Root cause of the multi-focus/multi-search/linked-terminal bug: the in-memory
id counter resets to 0 each daemon start, but restore() never advanced it past
restored ids. After a restart new_surface_id() re-minted existing ids → the same
surface_id appeared twice in a layout tree (rendered as two panels sharing focus,
search bar, and output channel — one ends up blank). Session-persistence made
restarts routine, surfacing the latent bug.
- restore() now reseeds the counter to max(restored id)+1
- ops::dedupe_leaves heals an already-corrupted persisted tree on load
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Addresses final-review findings: Cmd::Shutdown now snapshots all live surfaces
synchronously before exit (spec graceful-shutdown requirement); StoppedSnapshot
calls detachSurface on unmount to release the bridge output channel.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Adds Cmd::ClearEvents + Evt::EventsCleared: the daemon drops the persistent
event log (keeping next_id monotonic), persists, and broadcasts so every
client empties its list. A red trash icon next to 'Mark all read' triggers it;
disabled when the list is empty. Threaded through proto, the daemon handler,
the Tauri bridge, and socketBridge. Includes an EventLog::clear test.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The daemon outlives the GUI, so after an update an OLD daemon can keep serving
the socket and the new GUI just connects to it (stale code — e.g. the missing
TERM fix). Both binaries are now stamped with the git build id (build.rs):
the daemon reports it in `health.build`, and on connect the bridge compares it
to the GUI's own SPACESH_BUILD; on mismatch it shuts the daemon down and lets
ensure_daemon respawn the bundled (matching) one. No-op for unstamped dev
builds or daemons too old to report a build. Build id is shown in Settings.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
A GUI/launchd-spawned daemon has no TERM in its environment, so child shells
inherited none and tput/zsh/ncurses failed ('tput: No value for $TERM').
The PTY now defaults TERM=xterm-256color and COLORTERM=truecolor (matching
xterm.js) unless the caller already provides them. Adds a regression test.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Add a `pinned` bool to the workspace model, threaded through proto
(Workspace + WorkspaceView + SetWorkspaceMeta), the registry, the
set_workspace_meta handler, persistence, the CLI mapping, and the Tauri
bridge. serde(default) keeps existing state.json compatible (pinned=false).
Backs the sidebar Favorites section.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
A new surface spawned its shell immediately at the default 80x24, so the
shell (e.g. zsh + powerlevel10k) printed its first prompt before the GUI
fit-resized the panel. The post-print resize reflowed the grid, leaving the
prompt invisible until Enter and shifted by a line.
The surface actor now defers the child spawn: it waits for the first Resize
(spawning at that geometry) or, for headless/CLI surfaces that never attach a
GUI, falls back to the spec geometry after 250ms. Attaches received before
the spawn get an empty-grid snapshot plus a live subscription; pre-spawn
input is buffered and replayed. Adds two deterministic tests via `stty size`.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Add a config.rs module that loads optional ~/.spacesh/config.toml and a
default_shell() resolver: SPACESH_SHELL env -> config.toml default_shell ->
login shell from the passwd DB -> $SHELL -> /bin/sh. The passwd lookup
matters under launchd where $SHELL is absent. All plain-panel spawn sites in
server.rs now use the resolver. Adds toml + libc deps.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
lock_is_exclusive_within_process acquired the global ~/.spacesh/daemon.lock,
so it flaked whenever a real daemon was running. Add a SPACESH_LOCK env
override to lock_path() and point the test at a private temp file under a
serial() guard, making the suite deterministic regardless of a live daemon.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Wire Cmd::EventLog and Cmd::MarkRead to the live EventLog (replacing
NOT_IMPLEMENTED stubs). Cmd::Focus now calls mark_read for the surface,
persists the updated snapshot, and broadcasts Evt::EventsRead.
Add integration tests: event_log_query_and_mark_read and
focus_marks_surface_events_read.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Add event_log.rs with EventLog (capped VecDeque, monotonic ids, mark_read,
snapshot/restore) and EventLogState. Register mod in main.rs. Stub
Cmd::EventLog and Cmd::MarkRead arms in server.rs to keep the exhaustive
match compiling; full wiring follows in Task 4.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
vasyansk
