cthulhu/internal/app/braille_session.go

package app

import (
	"bufio"
	"bytes"
	"context"
	"encoding/json"
	"errors"
	"fmt"
	"io"
	"log/slog"
	"os/exec"
	"strings"
	"sync"
	"time"

	"git.stormux.org/storm/cthulhu/internal/braille"
)

// A helper connection attempt blocks inside BRLTTY until BRLTTY answers,
// which is exactly when a display becomes available: BRLTTY stalls BrlAPI
// while it looks for a switched-off (e.g. Bluetooth) display, and there is
// no cheaper presence query. The session therefore waits with the single
// helper instead of killing and respawning it. brailleReadyGiveUp only
// recycles a permanently wedged helper (e.g. BRLTTY died mid-handshake).
const (
	brailleReadyGiveUp = 5 * time.Minute
	brailleReadyNotice = 10 * time.Second
)

const brailleMaxBackoff = 30 * time.Second

// nextBrailleBackoff grows the delay between helper connection attempts so
// an absent display (switched off, BRLTTY not answering) is retried on a
// slow cadence instead of respawning the helper every few seconds forever.
func nextBrailleBackoff(current time.Duration) time.Duration {
	if current <= 0 {
		return time.Second
	}
	return min(current*2, brailleMaxBackoff)
}

func (worker *brailleWorker) run(ctx context.Context, done chan struct{}) {
	defer close(done)
	backoff := time.Duration(0)
	lastError := ""
	for ctx.Err() == nil {
		connectedFor, err := worker.session(ctx)
		// Only a healthy, established connection resets retry history. Waiting
		// for the helper's ready message can itself take several minutes.
		if connectedFor > 10*time.Second {
			backoff = 0
			lastError = ""
		}
		if err != nil && ctx.Err() == nil {
			if err.Error() != lastError {
				worker.logger.Warn("braille helper disconnected", "error", err)
				lastError = err.Error()
			} else {
				worker.logger.Debug("braille helper still unavailable", "error", err)
			}
		}
		backoff = nextBrailleBackoff(backoff)
		select {
		case <-ctx.Done():
			return
		case <-time.After(backoff):
		}
	}
}

// connectedFor excludes startup waits so the supervisor can distinguish an
// established connection from a slow failed handshake.
func (worker *brailleWorker) session(ctx context.Context) (connectedFor time.Duration, result error) {
	var connectedAt time.Time
	defer func() {
		if !connectedAt.IsZero() {
			connectedFor = time.Since(connectedAt)
		}
	}()
	defer func() { worker.mu.Lock(); worker.connected = false; worker.mu.Unlock() }()
	childCtx, cancel := context.WithCancel(ctx)
	defer cancel()
	command := exec.CommandContext(childCtx, worker.path, "--client")
	stdin, err := command.StdinPipe()
	if err != nil {
		return 0, err
	}
	stdout, err := command.StdoutPipe()
	if err != nil {
		return 0, err
	}
	var stderr bytes.Buffer
	command.Stderr = &stderr
	if err := command.Start(); err != nil {
		return 0, err
	}
	waitDone := make(chan struct{})
	var waitErr error
	go func() { defer close(waitDone); waitErr = command.Wait() }()
	var ioWorkers sync.WaitGroup
	// Kill the disposable native owner, close pipes to unblock I/O, then join
	// every goroutine before permitting reconnect or a new configured session.
	defer func() {
		cancel()
		_ = stdin.Close()
		_ = stdout.Close()
		<-waitDone
		ioWorkers.Wait()
		if result != nil {
			if detail := strings.TrimSpace(stderr.String()); detail != "" {
				result = fmt.Errorf("%w: %s", result, detail)
			} else if waitErr != nil {
				result = fmt.Errorf("%w: %v", result, waitErr)
			}
		}
	}()
	ready := make(chan error, 1)
	readErrors := make(chan error, 1)
	// Routing keys retain FIFO order. Backpressure stops this pipe reader at
	// 16 pending keys; cancellation releases it and retires the session queue.
	routes := make(chan braille.RoutingEvent, 16)
	ioWorkers.Add(1)
	go func() {
		defer ioWorkers.Done()
		scanner := bufio.NewScanner(stdout)
		scanner.Buffer(make([]byte, 1024), 1024)
		if !scanner.Scan() || scanner.Text() != "ready" {
			ready <- errors.New("braille helper sent an invalid ready message")
			return
		}
		ready <- nil
		for scanner.Scan() {
			var event braille.RoutingEvent
			if err := json.Unmarshal(scanner.Bytes(), &event); err != nil {
				readErrors <- errors.New("invalid braille routing event")
				return
			}
			select {
			case routes <- event:
			case <-childCtx.Done():
				return
			}
		}
		err := scanner.Err()
		if err == nil {
			err = io.EOF
		}
		readErrors <- err
	}()
	giveUp := worker.readyGiveUp
	if giveUp <= 0 {
		giveUp = brailleReadyGiveUp
	}
	giveUpTimer := time.NewTimer(giveUp)
	defer giveUpTimer.Stop()
	notice := time.NewTimer(brailleReadyNotice)
	defer notice.Stop()
readyWait:
	for {
		select {
		case err := <-ready:
			if err != nil {
				return 0, err
			}
			break readyWait
		case <-notice.C:
			worker.logger.Info("waiting for braille display")
		case <-giveUpTimer.C:
			return 0, errors.New("braille helper connection timed out")
		case <-ctx.Done():
			return 0, nil
		}
	}
	connectedAt = time.Now()
	worker.logger.Info("braille helper connected")
	worker.mu.Lock()
	worker.connected = true
	worker.mu.Unlock()
	writeError := make(chan error, 1)
	ioWorkers.Add(1)
	go func() {
		defer ioWorkers.Done()
		encoder := json.NewEncoder(stdin)
		for {
			frame := worker.current()
			if err := encoder.Encode(frame); err != nil {
				writeError <- err
				return
			}
			// This confirms the helper write, not the physical display. Text is
			// the uncontracted line; cursor offsets count Unicode characters.
			if worker.logger.Enabled(context.Background(), slog.LevelDebug) {
				worker.logger.Debug("sent braille frame", "frame_id", frame.ID,
					"braille_text", frame.Text, "cursor_offset", frame.Cursor,
					"contracted", frame.Contract, "table", frame.Table,
					"word_wrap", frame.WordWrap, "expand_at_cursor", frame.ExpandAtCursor)
			}
			select {
			case <-childCtx.Done():
				return
			case <-worker.wake:
			}
		}
	}()
	for {
		select {
		case <-waitDone:
			return 0, waitErr
		case err := <-writeError:
			return 0, err
		case err := <-readErrors:
			return 0, err
		case event := <-routes:
			worker.route(childCtx, event)
		case <-ctx.Done():
			return 0, nil
		}
	}
}