iop/apps/node/internal/adapters/cli/cli_test.go

package cli_test

import (
	"context"
	"fmt"
	"os"
	"path/filepath"
	"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 (sorted) order by checking that cleanup after a failure stops
// the earlier profiles regardless of map iteration order.
func TestCLIStartDeterministicOrder(t *testing.T) {
	if runtime.GOOS == "windows" {
		t.Skip("Unix shell required")
	}

	// "Z-alive" sorts before "Z-fail", so if deterministic order works,
	// Z-alive should be cleaned up when Z-fail fails.
	cfg := config.CLIConf{
		Enabled: true,
		Profiles: map[string]config.CLIProfileConf{
			"Z-alive": {
				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,
			},
			"Z-fail": {
				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)

	// All sessions must be cleared — both profiles should return no-session error.
	for _, model := range []string{"Z-alive", "Z-fail"} {
		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)
		}
	}
}

// TestCLIConcurrentExecuteAndStop verifies that concurrent Execute and Stop
// calls do not deadlock or panic, and the session map is safely protected.
// Executes are launched concurrently but run sequentially due to session mutex;
// the test uses short context timeouts so it completes without hanging.
func TestCLIConcurrentExecuteAndStop(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 0.05; printf "reply:%s\n" "$line"; done`},
				Persistent:            true,
				Terminal:              true,
				ResponseIdleTimeoutMS: 50, // short idle so completes quickly
				StartupIdleTimeoutMS:  50,
			},
		},
	}
	c := New(cfg, zap.NewNop())

	ctx := context.Background()
	if err := c.Start(ctx); err != nil {
		t.Fatalf("start: %v", err)
	}

	var wg sync.WaitGroup
	errCh := make(chan error, 10)

	// Goroutine 1: multiple sequential Execute calls (launched sequentially to
	// avoid session mutex deadlock from concurrent calls on same profile).
	wg.Add(1)
	go func() {
		defer wg.Done()
		for i := 0; i < 3; i++ {
			sink := &fakeSink{}
			execCtx, cancel := context.WithTimeout(ctx, 500*time.Millisecond)
			err := c.Execute(execCtx, noderuntime.ExecutionSpec{
				RunID: fmt.Sprintf("run-conc-%d", i),
				Model: "slow-echo",
				Input: map[string]any{"prompt": fmt.Sprintf("conc-%d", i)},
			}, sink)
			cancel()
			if err != nil {
				errCh <- fmt.Errorf("concurrent execute %d: %w", i, err)
			}
		}
	}()

	// Goroutine 2: Stop while executes are running
	wg.Add(1)
	go func() {
		defer wg.Done()
		time.Sleep(200 * time.Millisecond)
		_ = c.Stop(ctx)
	}()

	// Goroutine 3: more Execute after Stop (should get no-session error).
	wg.Add(1)
	go func() {
		defer wg.Done()
		time.Sleep(300 * time.Millisecond)
		for i := 0; i < 2; i++ {
			sink := &fakeSink{}
			execCtx, cancel := context.WithTimeout(ctx, 200*time.Millisecond)
			err := c.Execute(execCtx, noderuntime.ExecutionSpec{
				RunID: fmt.Sprintf("run-post-stop-%d", i),
				Model: "slow-echo",
				Input: map[string]any{"prompt": fmt.Sprintf("post-stop-%d", i)},
			}, sink)
			cancel()
			if err != nil {
				errCh <- fmt.Errorf("post-stop execute %d: %w", i, err)
			}
		}
	}()

	done := make(chan struct{})
	go func() {
		wg.Wait()
		close(done)
	}()

	select {
	case <-done:
		close(errCh)
		var errs []error
		for err := range errCh {
			errs = append(errs, err)
		}
		// It's expected that some executes fail due to context timeout or no-session.
		// The key is that no goroutine deadlocks or panics.
		_ = errs
	case <-time.After(10 * time.Second):
		t.Fatal("TestCLIConcurrentExecuteAndStop timed out — possible deadlock")
	}
}

// requireEventually polls fn until it returns true or the deadline exceeds.
// It helps flaky timing-dependent assertions such as process cleanup markers.
func requireEventually(t *testing.T, timeout, interval time.Duration, fn func() bool) {
	t.Helper()
	timer := time.NewTimer(timeout)
	defer timer.Stop()
	ticker := time.NewTicker(interval)
	defer ticker.Stop()
	for {
		if fn() {
			return
		}
		select {
		case <-timer.C:
			t.Fatal("condition not satisfied within timeout")
		case <-ticker.C:
		}
	}
}

// readMarker returns the content of a marker file or an empty string if missing.
func readMarker(path string) string {
	data, err := os.ReadFile(path)
	if err != nil {
		return ""
	}
	return strings.TrimSpace(string(data))
}

// TestCLIStartPartialRollbackWithMarkers verifies that when a persistent profile
// fails during startup, the previously started profiles are actually cleaned up
// by using START_MARKER and EXIT_MARKER in their output.
//
// Note: SIGKILL (used by Stop/Kill) cannot be caught by trap handlers, so we send
// SIGTERM first via a separate signal to trigger the EXIT trap. The test sends a
// prompt to profile-a (which wakes it up) and then relies on stopAllSessions -> Kill()
// which sends SIGKILL. Since SIGKILL is uncatchable, we instead verify START_MARKER
// to prove the session was actually started and subsequently removed.
func TestCLIStartPartialRollbackWithMarkers(t *testing.T) {
	if runtime.GOOS == "windows" {
		t.Skip("Unix shell required")
	}

	markerDir := t.TempDir()
	startMarker := "START_MARKER_A"
	startMarkerPath := filepath.Join(markerDir, "marker_a")
	// First profile writes START_MARKER on start.
	// Uses a simple approach: write marker immediately on startup.
	profileACmd := fmt.Sprintf(`mkdir -p %s; echo %s > %s/marker_a; while IFS= read -r line; do printf "reply:%%s\n" "$line"; done`,
		markerDir, startMarker, markerDir)

	cfg := config.CLIConf{
		Enabled: true,
		Profiles: map[string]config.CLIProfileConf{
			"profile-a": {
				Command:               "sh",
				Args:                  []string{"-c", profileACmd},
				Persistent:            true,
				Terminal:              true, // PTY so the process starts properly
				StartupIdleTimeoutMS:  50,
				ResponseIdleTimeoutMS: 200,
			},
			"profile-b": {
				Command:              "sh",
				Args:                 []string{"-c", "exit 2"},
				Persistent:           true,
				Terminal:             true,
				StartupIdleTimeoutMS: 50,
			},
		},
	}
	c := New(cfg, zap.NewNop())
	ctx := context.Background()

	// Start should fail because profile-b exits immediately.
	if err := c.Start(ctx); err == nil {
		t.Fatal("expected Start to fail when profile-b exits immediately")
	}

	// Stop must not panic after partial start failure.
	_ = c.Stop(ctx)

	// Give the cleanup goroutine time to finish.
	time.Sleep(300 * time.Millisecond)

	// Assertion 1: START_MARKER must exist (profile-a started successfully).
	requireEventually(t, 2*time.Second, 50*time.Millisecond, func() bool {
		content := readMarker(startMarkerPath)
		return content == startMarker
	})

	// Assertion 2: the session was removed — Execute should fail with no-session.
	sink := &fakeSink{}
	if err := c.Execute(ctx, noderuntime.ExecutionSpec{
		RunID: "post-rollback",
		Model: "profile-a",
		Input: map[string]any{"prompt": "test"},
	}, sink); err == nil {
		t.Fatal("expected Execute for profile-a to fail with no-session after rollback")
	}
}