gh-128639: Don't assume one thread in subinterpreter finalization (gh-128640)

Incidentally, this also fixed the warning not showing up if a subinterpreter wasn't
cleaned up via _interpreters.destroy. I had to update some of the tests as a result.

Lib/test/test_interpreters/test_api.py

@@ -647,6 +647,59 @@ class TestInterpreterClose(TestBase):
                     self.interp_exists(interpid))
 
 
+    def test_remaining_threads(self):
+        r_interp, w_interp = self.pipe()
+
+        FINISHED = b'F'
+
+        # It's unlikely, but technically speaking, it's possible
+        # that the thread could've finished before interp.close() is
+        # reached, so this test might not properly exercise the case.
+        # However, it's quite unlikely and I'm too lazy to deal with it.
+        interp = interpreters.create()
+        interp.exec(f"""if True:
+            import os
+            import threading
+            import time
+
+            def task():
+                time.sleep(1)
+                os.write({w_interp}, {FINISHED!r})
+
+            threads = [threading.Thread(target=task) for _ in range(3)]
+            for t in threads:
+                t.start()
+            """)
+        interp.close()
+
+        self.assertEqual(os.read(r_interp, 1), FINISHED)
+
+    def test_remaining_daemon_threads(self):
+        interp = _interpreters.create(
+            types.SimpleNamespace(
+                use_main_obmalloc=False,
+                allow_fork=False,
+                allow_exec=False,
+                allow_threads=True,
+                allow_daemon_threads=True,
+                check_multi_interp_extensions=True,
+                gil='own',
+            )
+        )
+        _interpreters.exec(interp, f"""if True:
+            import threading
+            import time
+
+            def task():
+                time.sleep(100)
+
+            threads = [threading.Thread(target=task, daemon=True) for _ in range(3)]
+            for t in threads:
+                t.start()
+            """)
+        _interpreters.destroy(interp)
+
+
 class TestInterpreterPrepareMain(TestBase):
 
     def test_empty(self):
@@ -755,7 +808,10 @@ class TestInterpreterExec(TestBase):
                 spam.eggs()
 
             interp = interpreters.create()
-            interp.exec(script)
+            try:
+                interp.exec(script)
+            finally:
+                interp.close()
             """)
 
         stdout, stderr = self.assert_python_failure(scriptfile)
@@ -764,7 +820,7 @@ class TestInterpreterExec(TestBase):
         #      File "{interpreters.__file__}", line 179, in exec
         self.assertEqual(stderr, dedent(f"""\
             Traceback (most recent call last):
-              File "{scriptfile}", line 9, in <module>
+              File "{scriptfile}", line 10, in <module>
                 interp.exec(script)
                 ~~~~~~~~~~~^^^^^^^^
               {interpmod_line.strip()}

Lib/test/test_interpreters/test_lifecycle.py

@@ -132,6 +132,7 @@ class StartupTests(TestBase):
                     'sub': sys.path[0],
                 }}, indent=4), flush=True)
                 """)
+            interp.close()
             '''
         # <tmp>/
         #   pkg/
@@ -172,7 +173,10 @@ class FinalizationTests(TestBase):
         argv = [sys.executable, '-c', '''if True:
             from test.support import interpreters
             interp = interpreters.create()
-            raise Exception
+            try:
+                raise Exception
+            finally:
+                interp.close()
             ''']
         proc = subprocess.run(argv, capture_output=True, text=True)
         self.assertIn('Traceback', proc.stderr)

Lib/test/test_threading.py

@@ -1689,10 +1689,7 @@ class SubinterpThreadingTests(BaseTestCase):
 
             _testcapi.run_in_subinterp(%r)
             """ % (subinterp_code,)
-        with test.support.SuppressCrashReport():
-            rc, out, err = assert_python_failure("-c", script)
-        self.assertIn("Fatal Python error: Py_EndInterpreter: "
-                      "not the last thread", err.decode())
+        assert_python_ok("-c", script)
 
     def _check_allowed(self, before_start='', *,
                        allowed=True,

Misc/NEWS.d/next/Core_and_Builtins/2025-01-08-12-52-47.gh-issue-128640.9nbh9z.rst

@@ -0,0 +1 @@
+Fix a crash when using threads inside of a subinterpreter.

Programs/_testembed.c

@@ -1395,9 +1395,12 @@ static int test_audit_subinterpreter(void)
     PySys_AddAuditHook(_audit_subinterpreter_hook, NULL);
     _testembed_initialize();
 
-    Py_NewInterpreter();
-    Py_NewInterpreter();
-    Py_NewInterpreter();
+    PyThreadState *tstate = PyThreadState_Get();
+    for (int i = 0; i < 3; ++i)
+    {
+        Py_EndInterpreter(Py_NewInterpreter());
+        PyThreadState_Swap(tstate);
+    }
 
     Py_Finalize();
 

Python/pylifecycle.c

@@ -1992,6 +1992,7 @@ resolve_final_tstate(_PyRuntimeState *runtime)
             }
             else {
                 /* Fall back to the current tstate.  It's better than nothing. */
+                // XXX No it's not
                 main_tstate = tstate;
             }
         }
@@ -2037,6 +2038,16 @@ _Py_Finalize(_PyRuntimeState *runtime)
 
     _PyAtExit_Call(tstate->interp);
 
+    /* Clean up any lingering subinterpreters.
+
+       Two preconditions need to be met here:
+
+        - This has to happen before _PyRuntimeState_SetFinalizing is
+          called, or else threads might get prematurely blocked.
+        - The world must not be stopped, as finalizers can run.
+    */
+    finalize_subinterpreters();
+
     assert(_PyThreadState_GET() == tstate);
 
     /* Copy the core config, PyInterpreterState_Delete() free
@@ -2124,9 +2135,6 @@ _Py_Finalize(_PyRuntimeState *runtime)
     _PyImport_FiniExternal(tstate->interp);
     finalize_modules(tstate);
 
-    /* Clean up any lingering subinterpreters. */
-    finalize_subinterpreters();
-
     /* Print debug stats if any */
     _PyEval_Fini();
 
@@ -2410,9 +2418,8 @@ Py_NewInterpreter(void)
     return tstate;
 }
 
-/* Delete an interpreter and its last thread.  This requires that the
-   given thread state is current, that the thread has no remaining
-   frames, and that it is its interpreter's only remaining thread.
+/* Delete an interpreter.  This requires that the given thread state
+   is current, and that the thread has no remaining frames.
    It is a fatal error to violate these constraints.
 
    (Py_FinalizeEx() doesn't have these constraints -- it zaps
@@ -2442,14 +2449,15 @@ Py_EndInterpreter(PyThreadState *tstate)
     _Py_FinishPendingCalls(tstate);
 
     _PyAtExit_Call(tstate->interp);
-
-    if (tstate != interp->threads.head || tstate->next != NULL) {
-        Py_FatalError("not the last thread");
-    }
+    _PyRuntimeState *runtime = interp->runtime;
+    _PyEval_StopTheWorldAll(runtime);
+    PyThreadState *list = _PyThreadState_RemoveExcept(tstate);
 
     /* Remaining daemon threads will automatically exit
        when they attempt to take the GIL (ex: PyEval_RestoreThread()). */
     _PyInterpreterState_SetFinalizing(interp, tstate);
+    _PyEval_StartTheWorldAll(runtime);
+    _PyThreadState_DeleteList(list, /*is_after_fork=*/0);
 
     // XXX Call something like _PyImport_Disable() here?
 
@@ -2480,6 +2488,8 @@ finalize_subinterpreters(void)
     PyInterpreterState *main_interp = _PyInterpreterState_Main();
     assert(final_tstate->interp == main_interp);
     _PyRuntimeState *runtime = main_interp->runtime;
+    assert(!runtime->stoptheworld.world_stopped);
+    assert(_PyRuntimeState_GetFinalizing(runtime) == NULL);
     struct pyinterpreters *interpreters = &runtime->interpreters;
 
     /* Get the first interpreter in the list. */
@@ -2508,27 +2518,17 @@ finalize_subinterpreters(void)
 
     /* Clean up all remaining subinterpreters. */
     while (interp != NULL) {
-        assert(!_PyInterpreterState_IsRunningMain(interp));
-
-        /* Find the tstate to use for fini.  We assume the interpreter
-           will have at most one tstate at this point. */
-        PyThreadState *tstate = interp->threads.head;
-        if (tstate != NULL) {
-            /* Ideally we would be able to use tstate as-is, and rely
-               on it being in a ready state: no exception set, not
-               running anything (tstate->current_frame), matching the
-               current thread ID (tstate->thread_id).  To play it safe,
-               we always delete it and use a fresh tstate instead. */
-            assert(tstate != final_tstate);
-            _PyThreadState_Attach(tstate);
-            PyThreadState_Clear(tstate);
-            _PyThreadState_Detach(tstate);
-            PyThreadState_Delete(tstate);
+        /* Make a tstate for finalization. */
+        PyThreadState *tstate = _PyThreadState_NewBound(interp, _PyThreadState_WHENCE_FINI);
+        if (tstate == NULL) {
+            // XXX Some graceful way to always get a thread state?
+            Py_FatalError("thread state allocation failed");
         }
-        tstate = _PyThreadState_NewBound(interp, _PyThreadState_WHENCE_FINI);
+
+        /* Enter the subinterpreter. */
+        _PyThreadState_Attach(tstate);
 
         /* Destroy the subinterpreter. */
-        _PyThreadState_Attach(tstate);
         Py_EndInterpreter(tstate);
         assert(_PyThreadState_GET() == NULL);