State replay & crash recovery

TL;DR. The intended journal-on-disk shape that turns crash recovery into mechanical replay: per-pane append-only PTY byte logs, capped, fed back into fresh libghostty Terminals on --recover. Not yet implemented; the bytes-on-wire choice (ADR-0013) makes the eventual implementation routine.

---

Status: Design intent. Not yet implemented as of 2026-05-26. Nothing in the server currently writes to disk; server.pid, per-pane journals, and --recover do not exist. The bytes-on-wire shape (ADR-0013) makes the implementation mechanical when its turn comes — the PTY byte stream we forward to clients is also exactly what a journal would record.

The intended shape: the server journals raw PTY output to disk, per pane, in journal/<pane_id>.log. Journals are append-only, fsync’d on close, and capped (default: 10 MB ring per pane).

On startup, if server.pid is stale, the server can be invoked with --recover. It reads each journal, replays it into a fresh libghostty_vt::Terminal, and reconstitutes sessions from a metadata file alongside the journals.

Crash recovery is therefore a property of the design, not an add-on. tmux loses everything on a daemon crash; phux will not.