feat: add leave_hook symmetric to enter_hook by tyvsmith · Pull Request #448 · feschber/lan-mouse

tyvsmith · 2026-05-26T23:34:20Z

Summary

Mirrors the existing enter_hook mechanism with a leave_hook that fires when capture for a client is released.

Motivation: external scripts that change host state on cross (e.g. swapping keyboard layout, toggling natural scroll, repositioning windows) currently have no symmetric "you're back" signal — enter_hook fires deterministically but the only way to detect release today is to poll hyprctl cursorpos / similar, which is fragile because the local cursor isn't actually pinned during capture on Wayland.

Design

A single new IPC capture event ICaptureEvent::ClientLeft(handle) emitted from release_capture() in src/capture.rs. Every release path funnels through that function:

release_bind chord
peer Leave event over the wire
CaptureRequest::Destroy when the active client is being torn down (with an explicit note about hostname-change client recreation)
CaptureRequest::Release

Wiring otherwise exactly mirrors enter_hook:

Layer	enter_hook	leave_hook
`src/config.rs`	`enter_hook: Option<String>`	+ `leave_hook: Option<String>`
`lan-mouse-ipc/src/lib.rs`	`enter_cmd` on `ClientConfig`, `UpdateEnterHook` request	+ `leave_cmd`, `UpdateLeaveHook`
`src/client.rs`	`set_enter_hook` / `get_enter_cmd`	+ `set_leave_hook` / `get_leave_cmd`
`src/service.rs`	`update_enter_hook`, `spawn_hook_command`, dispatch on `ClientEntered`	+ `update_leave_hook`, `spawn_leave_hook_command`, dispatch on `ClientLeft`
`lan-mouse-cli/src/lib.rs`	`--enter-hook` flag on `add-client`	+ `--leave-hook` flag

Docs

The existing enter_hook has been undocumented since #130. Second commit backfills both in the example config.toml so the per-client hook surface is discoverable.

Out of scope

GUI surface in lan-mouse-gtk for the new field (text-config only for now)
Unit tests (no test infrastructure exists in the project today; manual smoke-tested with notify-send hooks across crossing + release_bind release on Wayland/Hyprland)

Test plan

Builds clean: cargo build --release (1m 20s, no warnings)
cargo clippy --all-targets --release -- -D warnings clean
Manual smoke: enter_hook and leave_hook set to notify-send strings, daemon restarted from built binary, cross to client and release via chord — both notifications fire as expected
Verify behavior with peer-initiated Leave (cross-back from the remote side)
Verify behavior on cli deactivate of the active client

Per-client leave_hook fires sh -c <cmd> when capture for that client is released. Wired through a single release_capture() chokepoint covering release_bind, network send failure, Destroy-of-active-client, and Release request paths. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

The existing enter_hook field has been undocumented since it was added in feschber#130. Add it alongside the new leave_hook to the example config so both per-client hooks are discoverable. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Copilot

Pull request overview

Note

Copilot was unable to run its full agentic suite in this review.

Adds support for per-client “leave hooks” that run when capture is released for a client, wiring the new setting through config, IPC, CLI, and capture/service event handling.

Changes:

Introduce leave_hook / leave_cmd in config and IPC models, plus CLI flag support.
Emit a new ICaptureEvent::ClientLeft on release paths and handle it in the service.
Execute configured leave commands via sh -c when a client is left.

Reviewed changes

Copilot reviewed 7 out of 7 changed files in this pull request and generated 3 comments.

Show a summary per file

File	Description
src/service.rs	Handle `UpdateLeaveHook`, persist `leave_hook`, and spawn leave-hook commands on `ClientLeft`.
src/config.rs	Add `leave_hook` to TOML and runtime config conversion paths.
src/client.rs	Store and expose per-client `leave_cmd` in `ClientManager`.
src/capture.rs	Add `ClientLeft` event and ensure release paths trigger it.
lan-mouse-ipc/src/lib.rs	Extend IPC structs/enums with leave-hook fields and requests.
lan-mouse-cli/src/lib.rs	Add `--leave-hook` flag and send `UpdateLeaveHook` on create.
config.toml	Document and example-configure `leave_hook`.

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

tyvsmith · 2026-05-27T19:46:29Z

+            self.event_tx
+                .send(ICaptureEvent::ClientLeft(handle))
+                .expect("channel closed");


Leaving this as-is to stay consistent with the surrounding pattern: the two pre-existing event_tx.send(...).expect("channel closed") sites (CaptureBegin at L350 and ClientEntered at L371) use the same idiom, and request_rx.recv().expect("channel closed") at L303 follows the same convention. The event_tx receiver lives in the service task that owns the capture task — if it's gone the daemon is already tearing down. Changing only the new site would create a one-of-four inconsistency; if we want to soften this it should be a separate refactor across all four call sites.

tyvsmith · 2026-05-27T19:46:35Z

+            // fires and active_client is cleared (without this the
+            // active_client field would stay stale until the next
+            // Begin from a different handle).
+            self.release_capture(capture).await?;


The error-propagation surface doesn't actually change here. Inside release_capture() everything is either infallible-or-logged (the event_tx.send panics on closed channel like its siblings; the conn.send calls for key-up/modifier-reset/Leave all use if let Err(e) = ... { log::warn!(...) } — they never propagate). The only ?-propagated error is the final capture.release().await at the end, which is exactly what the old capture.release().await? returned. So this site propagates the same CaptureError as before — it just additionally fires the leave_hook and clears active_client along the way, both of which we want in this code path.

tyvsmith · 2026-05-27T19:46:36Z

+    fn spawn_leave_hook_command(&self, handle: ClientHandle) {
+        let Some(cmd) = self.client_manager.get_leave_cmd(handle) else {
+            return;
+        };
+        tokio::task::spawn_local(async move {
+            log::info!("spawning leave command!");
+            let mut child = match Command::new("sh").arg("-c").arg(cmd.as_str()).spawn() {
+                Ok(c) => c,
+                Err(e) => {
+                    log::warn!("could not execute leave cmd: {e}");
+                    return;
+                }
+            };
+            match child.wait().await {
+                Ok(s) => {
+                    if s.success() {
+                        log::info!("{cmd} exited successfully");
+                    } else {
+                        log::warn!("{cmd} exited with {s}");
+                    }
+                }
+                Err(e) => log::warn!("{cmd}: {e}"),
+            }
+        });
+    }


Done in 3313b2e — collapsed both into a single spawn_hook_command(handle, HookKind) with the kind carrying the human label into log messages (which now also include the client handle for both enter and leave).

The Swatinem/rust-cache action was restored before brew installed the native dependencies. When Homebrew updated glib (2.88.0 → newer) the cached Rust build artifacts still referenced the old versioned Cellar path (/opt/homebrew/Cellar/glib/2.88.0/lib), causing a linker failure: ld: library 'gio-2.0' not found Fix by: 1. Moving `brew install` before the rust-cache step so libs are current before the cache is consulted. 2. Capturing `brew list --versions glib gtk4 libadwaita` into a prefix-key so the cache key changes whenever those packages update, forcing a clean Rust build with fresh pkg-config paths. Linux and Windows cache behaviour is unchanged (MACOS_LIB_VER unset → empty prefix-key → same default v0- prefix). Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

tyvsmith · 2026-05-27T01:14:07Z

Reference implementation for hook driven input changes

Add to ~/.config/lan-mouse/config.toml

enter_hook = "lan-mouse-mode mac"
leave_hook = "lan-mouse-mode default"

Script that swaps super/alt & inverts scrolling for Hyprland host and MacOS client

The two spawn_*_hook_command functions were ~35 lines of near-byte- identical glue (only the get_*_cmd call and log prefixes differed). Replace with a single spawn_hook_command(handle, HookKind) and let the kind carry the human label into log lines (which now also include the client handle). No behaviour change. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

tyvsmith and others added 2 commits May 26, 2026 15:09

tyvsmith marked this pull request as ready for review May 26, 2026 23:38

Copilot AI review requested due to automatic review settings May 26, 2026 23:38

Copilot AI reviewed May 26, 2026

View reviewed changes

tyvsmith mentioned this pull request May 27, 2026

fix(ci): bust stale Homebrew glib cache on macOS runners #449

Open

3 tasks

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

feat: add leave_hook symmetric to enter_hook#448

feat: add leave_hook symmetric to enter_hook#448
tyvsmith wants to merge 4 commits into
feschber:mainfrom
tyvsmith:feat/leave-hook

tyvsmith commented May 26, 2026 •

edited

Loading

Uh oh!

Copilot AI left a comment

Uh oh!

tyvsmith May 27, 2026

Uh oh!

tyvsmith May 27, 2026

Uh oh!

tyvsmith May 27, 2026

Uh oh!

tyvsmith commented May 27, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Uh oh!

Conversation

tyvsmith commented May 26, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Design

Docs

Out of scope

Test plan

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

tyvsmith May 27, 2026

Choose a reason for hiding this comment

Uh oh!

tyvsmith May 27, 2026

Choose a reason for hiding this comment

Uh oh!

tyvsmith May 27, 2026

Choose a reason for hiding this comment

Uh oh!

tyvsmith commented May 27, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

tyvsmith commented May 26, 2026 •

edited

Loading