package cli_test import ( "context" "runtime" "strings" "sync" "testing" "time" "go.uber.org/zap" noderuntime "iop/apps/node/internal/runtime" "iop/packages/config" . "iop/apps/node/internal/adapters/cli" ) type fakeSink struct { mu sync.Mutex events []noderuntime.RuntimeEvent } func (f *fakeSink) Emit(_ context.Context, e noderuntime.RuntimeEvent) error { f.mu.Lock() defer f.mu.Unlock() f.events = append(f.events, e) return nil } func (f *fakeSink) Events() []noderuntime.RuntimeEvent { f.mu.Lock() defer f.mu.Unlock() cp := make([]noderuntime.RuntimeEvent, len(f.events)) copy(cp, f.events) return cp } func TestCLIExecuteOneShotPassesPromptAsArg(t *testing.T) { if runtime.GOOS == "windows" { t.Skip("sh not available on Windows") } cfg := config.CLIConf{ Enabled: true, Profiles: map[string]config.CLIProfileConf{ "echo": { Command: "sh", // $0 = "sh" (script name), $1 = prompt appended by executeOneShot Args: []string{"-c", `printf "reply:%s\n" "$1"`, "sh"}, }, }, } c := New(cfg, zap.NewNop()) sink := &fakeSink{} err := c.Execute(context.Background(), noderuntime.ExecutionSpec{ RunID: "run-1", Model: "echo", Input: map[string]any{"prompt": "hello"}, }, sink) if err != nil { t.Fatalf("execute: %v", err) } events := sink.Events() var types []string var deltas []string for _, e := range events { types = append(types, string(e.Type)) if e.Type == noderuntime.EventTypeDelta { deltas = append(deltas, e.Delta) } } if len(events) < 3 { t.Fatalf("expected at least 3 events (start/delta/complete), got %v", types) } if events[0].Type != noderuntime.EventTypeStart { t.Fatalf("first event should be start, got %q", events[0].Type) } if events[len(events)-1].Type != noderuntime.EventTypeComplete { t.Fatalf("last event should be complete, got %q", events[len(events)-1].Type) } combined := strings.Join(deltas, "") if !strings.Contains(combined, "reply:hello") { t.Fatalf("expected reply:hello in deltas, got %q", combined) } } func TestCLIExecutePersistentWaitsForSlowFirstOutput(t *testing.T) { if runtime.GOOS == "windows" { t.Skip("PTY not supported on Windows") } cfg := config.CLIConf{ Enabled: true, Profiles: map[string]config.CLIProfileConf{ "slow-echo": { Command: "sh", Args: []string{"-c", `stty -echo; while IFS= read -r line; do sleep 0.2; printf "reply:%s\n" "$line"; done`}, Persistent: true, Terminal: true, ResponseIdleTimeoutMS: 50, // shorter than the 200ms sleep StartupIdleTimeoutMS: 50, }, }, } c := New(cfg, zap.NewNop()) ctx := context.Background() if err := c.Start(ctx); err != nil { t.Fatalf("start: %v", err) } defer func() { _ = c.Stop(ctx) }() execCtx, cancel := context.WithTimeout(ctx, 1*time.Second) defer cancel() sink := &fakeSink{} err := c.Execute(execCtx, noderuntime.ExecutionSpec{ RunID: "run-1", Model: "slow-echo", Input: map[string]any{"prompt": "hello"}, }, sink) if err != nil { t.Fatalf("execute: %v", err) } events := sink.Events() var deltas []string var lastType noderuntime.EventType for _, e := range events { if e.Type == noderuntime.EventTypeDelta { deltas = append(deltas, e.Delta) } lastType = e.Type } if lastType != noderuntime.EventTypeComplete { t.Fatalf("expected last event to be complete, got %q", lastType) } combined := strings.Join(deltas, "") if !strings.Contains(combined, "reply:hello") { t.Fatalf("expected reply:hello in deltas, got %q", combined) } } func TestCLIExecutePersistentProcessExitReturnsError(t *testing.T) { if runtime.GOOS == "windows" { t.Skip("PTY not supported on Windows") } cfg := config.CLIConf{ Enabled: true, Profiles: map[string]config.CLIProfileConf{ "exit-on-input": { Command: "sh", Args: []string{"-c", `stty -echo; while IFS= read -r line; do printf "before-exit\n"; exit 2; done`}, Persistent: true, Terminal: true, ResponseIdleTimeoutMS: 500, StartupIdleTimeoutMS: 50, }, }, } c := New(cfg, zap.NewNop()) ctx := context.Background() if err := c.Start(ctx); err != nil { t.Fatalf("start: %v", err) } defer func() { _ = c.Stop(ctx) }() sink := &fakeSink{} err := c.Execute(ctx, noderuntime.ExecutionSpec{ RunID: "run-1", Model: "exit-on-input", Input: map[string]any{"prompt": "hello"}, }, sink) if err == nil { t.Fatal("expected non-nil error when persistent process exits") } events := sink.Events() var hasError bool for _, e := range events { if e.Type == noderuntime.EventTypeError { hasError = true } } if !hasError { t.Fatal("expected error event in emitted events") } if len(events) > 0 && events[len(events)-1].Type == noderuntime.EventTypeComplete { t.Fatal("last event should not be complete when process exits unexpectedly") } } func TestCLIStartPersistentReturnsErrorWhenProcessExitsDuringStartup(t *testing.T) { if runtime.GOOS == "windows" { t.Skip("PTY not supported on Windows") } cfg := config.CLIConf{ Enabled: true, Profiles: map[string]config.CLIProfileConf{ "exits-immediately": { Command: "sh", Args: []string{"-c", "exit 2"}, Persistent: true, Terminal: true, StartupIdleTimeoutMS: 50, }, }, } c := New(cfg, zap.NewNop()) ctx := context.Background() err := c.Start(ctx) if err == nil { t.Fatal("expected error when persistent process exits during startup") } } func TestCLIStartCleansUpStartedProfilesWhenLaterProfileFails(t *testing.T) { if runtime.GOOS == "windows" { t.Skip("Unix shell required") } cfg := config.CLIConf{ Enabled: true, Profiles: map[string]config.CLIProfileConf{ "ok-profile": { Command: "sh", Args: []string{"-c", `while IFS= read -r line; do printf "reply:%s\n" "$line"; done`}, Persistent: true, Terminal: false, StartupIdleTimeoutMS: 50, ResponseIdleTimeoutMS: 200, }, "fail-profile": { Command: "sh", Args: []string{"-c", "exit 2"}, Persistent: true, Terminal: true, StartupIdleTimeoutMS: 50, }, }, } c := New(cfg, zap.NewNop()) ctx := context.Background() if err := c.Start(ctx); err == nil { t.Fatal("expected Start to fail when one persistent profile exits immediately") } // Stop must not panic after partial start failure. _ = c.Stop(ctx) // Second Stop must also not panic (idempotency). _ = c.Stop(ctx) // Sessions must be cleared — Execute should return a no-session error for every profile. for _, model := range []string{"ok-profile", "fail-profile"} { sink := &fakeSink{} if err := c.Execute(ctx, noderuntime.ExecutionSpec{ RunID: "post-cleanup", Model: model, Input: map[string]any{"prompt": "test"}, }, sink); err == nil { t.Fatalf("expected Execute for %q to fail after cleanup, got nil", model) } } } func TestCLIStartPersistentTerminalAndExecuteWritesPrompt(t *testing.T) { if runtime.GOOS == "windows" { t.Skip("PTY not supported on Windows") } cfg := config.CLIConf{ Enabled: true, Profiles: map[string]config.CLIProfileConf{ "echo-persistent": { Command: "sh", Args: []string{"-c", `stty -echo; while IFS= read -r line; do printf "reply:%s\n" "$line"; done`}, Persistent: true, Terminal: true, ResponseIdleTimeoutMS: 300, StartupIdleTimeoutMS: 50, }, }, } c := New(cfg, zap.NewNop()) ctx := context.Background() if err := c.Start(ctx); err != nil { t.Fatalf("start: %v", err) } defer func() { _ = c.Stop(ctx) }() sink := &fakeSink{} err := c.Execute(ctx, noderuntime.ExecutionSpec{ RunID: "run-1", Model: "echo-persistent", Input: map[string]any{"prompt": "hello"}, }, sink) if err != nil { t.Fatalf("execute: %v", err) } events := sink.Events() var started, completed bool var deltas []string for _, e := range events { switch e.Type { case noderuntime.EventTypeStart: started = true case noderuntime.EventTypeComplete: completed = true case noderuntime.EventTypeDelta: deltas = append(deltas, e.Delta) } } if !started { t.Fatal("expected start event") } if !completed { t.Fatal("expected complete event") } combined := strings.Join(deltas, "") if !strings.Contains(combined, "reply:hello") { t.Fatalf("expected reply:hello in deltas, got %q", combined) } } // timeoutOnFirstRun ensures the context cancels during a persistent execute, // then verifies that late output from the first run does not leak into // a second run for the same profile. func TestCLIExecutePersistentTimeoutCleansSession(t *testing.T) { if runtime.GOOS == "windows" { t.Skip("PTY not supported on Windows") } cfg := config.CLIConf{ Enabled: true, Profiles: map[string]config.CLIProfileConf{ "slow-echo": { Command: "sh", Args: []string{"-c", `stty -echo; while IFS= read -r line; do sleep 0.1; printf "reply:%s\n" "$line"; done`}, Persistent: true, Terminal: true, ResponseIdleTimeoutMS: 5000, // long idle so output doesn't timeout naturally StartupIdleTimeoutMS: 50, }, }, } c := New(cfg, zap.NewNop()) ctx := context.Background() if err := c.Start(ctx); err != nil { t.Fatalf("start: %v", err) } defer func() { _ = c.Stop(ctx) }() // First run: timeout after 200ms firstCtx, firstCancel := context.WithTimeout(ctx, 200*time.Millisecond) defer firstCancel() sink1 := &fakeSink{} err := c.Execute(firstCtx, noderuntime.ExecutionSpec{ RunID: "run-1", Model: "slow-echo", Input: map[string]any{"prompt": "first"}, }, sink1) // We expect a context error (timeout or cancelled) if err == nil { t.Fatal("expected context error on first timeout execute") } // Give the background goroutine time to process any late output time.Sleep(300 * time.Millisecond) // Second run: should fail with no-session error because the session was // cleaned up after the first run's timeout. sink2 := &fakeSink{} err = c.Execute(ctx, noderuntime.ExecutionSpec{ RunID: "run-2", Model: "slow-echo", Input: map[string]any{"prompt": "second"}, }, sink2) if err == nil { t.Fatal("expected no-session error on second execute after timeout") } // Verify that the second run did NOT emit any delta events (no contamination) events2 := sink2.Events() for _, e := range events2 { if e.Type == noderuntime.EventTypeDelta { t.Fatalf("unexpected delta event in second run (session contamination): %q", e.Delta) } } } // TestCLIExecutePersistentConsecutiveExecutes verifies that the same persistent // profile can be executed multiple times in a row — the second Execute must // complete (emit a Complete event) without being blocked by the first run's // session mutex. func TestCLIExecutePersistentConsecutiveExecutes(t *testing.T) { if runtime.GOOS == "windows" { t.Skip("PTY not supported on Windows") } cfg := config.CLIConf{ Enabled: true, Profiles: map[string]config.CLIProfileConf{ "consecutive-echo": { Command: "sh", Args: []string{"-c", `stty -echo; while IFS= read -r line; do printf "reply:%s\n" "$line"; done`}, Persistent: true, Terminal: true, ResponseIdleTimeoutMS: 100, // short idle so both runs complete quickly StartupIdleTimeoutMS: 50, }, }, } c := New(cfg, zap.NewNop()) ctx := context.Background() if err := c.Start(ctx); err != nil { t.Fatalf("start: %v", err) } defer func() { _ = c.Stop(ctx) }() // First execute sink1 := &fakeSink{} err := c.Execute(ctx, noderuntime.ExecutionSpec{ RunID: "run-1", Model: "consecutive-echo", Input: map[string]any{"prompt": "first"}, }, sink1) if err != nil { t.Fatalf("first execute: %v", err) } events1 := sink1.Events() var lastType1 noderuntime.EventType for _, e := range events1 { lastType1 = e.Type } if lastType1 != noderuntime.EventTypeComplete { t.Fatalf("first run: expected last event to be complete, got %q", lastType1) } // Second execute on the same profile — must also complete without blocking. sink2 := &fakeSink{} err = c.Execute(ctx, noderuntime.ExecutionSpec{ RunID: "run-2", Model: "consecutive-echo", Input: map[string]any{"prompt": "second"}, }, sink2) if err != nil { t.Fatalf("second execute: %v", err) } events2 := sink2.Events() var lastType2 noderuntime.EventType for _, e := range events2 { lastType2 = e.Type } if lastType2 != noderuntime.EventTypeComplete { t.Fatalf("second run: expected last event to be complete, got %q", lastType2) } // Verify second run did NOT receive deltas from the first run. for _, e := range events2 { if e.Type == noderuntime.EventTypeDelta && strings.Contains(e.Delta, "first") { t.Fatalf("second run received first run's delta (session contamination): %q", e.Delta) } } } // TestCLIStartDeterministicOrder verifies that Start iterates profiles in // a deterministic order by checking that cleanup after a failure stops // the earlier profiles regardless of map iteration order. func TestCLIStartPartialRollbackChecksMarker(t *testing.T) { if runtime.GOOS == "windows" { t.Skip("Unix shell required") } // Use a profile that logs a start marker and waits for input, so it stays alive. cfg := config.CLIConf{ Enabled: true, Profiles: map[string]config.CLIProfileConf{ "alive-profile": { Command: "sh", Args: []string{"-c", `while IFS= read -r line; do printf "reply:%s\n" "$line"; done`}, Persistent: true, Terminal: false, StartupIdleTimeoutMS: 50, ResponseIdleTimeoutMS: 200, }, "fail-profile": { Command: "sh", Args: []string{"-c", "exit 2"}, Persistent: true, Terminal: true, StartupIdleTimeoutMS: 50, }, }, } c := New(cfg, zap.NewNop()) ctx := context.Background() if err := c.Start(ctx); err == nil { t.Fatal("expected Start to fail when one persistent profile exits immediately") } // Stop must not panic after partial start failure. _ = c.Stop(ctx) // All sessions must be cleared — both profiles should return no-session error. for _, model := range []string{"alive-profile", "fail-profile"} { sink := &fakeSink{} if err := c.Execute(ctx, noderuntime.ExecutionSpec{ RunID: "post-cleanup", Model: model, Input: map[string]any{"prompt": "test"}, }, sink); err == nil { t.Fatalf("expected Execute for %q to fail after cleanup, got nil", model) } } }