test: verify Windows Provider cancellation and descendant cleanup

README.md

@@ -94,3 +94,5 @@ child process tree. Ctrl+C during a Provider call returns 130.
 
 See [the startup/transport contract](docs/provider-startup.md) for error codes,
 compatibility and failure-injection commands.
+
+Since 2026.10.21.1, user cancellation returns 130 and a `cancelled` code without Provider-failure diagnostics or a synthetic fail verdict.

README.zh-CN.md

@@ -83,3 +83,5 @@ print/TUI 会显示准备信息。`--emit-events` 在 session 前每十秒发送
 超时或调用期间取消会终止子进程树；调用期间 Ctrl+C 返回 130。
 
 详见 [启动与传输契约](docs/provider-startup.md)。
+
+从 2026.10.21.1 起，用户取消返回 130 和 `cancelled`，不会误报 Provider 故障或产生伪造的 fail 判定。

docs/provider-startup.md

@@ -49,3 +49,5 @@ Run portable failure injection with `D2X=/absolute/path/to/d2x python tests/star
 `tests/e2e.sh` additionally covers save/advance, locking and print output flushing.
 CI stages candidate binaries with `provenance.json` containing repository, checkout
 SHA, binary version and SHA256. These are test artifacts, not stable releases.
+
+Since 2026.10.21.1, user cancellation returns 130 and a `cancelled` code without Provider-failure diagnostics or a synthetic fail verdict.

mcpp.toml

@@ -1,6 +1,6 @@
 [package]
 name        = "d2x"
-version     = "2026.10.20.2"
+version     = "2026.10.21.1"
 description = "AI-powered development assistant for C++ projects"
 license     = "Apache-2.0"
 repo        = "https://github.com/d2learn/d2x"

protocol/mcpp.toml

@@ -1,6 +1,6 @@
 [package]
 name        = "d2x-protocol"
-version     = "2026.10.20.2"
+version     = "2026.10.21.1"
 description = "d2x Provider/Frontend 协议层:类型、编解码、传输(参考实现;规范性载体是协议文档,跨语言合规判据是 conformance 套件)"
 license     = "Apache-2.0"
 

protocol/src/transport.cppm

@@ -121,7 +121,7 @@ export class ProcessProvider final : public IExerciseProvider {
                 mErrorCode, mCommand, args, std::filesystem::current_path().string(),
                 std::chrono::duration_cast<std::chrono::seconds>(std::chrono::steady_clock::now() - start).count(),
                 status.exit_code, out_tail, err_tail);
-            warn(mError);
+            if (!mCancelled) warn(mError);
         }
         return status;
     }

src/checker.cppm

@@ -122,9 +122,13 @@ export int run(const std::string& start_target = "", bool emit_events = false) {
     if (emit_events) sink = std::make_unique<emit::StdoutSink>();
     else sink = std::make_unique<emit::UiSink>();
     auto fail = [&](std::string_view code, std::string_view phase, std::string_view message) {
+        if (code == "cancelled") {
+            sink->error(code, phase, "Cancelled by user");
+            return 130;
+        }
         sink->error(code, phase, message);
         log::error("{}", message);
-        return code == "cancelled" ? 130 : 1;
+        return 1;
     };
     try {
     // 单实例锁:两个监听循环互踩的破坏是静默的(见 instance_lock 模块头注)。
@@ -170,6 +174,7 @@ export int run(const std::string& start_target = "", bool emit_events = false) {
     // 再打 "No targets found for checking."，两层都在误导学员。
     std::string provider_name;
     if (!provider.describe(provider_name)) {
+        if (provider.cancelled()) return fail("cancelled", "describe", "");
         log::error("Provider 无响应: {}", command);
         log::error("请确认该命令可执行，且实现了 d2x Provider 协议（describe/exercises/check）");
         return fail(provider.error_code(), "describe", provider.error() +
@@ -215,6 +220,7 @@ export int run(const std::string& start_target = "", bool emit_events = false) {
             };
 
             auto verdict = provider.check(exercise, progress);
+            if (provider.cancelled()) return fail("cancelled", "check", "");
             // verdict.output 里可能有传输层事后补充的说明(缺 verdict 的原因、
             // 活性超时的终止说明)——它们没走过流式回调,不补发就永远到不了
             // 前端/协议消费者。

src/config.cppm

@@ -17,7 +17,7 @@ export struct EnvVars;
 export class Config;
 
 struct Info {
-    static constexpr std::string_view VERSION = "2026.10.20.2";   // 日期版本制 YYYY.MM.DD.N(见 2026-07-24 设计文档决策记录)
+    static constexpr std::string_view VERSION = "2026.10.21.1";   // 日期版本制 YYYY.MM.DD.N(见 2026-07-24 设计文档决策记录)
     static constexpr std::string_view REPO = "https://github.com/d2learn/d2x";
 };
 

src/emit.cppm

@@ -144,7 +144,8 @@ public:
         refresh();
     }
     void error(std::string_view code, std::string_view phase, std::string_view message) override {
-        mStage.assign(phase); mOutcome = "error";
+        mStage = code == "cancelled" ? "cancelled" : std::string(phase);
+        mOutcome = code == "cancelled" ? "" : "error";
         mOutput = std::format("{}: {}", code, message);
         refresh();
     }

tests/startup_test.py

@@ -50,7 +50,10 @@ def provider():
         if mode == 'enumerate-exit':
             sys.exit(8)
     elif verb == 'check':
-        if mode == 'check-hang':
+        if mode in ('check-hang', 'check-tree'):
+            if mode == 'check-tree':
+                child = subprocess.Popen([sys.executable, '-c', 'import time; time.sleep(120)'])
+                Path('child.pid').write_text(str(child.pid))
             time.sleep(120)
         emit(event='verdict', outcome='pass', stage='run')
         if mode == 'check-exit':
@@ -147,33 +150,75 @@ def test_timeout_validation_and_env_precedence(self):
         p, _, _ = self.run_case(env={'D2X_PROVIDER_STARTUP_TIMEOUT': '5garbage'})
         self.assertNotEqual(p.returncode, 0)
 
-    @unittest.skipIf(os.name == 'nt', 'POSIX signal delivery; Windows Job timeout covered above')
     def test_cancel_reaps_process_tree(self):
-        with tempfile.TemporaryDirectory() as folder:
-            root = Path(folder)
-            command = ' '.join(shlex.quote(s) for s in [sys.executable, str(Path(__file__).resolve()), '--provider'])
-            (root / '.d2x.json').write_text(json.dumps({'buildtools': command}))
-            with subprocess.Popen([os.environ['D2X'], 'checker', '--emit-events'], cwd=root,
-                                  env={**os.environ, 'TEST_MODE': 'tree'}, stdout=subprocess.PIPE,
-                                  stderr=subprocess.PIPE, text=True) as process:
-                limit = time.monotonic() + 5
-                while not (root / 'child.pid').exists() and time.monotonic() < limit:
-                    time.sleep(.05)
-                self.assertTrue((root / 'child.pid').exists())
-                child = int((root / 'child.pid').read_text())
-                process.send_signal(signal.SIGINT)
-                out, err = process.communicate(timeout=5)
-                self.assertEqual(process.returncode, 130, err)
-                self.assertEqual(json.loads(out.splitlines()[-1])['code'], 'cancelled')
-                # A killed orphan may be a zombie until init reaps it; it must
-                # no longer execute, even when the direct shell already died.
-                state = Path(f'/proc/{child}/stat')
-                if state.exists():
-                    self.assertIn(state.read_text().split()[2], ['Z', 'X'])
-                else:
-                    with self.assertRaises(ProcessLookupError):
-                        os.kill(child, 0)
-                self.assertFalse((root / '.d2x/checker.lock').exists())
+        for mode in ['tree', 'check-tree']:
+            with tempfile.TemporaryDirectory() as folder:
+                root = Path(folder)
+                quote = (lambda s: '"' + s + '"') if os.name == 'nt' else shlex.quote
+                command = ' '.join(quote(s) for s in [sys.executable, str(Path(__file__).resolve()), '--provider'])
+                (root / '.d2x.json').write_text(json.dumps({'buildtools': command}))
+                (root / 'one.txt').write_text('one')
+                allocated_console = False
+                process = None
+                if os.name == 'nt':
+                    import ctypes
+                    api = ctypes.windll.kernel32
+                    if not api.GetConsoleCP():
+                        self.assertTrue(api.AllocConsole(), 'Cannot allocate the test console')
+                        allocated_console = True
+                try:
+                    process = subprocess.Popen([os.environ['D2X'], 'checker', '--emit-events'], cwd=root,
+                        env={**os.environ, 'TEST_MODE': mode}, stdout=subprocess.PIPE, stderr=subprocess.PIPE,
+                        text=True, encoding='utf-8', errors='replace',
+                        creationflags=subprocess.CREATE_NEW_PROCESS_GROUP if os.name == 'nt' else 0)
+                    limit = time.monotonic() + 5
+                    while not (root / 'child.pid').exists() and time.monotonic() < limit:
+                        time.sleep(.05)
+                    self.assertTrue((root / 'child.pid').exists())
+                    child = int((root / 'child.pid').read_text())
+                    process.send_signal(signal.CTRL_BREAK_EVENT if os.name == 'nt' else signal.SIGINT)
+                    out, err = process.communicate(timeout=5)
+                    self.assertEqual(process.returncode, 130, err)
+                    events = [json.loads(line) for line in out.splitlines()]
+                    self.assertEqual(events[-1]['code'], 'cancelled')
+                    self.assertEqual(events[-1]['phase'], 'describe' if mode == 'tree' else 'check')
+                    self.assertNotIn('Provider 无响应', err)
+                    self.assertNotIn('WARNING', err)
+                    self.assertFalse(any(e.get('event') == 'verdict' for e in events))
+                    if os.name == 'nt':
+                        api.OpenProcess.restype = ctypes.c_void_p
+                        api.GetExitCodeProcess.argtypes = [ctypes.c_void_p, ctypes.POINTER(ctypes.c_ulong)]
+                        api.CloseHandle.argtypes = [ctypes.c_void_p]
+                        api.WaitForSingleObject.argtypes = [ctypes.c_void_p, ctypes.c_ulong]
+                        handle = api.OpenProcess(0x101000, False, child)
+                        if handle:
+                            try:
+                                self.assertEqual(api.WaitForSingleObject(handle, 2000), 0)
+                                code = ctypes.c_ulong()
+                                api.GetExitCodeProcess(handle, ctypes.byref(code))
+                                self.assertNotEqual(code.value, 259, 'Cancelled descendant is still active')
+                            finally:
+                                api.CloseHandle(handle)
+                    else:
+                        limit = time.monotonic() + 2
+                        while True:
+                            state = subprocess.run(['ps', '-o', 'stat=', '-p', str(child)],
+                                                   capture_output=True, text=True).stdout.strip()
+                            if not state or state.startswith(('Z', 'X')):
+                                break
+                            self.assertLess(time.monotonic(), limit, 'Cancelled descendant is still active')
+                            time.sleep(.05)
+                    self.assertFalse((root / '.d2x/checker.lock').exists())
+                finally:
+                    if process is not None and process.poll() is None:
+                        process.terminate()
+                        try:
+                            process.wait(timeout=3)
+                        except subprocess.TimeoutExpired:
+                            process.kill()
+                            process.wait(timeout=3)
+                    if allocated_console:
+                        api.FreeConsole()
 
     def test_install_coordinates_course_dependencies(self):
         for failure in [False, True]: