feat(activity): use fsnotify for real-time feed (hq-ew1mbr.17)

Replace polling with filesystem watching for near-instant wake-up
(<50ms vs 250ms avg). Watches .beads/dolt/.dolt/noms for Dolt commits.
Falls back to polling if fsnotify unavailable.

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

cmd/bd/activity.go

@@ -152,7 +152,8 @@ func runActivityOnce(sinceTime time.Time) {
 	}
 }
 
-// runActivityFollow streams events in real-time
+// runActivityFollow streams events in real-time using filesystem watching.
+// Falls back to polling if fsnotify is not available.
 func runActivityFollow(sinceTime time.Time) {
 	// Start from now if no --since specified
 	lastPoll := time.Now().Add(-1 * time.Second)
@@ -181,9 +182,19 @@ func runActivityFollow(sinceTime time.Time) {
 		}
 	}
 
-	// Poll for new events
-	ticker := time.NewTicker(activityInterval)
-	defer ticker.Stop()
+	// Create filesystem watcher for near-instant wake-up
+	// Falls back to polling internally if fsnotify fails
+	beadsDir := filepath.Dir(dbPath)
+	watcher, err := NewActivityWatcher(beadsDir, activityInterval)
+	if err != nil {
+		// Watcher creation failed entirely - fall back to legacy polling
+		runActivityFollowPolling(sinceTime, lastPoll)
+		return
+	}
+	defer watcher.Close()
+
+	// Start watching
+	watcher.Start(rootCtx)
 
 	// Track consecutive failures for error reporting
 	consecutiveFailures := 0
@@ -194,7 +205,11 @@ func runActivityFollow(sinceTime time.Time) {
 		select {
 		case <-rootCtx.Done():
 			return
-		case <-ticker.C:
+		case _, ok := <-watcher.Events():
+			if !ok {
+				return // Watcher closed
+			}
+
 			newEvents, err := fetchMutations(lastPoll)
 			if err != nil {
 				consecutiveFailures++
@@ -246,6 +261,69 @@ func runActivityFollow(sinceTime time.Time) {
 	}
 }
 
+// runActivityFollowPolling is the legacy polling-based follow mode.
+// Used as fallback when ActivityWatcher cannot be created.
+func runActivityFollowPolling(sinceTime time.Time, lastPoll time.Time) {
+	ticker := time.NewTicker(activityInterval)
+	defer ticker.Stop()
+
+	consecutiveFailures := 0
+	const failureWarningThreshold = 5
+	lastWarningTime := time.Time{}
+
+	for {
+		select {
+		case <-rootCtx.Done():
+			return
+		case <-ticker.C:
+			newEvents, err := fetchMutations(lastPoll)
+			if err != nil {
+				consecutiveFailures++
+				if consecutiveFailures >= failureWarningThreshold {
+					if time.Since(lastWarningTime) >= 30*time.Second {
+						if jsonOutput {
+							errorEvent := map[string]interface{}{
+								"type":      "error",
+								"message":   fmt.Sprintf("daemon unreachable (%d failures)", consecutiveFailures),
+								"timestamp": time.Now().Format(time.RFC3339),
+							}
+							data, _ := json.Marshal(errorEvent)
+							fmt.Fprintln(os.Stderr, string(data))
+						} else {
+							timestamp := time.Now().Format("15:04:05")
+							fmt.Fprintf(os.Stderr, "[%s] %s daemon unreachable (%d consecutive failures)\n",
+								timestamp, ui.RenderWarn("!"), consecutiveFailures)
+						}
+						lastWarningTime = time.Now()
+					}
+				}
+				continue
+			}
+
+			if consecutiveFailures > 0 {
+				if consecutiveFailures >= failureWarningThreshold && !jsonOutput {
+					timestamp := time.Now().Format("15:04:05")
+					fmt.Fprintf(os.Stderr, "[%s] %s daemon reconnected\n", timestamp, ui.RenderPass("✓"))
+				}
+				consecutiveFailures = 0
+			}
+
+			newEvents = filterEvents(newEvents)
+			for _, e := range newEvents {
+				if jsonOutput {
+					data, _ := json.Marshal(formatEvent(e))
+					fmt.Println(string(data))
+				} else {
+					printEvent(e)
+				}
+				if e.Timestamp.After(lastPoll) {
+					lastPoll = e.Timestamp
+				}
+			}
+		}
+	}
+}
+
 // fetchMutations retrieves mutations from the daemon
 func fetchMutations(since time.Time) ([]rpc.MutationEvent, error) {
 	var sinceMillis int64

cmd/bd/activity_watcher.go

@@ -0,0 +1,234 @@
+package main
+
+import (
+	"context"
+	"os"
+	"path/filepath"
+	"sync"
+	"time"
+
+	"github.com/fsnotify/fsnotify"
+)
+
+// ActivityWatcher monitors the beads directory for changes using filesystem events.
+// Falls back to polling if fsnotify fails (some filesystems don't support it).
+type ActivityWatcher struct {
+	watcher      *fsnotify.Watcher
+	watchPaths   []string     // Paths being watched
+	pollingMode  bool         // True if using polling fallback
+	pollInterval time.Duration
+	events       chan struct{} // Sends wake-up signals on changes
+	cancel       context.CancelFunc
+	wg           sync.WaitGroup
+	mu           sync.Mutex
+
+	// Polling state
+	lastModTimes map[string]time.Time
+}
+
+// NewActivityWatcher creates a watcher for activity feed updates.
+// Watches the dolt noms directory for commits, falling back to polling if fsnotify fails.
+// The beadsDir should be the .beads directory path.
+// The pollInterval is used for polling fallback mode.
+func NewActivityWatcher(beadsDir string, pollInterval time.Duration) (*ActivityWatcher, error) {
+	aw := &ActivityWatcher{
+		pollInterval: pollInterval,
+		events:       make(chan struct{}, 1), // Buffered to avoid blocking
+		lastModTimes: make(map[string]time.Time),
+	}
+
+	// Determine watch paths - prefer dolt noms directory if it exists
+	doltNomsPath := filepath.Join(beadsDir, "dolt", ".dolt", "noms")
+	doltPath := filepath.Join(beadsDir, "dolt", ".dolt")
+	jsonlPath := filepath.Join(beadsDir, "issues.jsonl")
+
+	// Build list of paths to watch (in priority order)
+	var watchPaths []string
+	if stat, err := os.Stat(doltNomsPath); err == nil && stat.IsDir() {
+		// Watch dolt noms directory for commits
+		watchPaths = append(watchPaths, doltNomsPath)
+	} else if stat, err := os.Stat(doltPath); err == nil && stat.IsDir() {
+		// Fallback to .dolt directory
+		watchPaths = append(watchPaths, doltPath)
+	}
+	// Also watch JSONL for non-dolt or hybrid setups
+	if _, err := os.Stat(jsonlPath); err == nil {
+		watchPaths = append(watchPaths, jsonlPath)
+	}
+	// Watch the beads dir itself as last resort
+	if len(watchPaths) == 0 {
+		watchPaths = append(watchPaths, beadsDir)
+	}
+
+	aw.watchPaths = watchPaths
+
+	// Initialize modification times for polling
+	for _, p := range watchPaths {
+		if stat, err := os.Stat(p); err == nil {
+			aw.lastModTimes[p] = stat.ModTime()
+		}
+	}
+
+	// Try to create fsnotify watcher
+	watcher, err := fsnotify.NewWatcher()
+	if err != nil {
+		// Fall back to polling mode
+		aw.pollingMode = true
+		return aw, nil
+	}
+
+	// Add watches for each path
+	watchedAny := false
+	for _, p := range watchPaths {
+		if err := watcher.Add(p); err != nil {
+			// Log but continue - some paths may not be watchable
+			continue
+		}
+		watchedAny = true
+	}
+
+	if !watchedAny {
+		// No paths could be watched, fall back to polling
+		_ = watcher.Close()
+		aw.pollingMode = true
+		return aw, nil
+	}
+
+	aw.watcher = watcher
+	return aw, nil
+}
+
+// Events returns the channel that receives wake-up signals when changes are detected.
+// The channel sends an empty struct for each detected change (debounced).
+func (aw *ActivityWatcher) Events() <-chan struct{} {
+	return aw.events
+}
+
+// IsPolling returns true if the watcher is using polling fallback.
+func (aw *ActivityWatcher) IsPolling() bool {
+	return aw.pollingMode
+}
+
+// Start begins monitoring for changes.
+// Returns immediately, monitoring happens in background goroutine.
+func (aw *ActivityWatcher) Start(ctx context.Context) {
+	ctx, cancel := context.WithCancel(ctx)
+	aw.cancel = cancel
+
+	if aw.pollingMode {
+		aw.startPolling(ctx)
+	} else {
+		aw.startFSWatch(ctx)
+	}
+}
+
+// startFSWatch starts fsnotify-based watching.
+func (aw *ActivityWatcher) startFSWatch(ctx context.Context) {
+	aw.wg.Add(1)
+	go func() {
+		defer aw.wg.Done()
+
+		// Debounce: don't send more than one event per 50ms
+		var lastEvent time.Time
+		debounceWindow := 50 * time.Millisecond
+
+		for {
+			select {
+			case event, ok := <-aw.watcher.Events:
+				if !ok {
+					return
+				}
+
+				// Only trigger on write events
+				if event.Op&fsnotify.Write == 0 && event.Op&fsnotify.Create == 0 {
+					continue
+				}
+
+				// Debounce rapid events
+				now := time.Now()
+				if now.Sub(lastEvent) < debounceWindow {
+					continue
+				}
+				lastEvent = now
+
+				// Send non-blocking wake-up signal
+				select {
+				case aw.events <- struct{}{}:
+				default:
+					// Channel already has a pending event
+				}
+
+			case _, ok := <-aw.watcher.Errors:
+				if !ok {
+					return
+				}
+				// Log errors but continue watching
+
+			case <-ctx.Done():
+				return
+			}
+		}
+	}()
+}
+
+// startPolling starts polling-based change detection.
+func (aw *ActivityWatcher) startPolling(ctx context.Context) {
+	aw.wg.Add(1)
+	go func() {
+		defer aw.wg.Done()
+
+		ticker := time.NewTicker(aw.pollInterval)
+		defer ticker.Stop()
+
+		for {
+			select {
+			case <-ticker.C:
+				if aw.checkForChanges() {
+					// Send non-blocking wake-up signal
+					select {
+					case aw.events <- struct{}{}:
+					default:
+					}
+				}
+
+			case <-ctx.Done():
+				return
+			}
+		}
+	}()
+}
+
+// checkForChanges checks if any watched paths have been modified.
+func (aw *ActivityWatcher) checkForChanges() bool {
+	aw.mu.Lock()
+	defer aw.mu.Unlock()
+
+	changed := false
+	for _, p := range aw.watchPaths {
+		stat, err := os.Stat(p)
+		if err != nil {
+			continue
+		}
+
+		lastMod, exists := aw.lastModTimes[p]
+		if !exists || !stat.ModTime().Equal(lastMod) {
+			aw.lastModTimes[p] = stat.ModTime()
+			changed = true
+		}
+	}
+	return changed
+}
+
+// Close stops the watcher and releases resources.
+func (aw *ActivityWatcher) Close() error {
+	if aw.cancel != nil {
+		aw.cancel()
+	}
+	aw.wg.Wait()
+	close(aw.events)
+	if aw.watcher != nil {
+		return aw.watcher.Close()
+	}
+	return nil
+}
+