feat: cli auto-starts and stops locally-managed servers if no --api-url is passed

1 files changed, 283 insertions, 0 deletions

diff --git a/app/tests/test_backends_managed.py b/app/tests/test_backends_managed.py
new file mode 100644
index 0000000..cd1f03a
--- /dev/null
+++ b/app/tests/test_backends_managed.py
@@ -0,0 +1,283 @@
+"""Tests for the CLI's managed-server bootstrap (backends/managed.py).
+
+The CLI (without --api-url) calls ``ensure_running`` before converting and
+``ManagedServer.shutdown`` after; both are tested against mocked
+``servers``/``probe``/registry-detect so no process is ever spawned. The
+qwen paths exercise the real ``build_spec``/``model_for_voice_mode``
+mapping (pure config reads) to pin the model-per-voice-mode contract.
+"""
+
+import io
+import unittest
+from contextlib import redirect_stdout
+from types import SimpleNamespace
+from unittest.mock import ANY, patch
+
+from backends import ServerSpec, managed, servers
+from backends.managed import ManagedServer, ensure_running
+from backends.probe import (IDENTITY_AUDIOCPP, IDENTITY_QWEN_CLONE,
+                            IDENTITY_QWEN_CUSTOM, IDENTITY_QWEN_DESIGN)
+from converter.clients import (BACKEND_AUDIOCPP, BACKEND_QWEN,
+                               VOICE_MODE_CLONE, VOICE_MODE_CUSTOM,
+                               VOICE_MODE_DESIGN)
+
+
+def _spec(name="audiocpp", url="http://127.0.0.1:8080", identity=None):
+    return ServerSpec(name, url, ["/bin/fake_server"], identity=identity)
+
+
+def _status(specs, installed=True, label="audio.cpp"):
+    """A minimal stand-in for the registry's BackendStatus."""
+    return SimpleNamespace(servers=specs, installed=installed, label=label)
+
+
+class ShutdownTests(unittest.TestCase):
+    """ManagedServer.shutdown stops only what the run started."""
+
+    def test_noop_when_not_started(self):
+        with patch.object(servers, "stop") as mk_stop:
+            ManagedServer(_spec()).shutdown()
+        mk_stop.assert_not_called()
+
+    def test_stops_the_spec_when_started(self):
+        spec = _spec()
+        with patch.object(servers, "stop") as mk_stop:
+            ManagedServer(spec, started=True).shutdown()
+        mk_stop.assert_called_once_with(spec.name)
+
+
+class EnsureRunningTests(unittest.TestCase):
+    """ensure_running resolves the spec and boots or reuses the server."""
+
+    def setUp(self):
+        self.spec = _spec(identity=IDENTITY_AUDIOCPP)
+
+    def _run(self, backend=BACKEND_AUDIOCPP, voice_mode="custom_voice"):
+        out = io.StringIO()
+        with redirect_stdout(out):
+            result = ensure_running(backend, voice_mode)
+        return result, out.getvalue()
+
+    def test_not_installed_returns_none_with_warning(self):
+        with patch("backends.detect",
+                   return_value=_status([self.spec], installed=False)), \
+                patch.object(servers, "start") as mk_start:
+            result, output = self._run()
+        self.assertIsNone(result)
+        self.assertIn("not installed", output)
+        mk_start.assert_not_called()
+
+    def test_unknown_backend_returns_none_with_warning(self):
+        with patch("backends.detect", return_value=None), \
+                patch.object(servers, "start") as mk_start:
+            result, output = self._run(backend="gone")
+        self.assertIsNone(result)
+        self.assertIn("not installed", output)
+        mk_start.assert_not_called()
+
+    def test_no_specs_returns_none_with_warning(self):
+        with patch("backends.detect",
+                   return_value=_status([], installed=True)), \
+                patch.object(servers, "start") as mk_start:
+            result, _ = self._run()
+        self.assertIsNone(result)
+        mk_start.assert_not_called()
+
+    def test_starts_when_port_free(self):
+        with patch("backends.detect",
+                   return_value=_status([self.spec])), \
+                patch("backends.common.server_running",
+                      return_value=False), \
+                patch.object(servers, "start", return_value=True) as mk_start:
+            result, _ = self._run()
+        mk_start.assert_called_once_with(self.spec, progress=ANY)
+        self.assertTrue(result.ok)
+        self.assertTrue(result.started)
+        self.assertEqual(result.spec, self.spec)
+
+    def test_start_failure_reports_not_ok_and_never_stops(self):
+        with patch("backends.detect",
+                   return_value=_status([self.spec])), \
+                patch("backends.common.server_running",
+                      return_value=False), \
+                patch.object(servers, "start", return_value=False), \
+                patch.object(servers, "stop") as mk_stop:
+            result, _ = self._run()
+            result.shutdown()
+        self.assertFalse(result.ok)
+        self.assertFalse(result.started)
+        mk_stop.assert_not_called()
+
+    def test_running_server_is_reused_and_left_running(self):
+        with patch("backends.detect",
+                   return_value=_status([self.spec])), \
+                patch("backends.common.server_running",
+                      return_value=True), \
+                patch.object(servers, "start") as mk_start:
+            result, _ = self._run()
+            result.shutdown()
+        mk_start.assert_not_called()
+        self.assertTrue(result.ok)
+        self.assertFalse(result.started)
+
+    def test_running_event_marks_the_run_as_a_reuser(self):
+        """A server appearing under us between the port check and the spawn.
+
+        ``servers.start`` reports "running" instead of spawning in that
+        race; the run must then not stop the server at shutdown.
+        """
+        seen = []
+
+        def fake_start(spec, progress=None, cancel=None):
+            event = {"kind": "running", "name": spec.name, "url": spec.url}
+            seen.append(event)
+            if progress is not None:
+                progress(event)
+            return True
+
+        with patch("backends.detect",
+                   return_value=_status([self.spec])), \
+                patch("backends.common.server_running",
+                      return_value=False), \
+                patch.object(servers, "start", side_effect=fake_start), \
+                patch.object(servers, "_console_progress") as mk_print, \
+                patch.object(servers, "stop") as mk_stop:
+            result, _ = self._run()
+            result.shutdown()
+        self.assertTrue(result.ok)
+        self.assertFalse(result.started)
+        # The console still hears the event (delegated printer).
+        mk_print.assert_called_once_with(seen[0])
+        mk_stop.assert_not_called()
+
+    def test_keyboard_interrupt_stops_the_spawned_server(self):
+        with patch("backends.detect",
+                   return_value=_status([self.spec])), \
+                patch("backends.common.server_running",
+                      return_value=False), \
+                patch.object(servers, "start",
+                             side_effect=KeyboardInterrupt), \
+                patch.object(servers, "pid_for", return_value=4242), \
+                patch.object(servers, "stop") as mk_stop, \
+                self.assertRaises(KeyboardInterrupt):
+            self._run()
+        mk_stop.assert_called_once_with(self.spec.name)
+
+    def test_keyboard_interrupt_without_spawn_skips_the_stop(self):
+        with patch("backends.detect",
+                   return_value=_status([self.spec])), \
+                patch("backends.common.server_running",
+                      return_value=False), \
+                patch.object(servers, "start",
+                             side_effect=KeyboardInterrupt), \
+                patch.object(servers, "pid_for", return_value=None), \
+                patch.object(servers, "stop") as mk_stop, \
+                self.assertRaises(KeyboardInterrupt):
+            self._run()
+        mk_stop.assert_not_called()
+
+
+class QwenEnsureRunningTests(unittest.TestCase):
+    """qwen hosts one model per server: the running-model check is aware."""
+
+    def _detect_qwen(self):
+        from backends import qwen
+        return _status([qwen.build_spec("CustomVoice")],
+                       installed=True, label="qwen-tts")
+
+    def _run(self, voice_mode, backend=BACKEND_QWEN):
+        out = io.StringIO()
+        with redirect_stdout(out):
+            result = ensure_running(backend, voice_mode)
+        return result, out.getvalue()
+
+    def test_spec_aims_at_the_model_the_voice_mode_needs(self):
+        from backends import qwen
+        cases = [(VOICE_MODE_CLONE, "Base", IDENTITY_QWEN_CLONE),
+                 (VOICE_MODE_CUSTOM, "CustomVoice", IDENTITY_QWEN_CUSTOM),
+                 (VOICE_MODE_DESIGN, "VoiceDesign", IDENTITY_QWEN_DESIGN)]
+        for voice_mode, model, identity in cases:
+            with self.subTest(voice_mode=voice_mode):
+                with patch("backends.detect", return_value=self._detect_qwen()), \
+                        patch("backends.common.server_running",
+                              return_value=False), \
+                        patch.object(servers, "start",
+                                     return_value=True) as mk_start:
+                    result, _ = self._run(voice_mode)
+                spec = mk_start.call_args.args[0]
+                self.assertEqual(spec.identity, identity)
+                self.assertIn(qwen.MODEL_REPOS[model], spec.argv)
+                self.assertTrue(result.started)
+
+    def test_running_server_hosting_the_wanted_model_is_reused(self):
+        with patch("backends.detect", return_value=self._detect_qwen()), \
+                patch("backends.common.server_running", return_value=True), \
+                patch("backends.probe.identify_server",
+                      return_value=IDENTITY_QWEN_CLONE), \
+                patch.object(servers, "start") as mk_start, \
+                patch.object(servers, "stop") as mk_stop:
+            result, _ = self._run(VOICE_MODE_CLONE)
+            result.shutdown()
+        mk_start.assert_not_called()
+        mk_stop.assert_not_called()
+        self.assertTrue(result.ok)
+        self.assertFalse(result.started)
+
+    def test_managed_server_hosting_another_model_is_rebooted(self):
+        with patch("backends.detect", return_value=self._detect_qwen()), \
+                patch("backends.common.server_running", return_value=True), \
+                patch("backends.probe.identify_server",
+                      return_value=IDENTITY_QWEN_CUSTOM), \
+                patch.object(servers, "alive", return_value=True), \
+                patch.object(servers, "start", return_value=True) as mk_start, \
+                patch.object(servers, "stop") as mk_stop:
+            result, output = self._run(VOICE_MODE_CLONE)
+        self.assertIn("restarting", output)
+        mk_stop.assert_called_once_with("qwen")
+        spec = mk_start.call_args.args[0]
+        self.assertEqual(spec.identity, IDENTITY_QWEN_CLONE)
+        self.assertTrue(result.started)
+        # The rebooted server is this run's: shutdown stops it again.
+        with patch.object(servers, "stop") as mk_stop:
+            result.shutdown()
+        mk_stop.assert_called_once_with("qwen")
+
+    def test_foreign_server_hosting_another_model_refuses_the_run(self):
+        with patch("backends.detect", return_value=self._detect_qwen()), \
+                patch("backends.common.server_running", return_value=True), \
+                patch("backends.probe.identify_server",
+                      return_value=IDENTITY_QWEN_CUSTOM), \
+                patch.object(servers, "alive", return_value=False), \
+                patch.object(servers, "start") as mk_start, \
+                patch.object(servers, "stop") as mk_stop:
+            result, output = self._run(VOICE_MODE_CLONE)
+        self.assertFalse(result.ok)
+        self.assertIn("this run needs Base", output)
+        mk_start.assert_not_called()
+        mk_stop.assert_not_called()
+
+    def test_unidentified_running_server_counts_as_unknown(self):
+        """A port squatting service that probes as nothing known refuses."""
+        with patch("backends.detect", return_value=self._detect_qwen()), \
+                patch("backends.common.server_running", return_value=True), \
+                patch("backends.probe.identify_server", return_value=None), \
+                patch.object(servers, "alive", return_value=False), \
+                patch.object(servers, "start") as mk_start:
+            result, output = self._run(VOICE_MODE_CLONE)
+        self.assertFalse(result.ok)
+        self.assertIn("an unknown", output)
+        mk_start.assert_not_called()
+
+
+class ManagedModuleSmokeTests(unittest.TestCase):
+    """Import-surface sanity for the module the CLI wires in."""
+
+    def test_managed_server_defaults(self):
+        server = ManagedServer(_spec())
+        self.assertFalse(server.started)
+        self.assertTrue(server.ok)
+
+    def test_module_uses_the_servers_module_singletons(self):
+        # managed delegates to the same servers module the TUI uses, so
+        # pid/log files and console output stay identical.
+        self.assertIs(managed.servers, servers)