state of just created thread or fork/vfork trying to use main thread state in child process.