ctask/internal/session/adopt.go

package session

import (
	"fmt"
	"os"
	"path/filepath"
	"time"

	"github.com/warrenronsiek/ctask/internal/lockfile"
	"github.com/warrenronsiek/ctask/internal/shell"
	"github.com/warrenronsiek/ctask/internal/workspace"
)

// adoptAttacher and adoptPoll are test seams that *wrap* shell primitives
// rather than hand-rolling tmux invocations. Production calls go through
// shell.AttachSession / shell.PollSessionEnd. Tests override these
// variables; do not run such tests with t.Parallel().
var (
	adoptAttacher = shell.AttachSession
	adoptPoll     = shell.PollSessionEnd
)

// AdoptExistingPersistentSession is the adopted-reattach path. It is
// invoked when a tmux session for the workspace exists but the lease is
// missing, stale, or fresh-but-from-another-host (the cmd-layer dispatcher
// has already prompted for confirmation in the fresh_remote case before
// calling here — see cmd/persistent.go::confirmFreshRemoteAdoption).
//
// Eight-step flow (see v0.5.3-spec.md §3 path C):
//  1. Print one diagnostic line announcing the adoption.
//  2. Acquire write lock; under lock, re-read lease state. If now
//     fresh-local (concurrent adopter raced ahead), release the lock and
//     fall through to AttachExisting — the race winner owns the lease.
//     Otherwise, remove the orphaned lease, write a new lease for this
//     process, and bump task.yaml.UpdatedAt.
//  3. Capture a fresh start manifest. Without it, finalize has no
//     reliable diff baseline.
//  4. Start the heartbeat goroutine.
//  5. shell.AttachSession (returns nil on clean exit, error on non-zero).
//  6. Polling loop until tmux reports session gone — runs ONLY after a
//     successful attach.
//  7. Stop the heartbeat goroutine.
//  8. finalize with SessionOwnership="adopted" and AdoptedFromOrphanAt set.
//
// On attach failure (step 5 returns error), steps 6-8 are skipped and the
// error is returned — the user sees the underlying tmux failure.
func AdoptExistingPersistentSession(tmuxPath, sessionName, wsDir string, opts LaunchOpts) error {
	fmt.Fprintln(os.Stderr,
		"[ctask] adopting orphaned persistent session (previous owner exited without finalizing)")

	startTime := time.Now().UTC().Truncate(time.Second)
	adoptedAt := startTime
	leasePath := LeasePath(wsDir)

	var raced bool
	skipped, lockErr := lockfile.WithLock(
		ctaskWriteLockPath(wsDir),
		sessionWriteLockTimeout, sessionWriteLockStaleAfter,
		func() error {
			// Re-check under lock. If a concurrent adopter beat us to it,
			// fall through to passive reattach. No lease writes, no
			// task.yaml.UpdatedAt bump on this branch.
			if InspectLease(wsDir) == LeaseStateFreshLocal {
				raced = true
				return nil
			}
			if _, rmErr := CleanupStaleLease(leasePath, StaleLeaseAfter); rmErr != nil {
				fmt.Fprintf(os.Stderr,
					"[ctask] warning: failed to remove orphaned lease: %v\n", rmErr)
			}
			lease := NewLease(startTime, leaseAgentCommand(opts), opts.Mode)
			if err := WriteLease(leasePath, lease); err != nil {
				return fmt.Errorf("writing lease: %w", err)
			}
			// Adoption-only: bump task.yaml.UpdatedAt. Passive reattach
			// (and the race fall-through above) leave it untouched.
			metaPath := filepath.Join(wsDir, "task.yaml")
			if meta, err := workspace.ReadMeta(metaPath); err == nil && meta != nil {
				meta.UpdatedAt = startTime
				if err := workspace.WriteMeta(metaPath, meta); err != nil {
					fmt.Fprintf(os.Stderr,
						"[ctask] warning: failed to update task.yaml on adoption: %v\n", err)
				}
			}
			return nil
		},
	)
	if skipped {
		fmt.Fprintf(os.Stderr,
			"[ctask] Warning: could not acquire metadata lock; falling through to passive reattach\n")
		return AttachExisting(tmuxPath, sessionName)
	}
	if lockErr != nil {
		return fmt.Errorf("adoption lease write: %w", lockErr)
	}
	if raced {
		return AttachExisting(tmuxPath, sessionName)
	}

	// Step 3: fresh start manifest — load-bearing for finalize's diff baseline.
	startManifest, err := CaptureManifest(wsDir)
	if err != nil {
		fmt.Fprintf(os.Stderr,
			"[ctask] warning: failed to capture start manifest during adoption: %v; finalize diff will be skipped\n", err)
	} else {
		mPath := manifestStartPath(wsDir)
		if _, lockErr := lockfile.WithLock(
			ctaskWriteLockPath(wsDir),
			sessionWriteLockTimeout, sessionWriteLockStaleAfter,
			func() error { return WriteManifest(mPath, startManifest) },
		); lockErr != nil {
			fmt.Fprintf(os.Stderr,
				"[ctask] warning: failed to write start manifest: %v\n", lockErr)
			startManifest = nil
		}
	}

	// Step 4: heartbeat.
	hb := StartHeartbeat(leasePath, HeartbeatInterval)

	// Step 5: attach. Non-zero exit is a real failure; surface it
	// and skip steps 6-8 (polling and finalize). Stop the heartbeat first
	// so we don't leak the goroutine.
	if attachErr := adoptAttacher(tmuxPath, sessionName); attachErr != nil {
		hb.Stop()
		return attachErr
	}

	// Step 6: poll until session ends.
	adoptPoll(tmuxPath, sessionName, shell.PollInterval)

	// Step 7: stop heartbeat.
	hb.Stop()

	// Step 8: finalize with adoption fields.
	endTime := time.Now().UTC().Truncate(time.Second)
	if startManifest != nil {
		if err := finalizeAdopted(opts, wsDir, startManifest, startTime, endTime, adoptedAt); err != nil {
			fmt.Fprintf(os.Stderr, "[ctask] warning: adoption finalize failed: %v\n", err)
		}
	}
	return nil
}

// finalizeAdopted is the adoption-specific finalize path. Mirrors
// session.Run's finalize() but stamps SessionOwnership = "adopted" and
// AdoptedFromOrphanAt onto the summary.
func finalizeAdopted(opts LaunchOpts, wsDir string, startManifest *Manifest, startTime, endTime, adoptedAt time.Time) error {
	endManifest, err := CaptureManifest(wsDir)
	if err != nil {
		return fmt.Errorf("capturing end manifest: %w", err)
	}
	diff := DiffManifests(startManifest, endManifest)

	agentCmd := leaseAgentCommand(opts)
	if opts.Shell {
		agentCmd = "shell"
	}
	info := &SessionInfo{
		Agent:     agentCmd,
		Mode:      opts.Mode,
		StartTime: startTime,
		EndTime:   endTime,
		Diff:      diff,
	}

	sessionID := NewSessionID(currentHostname(), os.Getpid(), startTime)
	if l, err := ReadLease(LeasePath(wsDir)); err == nil && l != nil {
		sessionID = l.SessionID
	}

	summary := SummarizeFromDiff(
		sessionID, currentHostname(), agentCmd, opts.Mode,
		startTime, endTime, diff, endManifest,
	)
	summary.EndReason = "tmux_session_ended"
	summary.DetectedVia = "polling"
	summary.SessionOwnership = "adopted"
	summary.AdoptedFromOrphanAt = &adoptedAt

	skipped, lockErr := lockfile.WithLock(
		ctaskWriteLockPath(wsDir),
		sessionWriteLockTimeout, sessionWriteLockStaleAfter,
		func() error {
			if err := AppendSessionLog(wsDir, info); err != nil {
				fmt.Fprintf(os.Stderr, "[ctask] warning: append session log failed: %v\n", err)
			}
			if err := WriteSummary(SummaryPath(wsDir), summary); err != nil {
				return fmt.Errorf("write summary: %w", err)
			}
			if rmErr := os.Remove(LeasePath(wsDir)); rmErr != nil && !os.IsNotExist(rmErr) {
				fmt.Fprintf(os.Stderr, "[ctask] warning: could not remove lease: %v\n", rmErr)
			}
			if rmErr := os.Remove(manifestStartPath(wsDir)); rmErr != nil && !os.IsNotExist(rmErr) {
				fmt.Fprintf(os.Stderr, "[ctask] warning: could not remove start manifest: %v\n", rmErr)
			}
			return nil
		},
	)
	if skipped {
		fmt.Fprintf(os.Stderr,
			"[ctask] Warning: could not acquire metadata lock at adoption finalize; skipping summary write\n")
		return nil
	}
	return lockErr
}