| 1 | package bot |
| 2 | |
| 3 | import ( |
| 4 | "context" |
| 5 | "errors" |
| 6 | "io" |
| 7 | "log/slog" |
| 8 | "runtime" |
| 9 | "sync" |
| 10 | "testing" |
| 11 | "time" |
| 12 | |
| 13 | "reasonix/internal/control" |
| 14 | ) |
| 15 | |
| 16 | type closeProbeBotController struct { |
| 17 | *control.Controller |
| 18 | onClose func() |
| 19 | } |
| 20 | |
| 21 | type stopWaitBotController struct { |
| 22 | stubBotController |
| 23 | started chan struct{} |
| 24 | release chan struct{} |
| 25 | } |
| 26 | |
| 27 | func (c *stopWaitBotController) RunTurn(context.Context, string) error { |
| 28 | close(c.started) |
| 29 | <-c.release |
| 30 | return nil |
| 31 | } |
| 32 | |
| 33 | func (c *stopWaitBotController) SessionPath() string { return "" } |
| 34 | func (c *stopWaitBotController) WorkspaceRoot() string { return "" } |
| 35 | func (c *stopWaitBotController) Close() {} |
| 36 | |
| 37 | type countingStopAdapter struct { |
| 38 | *fakeAdapter |
| 39 | mu sync.Mutex |
| 40 | stopCalls int |
| 41 | } |
| 42 | |
| 43 | type cancelBlockingStartAdapter struct { |
| 44 | *fakeAdapter |
| 45 | entered chan struct{} |
| 46 | } |
| 47 | |
| 48 | func (a *cancelBlockingStartAdapter) Start(ctx context.Context) error { |
| 49 | close(a.entered) |
| 50 | <-ctx.Done() |
| 51 | return ctx.Err() |
| 52 | } |
| 53 | |
| 54 | func (a *countingStopAdapter) Stop() error { |
| 55 | a.mu.Lock() |
| 56 | a.stopCalls++ |
| 57 | a.mu.Unlock() |
| 58 | return a.fakeAdapter.Stop() |
| 59 | } |
| 60 | |
| 61 | func (a *countingStopAdapter) calls() int { |
| 62 | a.mu.Lock() |
| 63 | defer a.mu.Unlock() |
| 64 | return a.stopCalls |
| 65 | } |
| 66 | |
| 67 | func (c *closeProbeBotController) Close() { |
| 68 | if c.onClose != nil { |
| 69 | c.onClose() |
| 70 | } |
| 71 | } |
| 72 | |
| 73 | func TestBotGatewayStopClosesSessionsWithoutGatewayLock(t *testing.T) { |
| 74 | gw := &BotGateway{ |
| 75 | controllers: map[string]*sessionState{}, |
| 76 | } |
| 77 | closed := make(chan struct{}, 1) |
| 78 | gw.controllers["session"] = &sessionState{ |
| 79 | ctrl: &closeProbeBotController{ |
| 80 | Controller: control.New(control.Options{}), |
| 81 | onClose: func() { |
| 82 | gw.mu.Lock() |
| 83 | gw.mu.Unlock() //nolint:staticcheck // probe: lock must be immediately acquirable |
| 84 | closed <- struct{}{} |
| 85 | }, |
| 86 | }, |
| 87 | } |
| 88 | |
| 89 | done := make(chan struct{}) |
| 90 | go func() { |
| 91 | gw.Stop() |
| 92 | close(done) |
| 93 | }() |
| 94 | |
| 95 | select { |
| 96 | case <-done: |
| 97 | case <-time.After(time.Second): |
| 98 | t.Fatal("Stop blocked while closing a controller") |
| 99 | } |
| 100 | select { |
| 101 | case <-closed: |
| 102 | case <-time.After(time.Second): |
| 103 | t.Fatal("controller Close was not called") |
| 104 | } |
| 105 | if len(gw.controllers) != 0 { |
| 106 | t.Fatalf("controllers retained after Stop: %d", len(gw.controllers)) |
| 107 | } |
| 108 | } |
| 109 | |
| 110 | func TestBotGatewayStopWaitsForDispatchHandler(t *testing.T) { |
| 111 | adapter := newFakeAdapter(PlatformFeishu, "fake-feishu") |
| 112 | entered := make(chan struct{}) |
| 113 | release := make(chan struct{}) |
| 114 | gw := NewGatewayWithAdapterBindings(GatewayConfig{ |
| 115 | Enabled: map[Platform]bool{PlatformFeishu: true}, |
| 116 | Allowlist: AllowlistConfig{AllowAll: true}, |
| 117 | OnInbound: func(InboundMessage) { |
| 118 | close(entered) |
| 119 | <-release |
| 120 | }, |
| 121 | }, []AdapterBinding{{ID: "feishu-lark", Platform: PlatformFeishu, Adapter: adapter}}, slog.New(slog.NewTextHandler(io.Discard, nil))) |
| 122 | if err := gw.Start(context.Background()); err != nil { |
| 123 | t.Fatalf("Start: %v", err) |
| 124 | } |
| 125 | adapter.msgCh <- InboundMessage{ChatType: ChatDM, ChatID: "chat", UserID: "user", Text: "/status"} |
| 126 | select { |
| 127 | case <-entered: |
| 128 | case <-time.After(time.Second): |
| 129 | t.Fatal("dispatch handler did not start") |
| 130 | } |
| 131 | |
| 132 | done := make(chan struct{}) |
| 133 | go func() { |
| 134 | gw.Stop() |
| 135 | close(done) |
| 136 | }() |
| 137 | select { |
| 138 | case <-done: |
| 139 | t.Fatal("Stop returned while a dispatch handler was still running") |
| 140 | case <-time.After(50 * time.Millisecond): |
| 141 | } |
| 142 | close(release) |
| 143 | select { |
| 144 | case <-done: |
| 145 | case <-time.After(time.Second): |
| 146 | t.Fatal("Stop did not return after the dispatch handler exited") |
| 147 | } |
| 148 | } |
| 149 | |
| 150 | func TestBotGatewayStopWaitsForTurn(t *testing.T) { |
| 151 | adapter := newFakeAdapter(PlatformWeixin, "fake-weixin") |
| 152 | gw := NewGatewayWithAdapterBindings(GatewayConfig{ |
| 153 | Enabled: map[Platform]bool{PlatformWeixin: true}, |
| 154 | Allowlist: AllowlistConfig{AllowAll: true}, |
| 155 | }, []AdapterBinding{{ID: "weixin", Platform: PlatformWeixin, Adapter: adapter}}, slog.New(slog.NewTextHandler(io.Discard, nil))) |
| 156 | ctrl := &stopWaitBotController{started: make(chan struct{}), release: make(chan struct{})} |
| 157 | msg := InboundMessage{Platform: PlatformWeixin, ConnectionID: "weixin", ChatType: ChatDM, ChatID: "chat", UserID: "user", Text: "hello"} |
| 158 | key := BuildSessionKey(msg.Session()) |
| 159 | gw.controllers[key] = &sessionState{ctrl: ctrl, sink: &sessionEventSink{}} |
| 160 | if err := gw.Start(context.Background()); err != nil { |
| 161 | t.Fatalf("Start: %v", err) |
| 162 | } |
| 163 | adapter.msgCh <- msg |
| 164 | select { |
| 165 | case <-ctrl.started: |
| 166 | case <-time.After(time.Second): |
| 167 | t.Fatal("turn did not start") |
| 168 | } |
| 169 | |
| 170 | done := make(chan struct{}) |
| 171 | go func() { |
| 172 | gw.Stop() |
| 173 | close(done) |
| 174 | }() |
| 175 | select { |
| 176 | case <-done: |
| 177 | t.Fatal("Stop returned while a turn was still running") |
| 178 | case <-time.After(50 * time.Millisecond): |
| 179 | } |
| 180 | close(ctrl.release) |
| 181 | select { |
| 182 | case <-done: |
| 183 | case <-time.After(time.Second): |
| 184 | t.Fatal("Stop did not return after the turn exited") |
| 185 | } |
| 186 | } |
| 187 | |
| 188 | type typingHoldAdapter struct { |
| 189 | *fakeAdapter |
| 190 | entered chan struct{} // one send per turn parked in SendTyping |
| 191 | release chan struct{} |
| 192 | } |
| 193 | |
| 194 | func (a *typingHoldAdapter) SendTyping(context.Context, string) error { |
| 195 | a.entered <- struct{}{} |
| 196 | <-a.release |
| 197 | return nil |
| 198 | } |
| 199 | |
| 200 | type cancelPublishBotController struct { |
| 201 | stubBotController |
| 202 | closeEntered chan struct{} // one send when Close begins |
| 203 | closeHold chan struct{} // Close parks here, pinning Stop inside the closeSessions loop |
| 204 | turnCtx chan error // ctx.Err() observed on RunTurn entry |
| 205 | } |
| 206 | |
| 207 | type panickingRootBotController struct { |
| 208 | stubBotController |
| 209 | } |
| 210 | |
| 211 | func (panickingRootBotController) WorkspaceRoot() string { panic("workspace root unavailable") } |
| 212 | |
| 213 | func TestAPanicWhileClaimingASessionReleasesTheGatewayLock(t *testing.T) { |
| 214 | gw := NewGateway(GatewayConfig{Allowlist: AllowlistConfig{AllowAll: true}}, nil, slog.New(slog.NewTextHandler(io.Discard, nil))) |
| 215 | msg := InboundMessage{Platform: PlatformWeixin, ConnectionID: "weixin", ChatType: ChatDM, ChatID: "chat", UserID: "user"} |
| 216 | key := BuildSessionKey(msg.Session()) |
| 217 | gw.controllers[key] = &sessionState{ctrl: panickingRootBotController{}, sink: &sessionEventSink{}} |
| 218 | |
| 219 | panicked := make(chan any, 1) |
| 220 | go func() { |
| 221 | defer func() { panicked <- recover() }() |
| 222 | gw.getOrCreateSession(context.Background(), key, msg) |
| 223 | }() |
| 224 | if recovered := <-panicked; recovered == nil { |
| 225 | t.Fatal("the controller's panic did not reach the caller") |
| 226 | } |
| 227 | |
| 228 | deadline := time.Now().Add(time.Second) |
| 229 | for !gw.mu.TryLock() { |
| 230 | if time.Now().After(deadline) { |
| 231 | t.Fatal("the gateway lock is still held after the panic unwound") |
| 232 | } |
| 233 | time.Sleep(time.Millisecond) |
| 234 | } |
| 235 | gw.mu.Unlock() |
| 236 | } |
| 237 | |
| 238 | func (c *cancelPublishBotController) RunTurn(ctx context.Context, _ string) error { |
| 239 | c.turnCtx <- ctx.Err() |
| 240 | return nil |
| 241 | } |
| 242 | |
| 243 | func (c *cancelPublishBotController) SessionPath() string { return "" } |
| 244 | func (c *cancelPublishBotController) WorkspaceRoot() string { return "" } |
| 245 | func (c *cancelPublishBotController) Close() { |
| 246 | c.closeEntered <- struct{}{} |
| 247 | <-c.closeHold |
| 248 | } |
| 249 | |
| 250 | // Guards the cancel-publication window: runTurn publishes state.cancel under |
| 251 | // gw.mu only after the session is already visible in gw.controllers, so Stop |
| 252 | // can consume the field from another goroutine while the turn is still on its |
| 253 | // way to publication. TestBotGatewayStopWaitsForTurn stops only after RunTurn |
| 254 | // has begun (cancel already published) and never covers this window. |
| 255 | // |
| 256 | // Two sessions are needed because whichever state closeSessions visits first |
| 257 | // has its cancel read before Close signals the test — any write released off |
| 258 | // that signal is ordered after the read and invisible to the race detector. |
| 259 | // The first state's blocking Close pins Stop mid-loop instead, so the second |
| 260 | // state's cancel read happens after the turns were let through to publish. |
| 261 | // Run with -race: an unlocked read of state.cancel here is a data race. |
| 262 | func TestBotGatewayStopBeforeTurnCancelPublication(t *testing.T) { |
| 263 | adapter := &typingHoldAdapter{ |
| 264 | fakeAdapter: newFakeAdapter(PlatformWeixin, "fake-weixin"), |
| 265 | entered: make(chan struct{}, 2), |
| 266 | release: make(chan struct{}), |
| 267 | } |
| 268 | gw := NewGatewayWithAdapterBindings(GatewayConfig{ |
| 269 | Enabled: map[Platform]bool{PlatformWeixin: true}, |
| 270 | Allowlist: AllowlistConfig{AllowAll: true}, |
| 271 | }, []AdapterBinding{{ID: "weixin", Platform: PlatformWeixin, Adapter: adapter}}, slog.New(slog.NewTextHandler(io.Discard, nil))) |
| 272 | closeEntered := make(chan struct{}, 2) |
| 273 | closeHold := make(chan struct{}) |
| 274 | turnCtx := make(chan error, 2) |
| 275 | chats := []string{"chat-a", "chat-b"} |
| 276 | for _, chat := range chats { |
| 277 | msg := InboundMessage{Platform: PlatformWeixin, ConnectionID: "weixin", ChatType: ChatDM, ChatID: chat, UserID: "user"} |
| 278 | gw.controllers[BuildSessionKey(msg.Session())] = &sessionState{ |
| 279 | ctrl: &cancelPublishBotController{closeEntered: closeEntered, closeHold: closeHold, turnCtx: turnCtx}, |
| 280 | sink: &sessionEventSink{}, |
| 281 | } |
| 282 | } |
| 283 | if err := gw.Start(context.Background()); err != nil { |
| 284 | t.Fatalf("Start: %v", err) |
| 285 | } |
| 286 | for _, chat := range chats { |
| 287 | adapter.msgCh <- InboundMessage{Platform: PlatformWeixin, ConnectionID: "weixin", ChatType: ChatDM, ChatID: chat, UserID: "user", Text: "hello"} |
| 288 | } |
| 289 | for range chats { |
| 290 | select { |
| 291 | case <-adapter.entered: |
| 292 | case <-time.After(time.Second): |
| 293 | t.Fatal("turn did not reach the pre-publication window") |
| 294 | } |
| 295 | } |
| 296 | |
| 297 | done := make(chan struct{}) |
| 298 | go func() { |
| 299 | gw.Stop() |
| 300 | close(done) |
| 301 | }() |
| 302 | // Stop is parked inside the closeSessions loop: the first state's cancel is |
| 303 | // consumed and its Close is held; the second state's cancel is still unread. |
| 304 | select { |
| 305 | case <-closeEntered: |
| 306 | case <-time.After(time.Second): |
| 307 | t.Fatal("Stop did not reach the session close loop") |
| 308 | } |
| 309 | // Let both turns publish their cancel funcs while Stop stays parked. The |
| 310 | // sleep is deliberate and cannot become a channel handshake: observing the |
| 311 | // publication would order the write before Stop's read and hide the race |
| 312 | // from the detector. |
| 313 | close(adapter.release) |
| 314 | time.Sleep(200 * time.Millisecond) |
| 315 | close(closeHold) |
| 316 | |
| 317 | for range chats { |
| 318 | select { |
| 319 | case err := <-turnCtx: |
| 320 | if err == nil { |
| 321 | t.Fatal("turn ran with a live context after Stop closed its session") |
| 322 | } |
| 323 | case <-time.After(time.Second): |
| 324 | t.Fatal("turn did not run to completion") |
| 325 | } |
| 326 | } |
| 327 | select { |
| 328 | case <-done: |
| 329 | case <-time.After(time.Second): |
| 330 | t.Fatal("Stop did not return after the late turns exited") |
| 331 | } |
| 332 | } |
| 333 | |
| 334 | func TestBotGatewayStopIsIdempotent(t *testing.T) { |
| 335 | adapter := &countingStopAdapter{fakeAdapter: newFakeAdapter(PlatformFeishu, "fake-feishu")} |
| 336 | gw := NewGatewayWithAdapterBindings(GatewayConfig{ |
| 337 | Enabled: map[Platform]bool{PlatformFeishu: true}, |
| 338 | }, []AdapterBinding{{ID: "feishu-lark", Platform: PlatformFeishu, Adapter: adapter}}, slog.New(slog.NewTextHandler(io.Discard, nil))) |
| 339 | if err := gw.Start(context.Background()); err != nil { |
| 340 | t.Fatalf("Start: %v", err) |
| 341 | } |
| 342 | |
| 343 | gw.Stop() |
| 344 | gw.Stop() |
| 345 | if got := adapter.calls(); got != 1 { |
| 346 | t.Fatalf("adapter Stop calls = %d, want 1", got) |
| 347 | } |
| 348 | } |
| 349 | |
| 350 | func TestBotGatewayStopCancelsConcurrentStart(t *testing.T) { |
| 351 | adapter := &cancelBlockingStartAdapter{ |
| 352 | fakeAdapter: newFakeAdapter(PlatformFeishu, "fake-feishu"), |
| 353 | entered: make(chan struct{}), |
| 354 | } |
| 355 | gw := NewGatewayWithAdapterBindings(GatewayConfig{ |
| 356 | Enabled: map[Platform]bool{PlatformFeishu: true}, |
| 357 | }, []AdapterBinding{{ID: "feishu-lark", Platform: PlatformFeishu, Adapter: adapter}}, slog.New(slog.NewTextHandler(io.Discard, nil))) |
| 358 | startDone := make(chan error, 1) |
| 359 | go func() { startDone <- gw.Start(context.Background()) }() |
| 360 | select { |
| 361 | case <-adapter.entered: |
| 362 | case <-time.After(time.Second): |
| 363 | t.Fatal("adapter Start did not begin") |
| 364 | } |
| 365 | |
| 366 | stopDone := make(chan struct{}) |
| 367 | go func() { |
| 368 | gw.Stop() |
| 369 | close(stopDone) |
| 370 | }() |
| 371 | select { |
| 372 | case err := <-startDone: |
| 373 | if !errors.Is(err, context.Canceled) { |
| 374 | t.Fatalf("Start error = %v, want context canceled", err) |
| 375 | } |
| 376 | case <-time.After(time.Second): |
| 377 | t.Fatal("Stop did not cancel the in-progress Start") |
| 378 | } |
| 379 | select { |
| 380 | case <-stopDone: |
| 381 | case <-time.After(time.Second): |
| 382 | t.Fatal("Stop did not finish after Start returned") |
| 383 | } |
| 384 | } |
| 385 | |
| 386 | // Guards the gw.cfg.Channels / gw.cfg.ConnectionChannels / gw.cfg.ToolApprovalMode |
| 387 | // snapshot locking: approval-mode writers mutate those under gw.mu while |
| 388 | // sessionOptionsForMessage and the project/session index builders read them. |
| 389 | // Run with -race; a lock-free read is a concurrent map read/write crash. |
| 390 | func TestBotGatewayToolApprovalModeConcurrentWithConfigReaders(t *testing.T) { |
| 391 | t.Setenv("REASONIX_HOME", t.TempDir()) |
| 392 | gw := &BotGateway{ |
| 393 | cfg: GatewayConfig{ |
| 394 | WorkspaceRoot: t.TempDir(), |
| 395 | Channels: map[Platform]ChannelConfig{ |
| 396 | PlatformFeishu: {ToolApprovalMode: control.ToolApprovalAsk}, |
| 397 | }, |
| 398 | ConnectionChannels: map[string]ChannelConfig{ |
| 399 | "feishu-lark": {ToolApprovalMode: control.ToolApprovalAsk}, |
| 400 | }, |
| 401 | }, |
| 402 | controllers: map[string]*sessionState{}, |
| 403 | sessionOverrides: map[string]sessionRuntimeOverride{}, |
| 404 | logger: slog.New(slog.NewTextHandler(io.Discard, nil)), |
| 405 | } |
| 406 | |
| 407 | // Keep both sides finite. The former stop-driven writer kept allocating until |
| 408 | // the reader returned and repeatedly tripped Go's Windows GC in normal CI. |
| 409 | // The shared start edge orders neither loop after the other, so -race still |
| 410 | // observes any missing map lock even if one goroutine happens to run first. |
| 411 | const iterations = 200 |
| 412 | start := make(chan struct{}) |
| 413 | writerDone := make(chan struct{}) |
| 414 | go func() { |
| 415 | defer close(writerDone) |
| 416 | <-start |
| 417 | modes := []string{control.ToolApprovalYolo, control.ToolApprovalAsk, control.ToolApprovalAuto} |
| 418 | for i := range iterations { |
| 419 | mode := modes[i%len(modes)] |
| 420 | gw.UpdateConnectionToolApprovalMode("feishu-lark", mode) |
| 421 | gw.mu.Lock() |
| 422 | gw.updateToolApprovalModeDefaultLocked(InboundMessage{Platform: PlatformFeishu}, mode) |
| 423 | gw.updateToolApprovalModeDefaultLocked(InboundMessage{}, mode) |
| 424 | gw.mu.Unlock() |
| 425 | runtime.Gosched() |
| 426 | } |
| 427 | }() |
| 428 | |
| 429 | close(start) |
| 430 | connMsg := InboundMessage{Platform: PlatformFeishu, ConnectionID: "feishu-lark", ChatType: ChatDM, ChatID: "chat", UserID: "user"} |
| 431 | for range iterations { |
| 432 | gw.sessionOptionsForMessage(connMsg) |
| 433 | gw.sessionOptionsForMessage(InboundMessage{Platform: PlatformFeishu}) |
| 434 | projects := gw.buildProjectIndex() |
| 435 | gw.buildSessionIndex(projects) |
| 436 | runtime.Gosched() |
| 437 | } |
| 438 | <-writerDone |
| 439 | } |
| 440 |