gh-105716: Support Background Threads in Subinterpreters Consistently (…

…gh-109921)

The existence of background threads running on a subinterpreter was preventing interpreters from getting properly destroyed, as well as impacting the ability to run the interpreter again. It also affected how we wait for non-daemon threads to finish.

We add PyInterpreterState.threads.main, with some internal C-API functions.

Include/cpython/pystate.h

@@ -211,6 +211,7 @@ struct _ts {
  * if it is NULL. */
 PyAPI_FUNC(PyThreadState *) _PyThreadState_UncheckedGet(void);
 
+
 // Disable tracing and profiling.
 PyAPI_FUNC(void) PyThreadState_EnterTracing(PyThreadState *tstate);
 

Include/internal/pycore_interp.h

@@ -73,6 +73,8 @@ struct _is {
         uint64_t next_unique_id;
         /* The linked list of threads, newest first. */
         PyThreadState *head;
+        /* The thread currently executing in the __main__ module, if any. */
+        PyThreadState *main;
         /* Used in Modules/_threadmodule.c. */
         long count;
         /* Support for runtime thread stack size tuning.

Include/internal/pycore_pystate.h

@@ -44,6 +44,11 @@ _Py_IsMainInterpreterFinalizing(PyInterpreterState *interp)
             interp == &_PyRuntime._main_interpreter);
 }
 
+// Export for _xxsubinterpreters module.
+PyAPI_FUNC(int) _PyInterpreterState_SetRunningMain(PyInterpreterState *);
+PyAPI_FUNC(void) _PyInterpreterState_SetNotRunningMain(PyInterpreterState *);
+PyAPI_FUNC(int) _PyInterpreterState_IsRunningMain(PyInterpreterState *);
+
 
 static inline const PyConfig *
 _Py_GetMainConfig(void)

Lib/test/test_interpreters.py

@@ -261,6 +261,16 @@ def test_subinterpreter(self):
             self.assertTrue(interp.is_running())
         self.assertFalse(interp.is_running())
 
+    def test_finished(self):
+        r, w = os.pipe()
+        interp = interpreters.create()
+        interp.run(f"""if True:
+            import os
+            os.write({w}, b'x')
+            """)
+        self.assertFalse(interp.is_running())
+        self.assertEqual(os.read(r, 1), b'x')
+
     def test_from_subinterpreter(self):
         interp = interpreters.create()
         out = _run_output(interp, dedent(f"""
@@ -288,6 +298,31 @@ def test_bad_id(self):
         with self.assertRaises(ValueError):
             interp.is_running()
 
+    def test_with_only_background_threads(self):
+        r_interp, w_interp = os.pipe()
+        r_thread, w_thread = os.pipe()
+
+        DONE = b'D'
+        FINISHED = b'F'
+
+        interp = interpreters.create()
+        interp.run(f"""if True:
+            import os
+            import threading
+
+            def task():
+                v = os.read({r_thread}, 1)
+                assert v == {DONE!r}
+                os.write({w_interp}, {FINISHED!r})
+            t = threading.Thread(target=task)
+            t.start()
+            """)
+        self.assertFalse(interp.is_running())
+
+        os.write(w_thread, DONE)
+        interp.run('t.join()')
+        self.assertEqual(os.read(r_interp, 1), FINISHED)
+
 
 class TestInterpreterClose(TestBase):
 
@@ -389,6 +424,37 @@ def test_still_running(self):
                 interp.close()
             self.assertTrue(interp.is_running())
 
+    def test_subthreads_still_running(self):
+        r_interp, w_interp = os.pipe()
+        r_thread, w_thread = os.pipe()
+
+        FINISHED = b'F'
+
+        interp = interpreters.create()
+        interp.run(f"""if True:
+            import os
+            import threading
+            import time
+
+            done = False
+
+            def notify_fini():
+                global done
+                done = True
+                t.join()
+            threading._register_atexit(notify_fini)
+
+            def task():
+                while not done:
+                    time.sleep(0.1)
+                os.write({w_interp}, {FINISHED!r})
+            t = threading.Thread(target=task)
+            t.start()
+            """)
+        interp.close()
+
+        self.assertEqual(os.read(r_interp, 1), FINISHED)
+
 
 class TestInterpreterRun(TestBase):
 
@@ -465,6 +531,37 @@ def test_bytes_for_script(self):
         with self.assertRaises(TypeError):
             interp.run(b'print("spam")')
 
+    def test_with_background_threads_still_running(self):
+        r_interp, w_interp = os.pipe()
+        r_thread, w_thread = os.pipe()
+
+        RAN = b'R'
+        DONE = b'D'
+        FINISHED = b'F'
+
+        interp = interpreters.create()
+        interp.run(f"""if True:
+            import os
+            import threading
+
+            def task():
+                v = os.read({r_thread}, 1)
+                assert v == {DONE!r}
+                os.write({w_interp}, {FINISHED!r})
+            t = threading.Thread(target=task)
+            t.start()
+            os.write({w_interp}, {RAN!r})
+            """)
+        interp.run(f"""if True:
+            os.write({w_interp}, {RAN!r})
+            """)
+
+        os.write(w_thread, DONE)
+        interp.run('t.join()')
+        self.assertEqual(os.read(r_interp, 1), RAN)
+        self.assertEqual(os.read(r_interp, 1), RAN)
+        self.assertEqual(os.read(r_interp, 1), FINISHED)
+
     # test_xxsubinterpreters covers the remaining Interpreter.run() behavior.
 
 

Lib/test/test_threading.py

@@ -26,6 +26,11 @@
 from test import lock_tests
 from test import support
 
+try:
+    from test.support import interpreters
+except ModuleNotFoundError:
+    interpreters = None
+
 threading_helper.requires_working_threading(module=True)
 
 # Between fork() and exec(), only async-safe functions are allowed (issues
@@ -52,6 +57,12 @@ def skip_unless_reliable_fork(test):
     return test
 
 
+def requires_subinterpreters(meth):
+    """Decorator to skip a test if subinterpreters are not supported."""
+    return unittest.skipIf(interpreters is None,
+                           'subinterpreters required')(meth)
+
+
 def restore_default_excepthook(testcase):
     testcase.addCleanup(setattr, threading, 'excepthook', threading.excepthook)
     threading.excepthook = threading.__excepthook__
@@ -1311,6 +1322,44 @@ def f():
         # The thread was joined properly.
         self.assertEqual(os.read(r, 1), b"x")
 
+    @requires_subinterpreters
+    def test_threads_join_with_no_main(self):
+        r_interp, w_interp = self.pipe()
+
+        INTERP = b'I'
+        FINI = b'F'
+        DONE = b'D'
+
+        interp = interpreters.create()
+        interp.run(f"""if True:
+            import os
+            import threading
+            import time
+
+            done = False
+
+            def notify_fini():
+                global done
+                done = True
+                os.write({w_interp}, {FINI!r})
+                t.join()
+            threading._register_atexit(notify_fini)
+
+            def task():
+                while not done:
+                    time.sleep(0.1)
+                os.write({w_interp}, {DONE!r})
+            t = threading.Thread(target=task)
+            t.start()
+
+            os.write({w_interp}, {INTERP!r})
+            """)
+        interp.close()
+
+        self.assertEqual(os.read(r_interp, 1), INTERP)
+        self.assertEqual(os.read(r_interp, 1), FINI)
+        self.assertEqual(os.read(r_interp, 1), DONE)
+
     @cpython_only
     def test_daemon_threads_fatal_error(self):
         subinterp_code = f"""if 1:

Lib/threading.py

@@ -38,6 +38,7 @@
 _allocate_lock = _thread.allocate_lock
 _set_sentinel = _thread._set_sentinel
 get_ident = _thread.get_ident
+_is_main_interpreter = _thread._is_main_interpreter
 try:
     get_native_id = _thread.get_native_id
     _HAVE_THREAD_NATIVE_ID = True
@@ -1574,7 +1575,7 @@ def _shutdown():
     # the main thread's tstate_lock - that won't happen until the interpreter
     # is nearly dead.  So we release it here.  Note that just calling _stop()
     # isn't enough:  other threads may already be waiting on _tstate_lock.
-    if _main_thread._is_stopped:
+    if _main_thread._is_stopped and _is_main_interpreter():
         # _shutdown() was already called
         return
 
@@ -1627,6 +1628,7 @@ def main_thread():
     In normal conditions, the main thread is the thread from which the
     Python interpreter was started.
     """
+    # XXX Figure this out for subinterpreters.  (See gh-75698.)
     return _main_thread
 
 # get thread-local implementation, either from the thread

Misc/NEWS.d/next/Core and Builtins/2023-09-26-14-00-25.gh-issue-105716.SUJkW1.rst

@@ -0,0 +1,3 @@
+Subinterpreters now correctly handle the case where they have threads
+running in the background.  Before, such threads would interfere with
+cleaning up and destroying them, as well as prevent running another script.

Modules/_threadmodule.c

@@ -1605,6 +1605,18 @@ PyDoc_STRVAR(excepthook_doc,
 \n\
 Handle uncaught Thread.run() exception.");
 
+static PyObject *
+thread__is_main_interpreter(PyObject *module, PyObject *Py_UNUSED(ignored))
+{
+    PyInterpreterState *interp = _PyInterpreterState_GET();
+    return PyBool_FromLong(_Py_IsMainInterpreter(interp));
+}
+
+PyDoc_STRVAR(thread__is_main_interpreter_doc,
+"_is_main_interpreter()\n\
+\n\
+Return True if the current interpreter is the main Python interpreter.");
+
 static PyMethodDef thread_methods[] = {
     {"start_new_thread",        (PyCFunction)thread_PyThread_start_new_thread,
      METH_VARARGS, start_new_doc},
@@ -1634,8 +1646,10 @@ static PyMethodDef thread_methods[] = {
      METH_VARARGS, stack_size_doc},
     {"_set_sentinel",           thread__set_sentinel,
      METH_NOARGS, _set_sentinel_doc},
-    {"_excepthook",              thread_excepthook,
+    {"_excepthook",             thread_excepthook,
      METH_O, excepthook_doc},
+    {"_is_main_interpreter",    thread__is_main_interpreter,
+     METH_NOARGS, thread__is_main_interpreter_doc},
     {NULL,                      NULL}           /* sentinel */
 };