python
diff --git a/Lib/test/test_threading.py b/Lib/test/test_threading.py
@@ -345,7 +345,7 @@ def run(self):
 
     def test_limbo_cleanup(self):
         # Issue 7481: Failure to start thread should cleanup the limbo map.
-        def fail_new_thread(*args):
+        def fail_new_thread(*args, **kwargs):
             raise threading.ThreadError()
         _start_new_thread = threading._start_new_thread
         threading._start_new_thread = fail_new_thread

diff --git a/Lib/threading.py b/Lib/threading.py
@@ -49,6 +49,11 @@
 except AttributeError:
     _CRLock = None
 TIMEOUT_MAX = _thread.TIMEOUT_MAX
+_wait_for_threads_fini = _thread._wait_for_threads_fini
+try:
+    _internal_after_fork = _thread._after_fork
+except AttributeError:
+    _internal_after_fork = None
 del _thread
 
 
@@ -968,7 +973,7 @@ def start(self):
         with _active_limbo_lock:
             _limbo[self] = self
         try:
-            _start_new_thread(self._bootstrap, ())
+            _start_new_thread(self._bootstrap, (), daemonic=self._daemonic)
         except Exception:
             with _active_limbo_lock:
                 del _limbo[self]
@@ -1589,6 +1594,7 @@ def _shutdown():
         pass
 
     # Join all non-deamon threads
+    # XXX We should be able to drop this in favor of _wait_for_threads_fini().
     while True:
         with _shutdown_locks_lock:
             locks = list(_shutdown_locks)
@@ -1605,6 +1611,9 @@ def _shutdown():
         # new threads can be spawned while we were waiting for the other
         # threads to complete
 
+    # Wait for all non-daemon threads to be finalized.
+    _wait_for_threads_fini()
+
 
 def main_thread():
     """Return the main thread object.
@@ -1677,4 +1686,6 @@ def _after_fork():
 
 
 if hasattr(_os, "register_at_fork"):
+    if _internal_after_fork is not None:
+        _os.register_at_fork(after_in_child=_internal_after_fork)
     _os.register_at_fork(after_in_child=_after_fork)