Shell 33: app-shell-runtime (#825) #911

Closed
charles wants to merge 0 commits from shell-33-app-shell-runtime into shell
Owner

Stacked on: shell-17-platform-single-instance (the other listed branches, shell-7, 8, 10, 13, 16, 18, 19, 21, 22, 23, 24 and 32, are already on shell). Merging shell-17 conflicted only in crates/fluxer-platform/Cargo.toml: the two windows feature lists were joined.

What this builds (Desktop shell plan, Task 33)

  • main, single instance first (spec §7.2, #754). This runs right after Velopack's hook and the argument parse, and before logging, settings and the runtime. It is skipped for --debug-info and --check-noise-suppression. The scope is the data directory; with --devserver it is the devserver's scratch directory. shell_rt::start_instance does the work:
    • A second launch sends Activate (with XDG_ACTIVATION_TOKEN), OpenUrl or Task, then exits 0.
    • If the answer is busy or nothing comes back within 2 s, it waits up to 10 s for the lock and takes over.
    • --relaunched-from <pid> waits for that process first. If the process still runs after that, the launch sends it UpdateRelaunch, which makes it quit.
    • When the channel cannot be made, the client runs without it, and a warning is logged.
    • The legacy folder migration now runs before the lock. Without XDG_RUNTIME_DIR the lock lives in the data directory, so taking it first would create that directory and break the migration.
  • New options (config.rs): --autostart, --url <URL>, a bare fluxer:// or flaxative:// link (as %U and the Windows handler pass it; any other bare argument is a usage error), --task open-settings, --reset-window-state, --no-tray and the hidden --relaunched-from <pid>. Two helpers go with them: Config::launch() builds the core Launch, and Config::instance_message() builds the message a second launch sends.
  • shell_rt::ShellRuntime. The traits are injected through ShellPlatform: system() in main, and inert() by default, so tests never touch the desktop.
    • Tray: made at start. With --no-tray it is never made and Disabled is reported; when the system has no tray, NoHost is reported. Each ShellCommand::Tray / RemoveTray goes to the tray. Its events come back: a click → Tray(ToggleWindow), a double click → ShowWindow, a menu row → Tray(action), availability → ShellEvent::TrayAvailability. The icon is removed when the shutdown begins.
    • Autostart: the state is read at start and again after each SetAutostart (enable or disable).
    • Badge (SetBadge): set through the badge of the window the UI's WindowHost reports (shell_rt::window_host / Runtime::with_window_host; main sets Services::window_host). It is rebuilt when a window opens, re-applied on Show / Raise / Activate, and follows "Use dark taskbar badge".
    • Displays: read once at start and dispatched with the startup load. main waits (at most 5 s) for the first snapshot with settings before the UI starts. It also sets Services::launch, so start minimised and the restored placement see the right values.
    • sound_allowed(): asked every second on a blocking thread; a change is reported as ShellEvent::SoundAllowed (see deviations).
    • Desktop entry and URL schemes: installed at start (Linux: entry plus xdg-mime claim; Windows: HKCU classes), and on InstallDesktopEntry.
    • InstanceServer: answers each message ok at once and hands it to the store as ShellEvent::Instance. Once the shutdown began it answers busy.
    • Order: blocking work runs on one thread of its own, in order.
  • Quit and restart (runtime.rs): ShellCommand::Quit / Restart tell the window WindowCommand::Quit and end the loop; Handle::stopped_on_request() reports this and main's supervisor exits 0 on it. The existing shutdown follows, so a ready update still installs on quit. Then, as #754 asks, the instance listener is closed and the update lock released (UpdateRuntime::release_lock). Only after that does the relaunch run: through ShellPlatform::relaunch (appimage::relaunch of Launch::executable), or through the installer's relaunch when an update is being installed.
  • Relaunch arguments (relaunch_args): this process's arguments without --autostart, --reset-window-state, --url, --task, a bare link or an earlier --relaunched-from, plus --relaunched-from <pid>. The update's relaunch uses the same rule; before this change it passed every argument as given.
  • A link or task given to the first launch is dispatched after the startup load. A link waits for the login; the core keeps it until then.
  • Tray actions: "Copy build info", "Update now" and "Check for updates" run through the core's existing Effect::Clipboard and UpdateCommand paths (Task 8); nothing new was needed here.

Tests

apps/fluxer/tests/shell_runtime.rs runs against the devserver shell scenario. Every OS piece is a fake (FakeTray, FakeAutostart, FakeBadge, a recording relauncher), and the single-instance channel lives in a temporary directory. Tests:

  • second_start_activates_and_exits
  • link_argument_reaches_running_client (a bare link, and --task open-settings)
  • link_given_to_the_first_launch_runs_after_login
  • busy_during_shutdown_then_takeover
  • relaunch_releases_locks_first
  • relaunch_arguments_leave_one_shot_options_out
  • tray_follows_voice_and_unread
  • tray_actions_run
  • close_to_tray_keeps_voice
  • no_tray_close_quits (no host, and --no-tray)
  • autostart_reconciled_at_start
  • badge_set_and_cleared
  • start_minimised_only_with_autostart
  • reset_window_state_option

There are also unit tests for the config options, the tray event mapping and the relaunch arguments. FLUXER_SCREENSHOT_TESTS=1 cargo xtask ci passes locally. No FLUXER_DESKTOP_TESTS tests were added or run: no real tray, dialog or desktop entry was touched.

Deviations from the plan

  • sound_allowed() is polled every second and reported when it changes, not asked at the moment a sound is about to play. The gate lives in the pure core, which cannot wait for a D-Bus answer, so the core always has an answer at most 1 s old. This matches the original's 1 s cache.
  • busy_during_shutdown_then_takeover drives the InstanceServer the runtime uses (begin_shutdown, then close) directly. The runtime's own shutdown with the devserver is too quick to send a message in the middle of it reliably. relaunch_releases_locks_first covers the runtime path: the lock is free when the relauncher is called.
  • Update lock: under --devserver the updater takes no update lock, so the relaunch test checks the instance lock only. release_lock is a one-line call before the install and the relaunch.
  • --reset-window-state ignores the stored placement (core start_placement). The stored values are not deleted: the window's first resize or move overwrites them.
  • Tray activation token: the token from the tray's Activate is dropped for now. Raising the window with it on Wayland is Desktop shell 30's job.
  • desktop_rt.rs: listed in the plan's Files, but it needed no change.

For Charles to check on Windows (cannot run here)

  • A second launch focuses the running client.
  • flaxative:// links from the browser reach the running client through --url "%1".
  • The tray's "Restart Flaxative" starts a new client once the old one has gone.
  • The overlay badge appears on the task bar button after the window is hidden and shown again.

Closes #825

🤖 Generated with Claude Code

https://claude.ai/code/session_0161qMQMsCyJhjYw4gqa6e4K

Stacked on: shell-17-platform-single-instance (the other listed branches, shell-7, 8, 10, 13, 16, 18, 19, 21, 22, 23, 24 and 32, are already on `shell`). Merging shell-17 conflicted only in `crates/fluxer-platform/Cargo.toml`: the two `windows` feature lists were joined. ## What this builds (Desktop shell plan, Task 33) - **`main`, single instance first** (spec §7.2, #754). This runs right after Velopack's hook and the argument parse, and before logging, settings and the runtime. It is skipped for `--debug-info` and `--check-noise-suppression`. The scope is the data directory; with `--devserver` it is the devserver's scratch directory. `shell_rt::start_instance` does the work: - A second launch sends `Activate` (with `XDG_ACTIVATION_TOKEN`), `OpenUrl` or `Task`, then exits 0. - If the answer is `busy` or nothing comes back within 2 s, it waits up to 10 s for the lock and takes over. - `--relaunched-from <pid>` waits for that process first. If the process still runs after that, the launch sends it `UpdateRelaunch`, which makes it quit. - When the channel cannot be made, the client runs without it, and a warning is logged. - The legacy folder migration now runs before the lock. Without `XDG_RUNTIME_DIR` the lock lives in the data directory, so taking it first would create that directory and break the migration. - **New options** (`config.rs`): `--autostart`, `--url <URL>`, a bare `fluxer://` or `flaxative://` link (as `%U` and the Windows handler pass it; any other bare argument is a usage error), `--task open-settings`, `--reset-window-state`, `--no-tray` and the hidden `--relaunched-from <pid>`. Two helpers go with them: `Config::launch()` builds the core `Launch`, and `Config::instance_message()` builds the message a second launch sends. - **`shell_rt::ShellRuntime`**. The traits are injected through `ShellPlatform`: `system()` in `main`, and `inert()` by default, so tests never touch the desktop. - **Tray:** made at start. With `--no-tray` it is never made and `Disabled` is reported; when the system has no tray, `NoHost` is reported. Each `ShellCommand::Tray` / `RemoveTray` goes to the tray. Its events come back: a click → `Tray(ToggleWindow)`, a double click → `ShowWindow`, a menu row → `Tray(action)`, availability → `ShellEvent::TrayAvailability`. The icon is removed when the shutdown begins. - **Autostart:** the state is read at start and again after each `SetAutostart` (enable or disable). - **Badge (`SetBadge`):** set through the badge of the window the UI's `WindowHost` reports (`shell_rt::window_host` / `Runtime::with_window_host`; `main` sets `Services::window_host`). It is rebuilt when a window opens, re-applied on `Show` / `Raise` / `Activate`, and follows "Use dark taskbar badge". - **Displays:** read once at start and dispatched with the startup load. `main` waits (at most 5 s) for the first snapshot with settings before the UI starts. It also sets `Services::launch`, so start minimised and the restored placement see the right values. - **`sound_allowed()`:** asked every second on a blocking thread; a change is reported as `ShellEvent::SoundAllowed` (see deviations). - **Desktop entry and URL schemes:** installed at start (Linux: entry plus `xdg-mime` claim; Windows: `HKCU` classes), and on `InstallDesktopEntry`. - **`InstanceServer`:** answers each message `ok` at once and hands it to the store as `ShellEvent::Instance`. Once the shutdown began it answers `busy`. - **Order:** blocking work runs on one thread of its own, in order. - **Quit and restart** (`runtime.rs`): `ShellCommand::Quit` / `Restart` tell the window `WindowCommand::Quit` and end the loop; `Handle::stopped_on_request()` reports this and `main`'s supervisor exits 0 on it. The existing shutdown follows, so a ready update still installs on quit. Then, as #754 asks, the instance listener is closed and the update lock released (`UpdateRuntime::release_lock`). Only after that does the relaunch run: through `ShellPlatform::relaunch` (`appimage::relaunch` of `Launch::executable`), or through the installer's relaunch when an update is being installed. - **Relaunch arguments** (`relaunch_args`): this process's arguments without `--autostart`, `--reset-window-state`, `--url`, `--task`, a bare link or an earlier `--relaunched-from`, plus `--relaunched-from <pid>`. The update's relaunch uses the same rule; before this change it passed every argument as given. - A link or task given to the first launch is dispatched after the startup load. A link waits for the login; the core keeps it until then. - **Tray actions:** "Copy build info", "Update now" and "Check for updates" run through the core's existing `Effect::Clipboard` and `UpdateCommand` paths (Task 8); nothing new was needed here. ## Tests `apps/fluxer/tests/shell_runtime.rs` runs against the devserver `shell` scenario. Every OS piece is a fake (`FakeTray`, `FakeAutostart`, `FakeBadge`, a recording relauncher), and the single-instance channel lives in a temporary directory. Tests: - `second_start_activates_and_exits` - `link_argument_reaches_running_client` (a bare link, and `--task open-settings`) - `link_given_to_the_first_launch_runs_after_login` - `busy_during_shutdown_then_takeover` - `relaunch_releases_locks_first` - `relaunch_arguments_leave_one_shot_options_out` - `tray_follows_voice_and_unread` - `tray_actions_run` - `close_to_tray_keeps_voice` - `no_tray_close_quits` (no host, and `--no-tray`) - `autostart_reconciled_at_start` - `badge_set_and_cleared` - `start_minimised_only_with_autostart` - `reset_window_state_option` There are also unit tests for the config options, the tray event mapping and the relaunch arguments. `FLUXER_SCREENSHOT_TESTS=1 cargo xtask ci` passes locally. No `FLUXER_DESKTOP_TESTS` tests were added or run: no real tray, dialog or desktop entry was touched. ## Deviations from the plan - **`sound_allowed()`** is polled every second and reported when it changes, not asked at the moment a sound is about to play. The gate lives in the pure core, which cannot wait for a D-Bus answer, so the core always has an answer at most 1 s old. This matches the original's 1 s cache. - **`busy_during_shutdown_then_takeover`** drives the `InstanceServer` the runtime uses (`begin_shutdown`, then `close`) directly. The runtime's own shutdown with the devserver is too quick to send a message in the middle of it reliably. `relaunch_releases_locks_first` covers the runtime path: the lock is free when the relauncher is called. - **Update lock:** under `--devserver` the updater takes no update lock, so the relaunch test checks the instance lock only. `release_lock` is a one-line call before the install and the relaunch. - **`--reset-window-state`** ignores the stored placement (core `start_placement`). The stored values are not deleted: the window's first resize or move overwrites them. - **Tray activation token:** the token from the tray's `Activate` is dropped for now. Raising the window with it on Wayland is Desktop shell 30's job. - **`desktop_rt.rs`:** listed in the plan's Files, but it needed no change. ## For Charles to check on Windows (cannot run here) - A second launch focuses the running client. - `flaxative://` links from the browser reach the running client through `--url "%1"`. - The tray's "Restart Flaxative" starts a new client once the old one has gone. - The overlay badge appears on the task bar button after the window is hidden and shown again. Closes #825 🤖 Generated with [Claude Code](https://claude.com/claude-code) https://claude.ai/code/session_0161qMQMsCyJhjYw4gqa6e4K
One client per data directory (Desktop shell spec §7.2, §12): a flock and
a 0700/0600 Unix socket with the peer uid checked, or a Local\ mutex and a
first-instance, user-only, local-only named pipe on Windows. Frames of at
most 4 KiB, read within 2 s, answered ok / busy / error. wait_and_acquire
and wait_for_pid_exit for the updater's handover (#754); UpdateLock gains
release() for the relaunch.

Refs #809

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0161qMQMsCyJhjYw4gqa6e4K
- tests: `Path` imported only by the Unix test that uses it (Windows clippy
  -D warnings: unused import).
- A full message queue is answered `error` (Refused) instead of `busy`, which
  stays for a stopped listener (shutting down).
- `acquire_in` and `send_frame` called inside an async runtime run their own
  runtime on a thread of their own instead of panicking.
- `wait_for_pid_exit`: ACCESS_DENIED from OpenProcess (an elevated or another
  user's process) counts as running on Windows; without /proc, Unix asks `ps`
  instead of `kill -0` (EPERM).

Refs #809

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0161qMQMsCyJhjYw4gqa6e4K
The lock test expected the flock free right after drop, but other tests in the
same binary fork child processes that share the open file until they exec. It
now retries for a bounded time after drop and after release.

Docs: ANSWER_LIMIT vs the second launch's 2 s limit, and Refused means exit
without waiting for the lock.

Refs #809

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0161qMQMsCyJhjYw4gqa6e4K
# Conflicts:
#	crates/fluxer-platform/Cargo.toml
Refs #825

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0161qMQMsCyJhjYw4gqa6e4K
The single instance first in main (a second launch hands over its message,
busy/no answer waits and takes over, --relaunched-from waits for the old
process), the new options (--autostart, --url and a bare link, --task,
--reset-window-state, --no-tray, --relaunched-from), and ShellRuntime: the
tray, autostart, badge through the WindowHost, displays, sound state, desktop
entry, quit and restart with the instance and update locks released before
any relaunch (#754).

Refs #825

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0161qMQMsCyJhjYw4gqa6e4K
Author
Owner

Independent review: Shell 33 (app shell runtime)

Reviewed the diff 11933dd1..origin/shell-33-app-shell-runtime (78268efe), read cold.

Gate: FLUXER_SCREENSHOT_TESTS=1 cargo xtask ci run in a clean review worktree: all steps passed (exit 0). All 12 planned tests in apps/fluxer/tests/shell_runtime.rs ran and passed, plus link_given_to_the_first_launch_runs_after_login and relaunch_arguments_leave_one_shot_options_out.

What I checked

  • Planned tests (1): all 12 names from the plan are there and assert what the plan says: Activate carries the launcher token and becomes WindowCommand::Activate, or Raise without one; a bare link reaches the running client's invite modal and --task open-settings opens settings; a launch during shutdown gets busy and takes over within 10 s; the instance lock is free when the relaunch starts and the arguments drop --autostart; close-to-tray keeps voice; with no tray, or with --no-tray, close quits and never hides; autostart is reconciled to off (§18 Q5); the badge counts and clears; start minimised only with --autostart; --reset-window-state.
  • Interfaces (2): --autostart, --url plus a bare link, --task, --reset-window-state, --no-tray and the hidden --relaunched-from are all present. The single instance is taken before logging and settings, and is skipped for --debug-info, --help and --check-noise-suppression. The second launch exits 0. Busy or no answer leads to wait_and_acquire (10 s). ShellRuntime runs the tray, SetAutostart, SetBadge through the WindowHost, the desktop entry at start, Restart and Quit. "Copy build info", "Update now" and "Check for updates" go through the existing paths (core tray.rs). Both locks are released before the install or relaunch.
  • Invariants (3): fluxer-core and fluxer-ui are untouched. InstanceMessage and ActivationToken hide links and tokens in their Debug output, so the received = ?message log line leaks nothing. No secrets, paths or links are logged. The Windows cfg block (url_scheme::register(&SystemRegistry, …)) uses paths that exist in fluxer-platform (registry::SystemRegistry is cfg(windows) and re-exported; url_scheme is windows.rs, compiled on every OS), and the types line up.
  • Parity (4), checked against the fork: POLICY_CACHE_TTL_MS = 1_000 (fluxer_desktop/src/main/NotificationState.ts:18) matches SOUND_POLL. AUTOSTART_LAUNCH_ARG = '--autostart' (AutostartLaunch.ts:6) matches. The original ignores a second instance while quitting (ModuleBootHandoff.ts:94-98); this branch answers busy, which is what spec §7.2 / #754 deliberately changed. The original relaunches with the full process.argv.slice(1) (LinuxLaunchPath.ts:46); the spec drops --autostart, and this branch also drops the one-shot link, task and reset options, which is sensible.
  • Review Focus 5: close follows the store per tray availability; --no-tray reports Disabled and no tray is made, so the window is never left unreachable.

Findings

  1. minor: shell_rt::poll_sound asks sound_allowed() every second for as long as the process runs. On Linux each call opens a new session-bus connection and makes up to two round trips (notify/state_linux.rs:38-60). That is about 86 400 bus connections a day while idle, and a debug line every second when there is no bus. Spec §9 / plan say to ask when a sound is about to play, cached 1 s; the deviation is documented. Suggested follow-up: keep one connection, or poll only while sounds are enabled or the window is unfocused.
  2. minor: relaunch_releases_locks_first checks only the instance lock. The plan says "releases the instance and update locks before any relaunch", and the update lock release (update.release_lock() in runtime.rs) has no assertion.
  3. minor: --reset-window-state ignores the stored placement but does not reset it (spec §7.1: "ignores and resets it"). The deviation is documented, and the stored state is overwritten by the first move, resize or close save.
  4. minor: the update lock and the instance lock are released before installer.apply, not only right before the relaunch, so a launch during the install can become primary and start its own update check. The window is short, and the plan names this order, so this is noted only.
  5. minor: the first Persisted dispatch awaits the displays probe (spawn_blocking, no timeout). An X server that hangs on connect would delay the load past main's 5 s settings wait. Consider a short timeout.
  6. minor / info: the tray's Activate { token } is dropped (left to Task 30), and busy-during-shutdown is tested on InstanceServer rather than through the runtime. By reading the code, begin_shutdown() runs right after the loop ends and before the quit-time update wait, which is what #754 needs.

Windows, by hand (for Task 44): the URL scheme registration under HKCU classes, the badge overlay through the WindowHost, and Restart through the Velopack stub (launch.executable) with --relaunched-from.

Verdict: approved. No blocking findings; local gate green.

## Independent review: Shell 33 (app shell runtime) Reviewed the diff `11933dd1..origin/shell-33-app-shell-runtime` (78268efe), read cold. **Gate:** `FLUXER_SCREENSHOT_TESTS=1 cargo xtask ci` run in a clean review worktree: **all steps passed** (exit 0). All 12 planned tests in `apps/fluxer/tests/shell_runtime.rs` ran and passed, plus `link_given_to_the_first_launch_runs_after_login` and `relaunch_arguments_leave_one_shot_options_out`. ### What I checked - **Planned tests (1):** all 12 names from the plan are there and assert what the plan says: Activate carries the launcher token and becomes `WindowCommand::Activate`, or `Raise` without one; a bare link reaches the running client's invite modal and `--task open-settings` opens settings; a launch during shutdown gets `busy` and takes over within 10 s; the instance lock is free when the relaunch starts and the arguments drop `--autostart`; close-to-tray keeps voice; with no tray, or with `--no-tray`, close quits and never hides; autostart is reconciled to off (§18 Q5); the badge counts and clears; start minimised only with `--autostart`; `--reset-window-state`. - **Interfaces (2):** `--autostart`, `--url` plus a bare link, `--task`, `--reset-window-state`, `--no-tray` and the hidden `--relaunched-from` are all present. The single instance is taken before logging and settings, and is skipped for `--debug-info`, `--help` and `--check-noise-suppression`. The second launch exits 0. `Busy` or no answer leads to `wait_and_acquire` (10 s). `ShellRuntime` runs the tray, `SetAutostart`, `SetBadge` through the `WindowHost`, the desktop entry at start, `Restart` and `Quit`. "Copy build info", "Update now" and "Check for updates" go through the existing paths (core `tray.rs`). Both locks are released before the install or relaunch. - **Invariants (3):** fluxer-core and fluxer-ui are untouched. `InstanceMessage` and `ActivationToken` hide links and tokens in their `Debug` output, so the `received = ?message` log line leaks nothing. No secrets, paths or links are logged. The Windows `cfg` block (`url_scheme::register(&SystemRegistry, …)`) uses paths that exist in `fluxer-platform` (`registry::SystemRegistry` is cfg(windows) and re-exported; `url_scheme` is `windows.rs`, compiled on every OS), and the types line up. - **Parity (4), checked against the fork:** `POLICY_CACHE_TTL_MS = 1_000` (`fluxer_desktop/src/main/NotificationState.ts:18`) matches `SOUND_POLL`. `AUTOSTART_LAUNCH_ARG = '--autostart'` (`AutostartLaunch.ts:6`) matches. The original *ignores* a second instance while quitting (`ModuleBootHandoff.ts:94-98`); this branch answers `busy`, which is what spec §7.2 / #754 deliberately changed. The original relaunches with the full `process.argv.slice(1)` (`LinuxLaunchPath.ts:46`); the spec drops `--autostart`, and this branch also drops the one-shot link, task and reset options, which is sensible. - **Review Focus 5:** close follows the store per tray availability; `--no-tray` reports `Disabled` and no tray is made, so the window is never left unreachable. ### Findings 1. **minor**: `shell_rt::poll_sound` asks `sound_allowed()` every second for as long as the process runs. On Linux each call opens a **new** session-bus connection and makes up to two round trips (`notify/state_linux.rs:38-60`). That is about 86 400 bus connections a day while idle, and a debug line every second when there is no bus. Spec §9 / plan say to ask when a sound is about to play, cached 1 s; the deviation is documented. Suggested follow-up: keep one connection, or poll only while sounds are enabled or the window is unfocused. 2. **minor**: `relaunch_releases_locks_first` checks only the instance lock. The plan says "releases the instance **and update** locks before any relaunch", and the update lock release (`update.release_lock()` in `runtime.rs`) has no assertion. 3. **minor**: `--reset-window-state` ignores the stored placement but does not reset it (spec §7.1: "ignores and resets it"). The deviation is documented, and the stored state is overwritten by the first move, resize or close save. 4. **minor**: the update lock and the instance lock are released before `installer.apply`, not only right before the relaunch, so a launch during the install can become primary and start its own update check. The window is short, and the plan names this order, so this is noted only. 5. **minor**: the first `Persisted` dispatch awaits the displays probe (`spawn_blocking`, no timeout). An X server that hangs on connect would delay the load past `main`'s 5 s settings wait. Consider a short timeout. 6. **minor / info**: the tray's `Activate { token }` is dropped (left to Task 30), and busy-during-shutdown is tested on `InstanceServer` rather than through the runtime. By reading the code, `begin_shutdown()` runs right after the loop ends and before the quit-time update wait, which is what #754 needs. **Windows, by hand (for Task 44):** the URL scheme registration under HKCU classes, the badge overlay through the `WindowHost`, and Restart through the Velopack stub (`launch.executable`) with `--relaunched-from`. **Verdict: approved.** No blocking findings; local gate green.
Author
Owner

Merged into shell by hand (the merge API returns 405); the commits are in the branch.

Merged into `shell` by hand (the merge API returns 405); the commits are in the branch.
charles closed this pull request 2026-11-18 20:41:07 +00:00

Pull request closed

Sign in to join this conversation.
No description provided.