fix: quit in ::unlocked instead of after unlock() (v0.6.17)

The real cause of the unlock SIGSEGV: gtk4-session-lock destroys the lock
windows itself when the lock ends (per its header), so unlock_callback's
`unlock(); app.quit();` destroyed them a second time — surface already gone
→ gdk_surface_get_display NULL → crash in gtk_window_destroy. Reproduces on
a plain single-monitor lock/unlock, no suspend needed.

unlock_callback now calls only unlock(); a new connect_unlocked handler
quits the app after the library finishes teardown. Mirrors the upstream
examples/simple.rs exactly.

Reverts the v0.6.15/v0.6.16 monitor-pruning + LockscreenHandles.monitor:
it was a misdiagnosis (the crash is monitor-independent) and manually
manipulated library-managed lock windows, which the upstream example
explicitly warns against. Monitor removal is left to the library.

CLAUDE.md

@@ -42,7 +42,7 @@ LD_PRELOAD=/usr/lib/libgtk4-layer-shell.so ./target/release/moonlock
 - `i18n.rs` — Locale-Erkennung (OnceLock-cached) und String-Tabellen (DE/EN), faillock_warning mit konfigurierbarem max_attempts
 - `config.rs` — TOML-Config (background_path, background_blur clamped [0,200], fingerprint_enabled als Option<bool>) + Wallpaper-Fallback + Symlink-Rejection via symlink_metadata + Parse-Error-Logging
 - `lockscreen.rs` — GTK4 UI via LockscreenHandles, PAM-Auth via gio::spawn_blocking mit 30s Timeout und Generation Counter, FP-Success ruft unlock_callback direkt (PAM-Stack ohne account-Modul, Lockout via auth-Pfad und MAX_FP_ATTEMPTS), Zeroizing<String> für Passwort, Power-Confirm, GPU-Blur via GskBlurNode (Downscale auf max 1920px), Blur/Avatar-Cache für Multi-Monitor
-- `main.rs` — Entry Point, Panic-Hook (vor Logging), Root-Check, ext-session-lock-v1 (Pflicht in Release), Monitor-Hotplug via `connect_monitor`-Signal (v1_2), shared Blur/Avatar-Caches in Rc, systemd-Journal-Logging, Debug-Level per `MOONLOCK_DEBUG` Env-Var, async fprintd-Init nach window.present()
+- `main.rs` — Entry Point, Panic-Hook (vor Logging), Root-Check, ext-session-lock-v1 (Pflicht in Release), Monitor-Hotplug via `connect_monitor`-Signal (v1_2), Unlock via `unlock()` + `connect_unlocked`→`app.quit()` (Lib zerstört die Lock-Fenster selbst beim Lock-Ende — kein eigenes Destroy/quit, sonst Double-Destroy-SIGSEGV), shared Blur/Avatar-Caches in Rc, systemd-Journal-Logging, Debug-Level per `MOONLOCK_DEBUG` Env-Var, async fprintd-Init nach window.present()
 
 ## Sicherheit
 

Cargo.lock

@@ -575,7 +575,7 @@ dependencies = [
 
 [[package]]
 name = "moonlock"
-version = "0.6.14"
+version = "0.6.17"
 dependencies = [
  "gdk-pixbuf",
  "gdk4",

Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "moonlock"
-version = "0.6.14"
+version = "0.6.17"
 edition = "2024"
 description = "A secure Wayland lockscreen with GTK4, PAM and fingerprint support"
 license = "MIT"

DECISIONS.md

@@ -2,6 +2,20 @@
 
 Architectural and design decisions for Moonlock, in reverse chronological order.
 
+## 2026-06-02 – Real fix for the unlock SIGSEGV: quit in ::unlocked, never destroy windows ourselves (v0.6.17)
+
+- **Who**: ClaudeCode, Dom
+- **Why**: The v0.6.15 "stale per-monitor window" theory was a **misdiagnosis**. A clean manual `target/release/moonlock` lock→unlock (single monitor, no suspend, no monitor removal) still crashed, preceded by `Gdk-CRITICAL: gdk_surface_get_display: assertion 'GDK_IS_SURFACE (surface)' failed`. The v0.6.16 backtrace's top app frame is the `unlock_callback` closure, which ran `lock.unlock(); app.quit();`. Per the gtk4-session-lock header (`assign_window_to_monitor`: "the window will be unmapped and `gtk_window_destroy()` called on it when the current lock ends") the **library destroys the lock windows itself** on unlock. `app.quit()` then destroyed the same windows again — surface already gone → SIGSEGV. A double-destroy in the teardown sequence, entirely independent of monitors.
+- **Tradeoffs**: Reverted the v0.6.15/v0.6.16 monitor-pruning + `LockscreenHandles.monitor` field: it targeted the wrong cause, never fired in testing, and manually called `app.remove_window()` on library-managed lock windows — exactly what the upstream example warns against ("does NOT manually destroy or close the lock windows"). Monitor-removal-during-lock is left entirely to the library (which already auto-unmaps+dereferences such windows). Three attempts total: idle_add deferral (wrong: reentrancy theory), monitor-pruning (wrong: misdiagnosis), and this one — verified against both the C header and the upstream `examples/simple.rs`.
+- **How**: `unlock_callback` now calls only `lock.unlock()`. A new `lock.connect_unlocked(|_| app.quit())` quits the app only after the library finishes its teardown and fires `::unlocked`. Mirrors the canonical gtk4-session-lock usage exactly.
+
+## 2026-06-02 – Prune per-monitor windows on monitor removal to fix resume-unlock SIGSEGV (v0.6.15)
+
+- **Who**: ClaudeCode, Dom
+- **Why**: Chronic SIGSEGV (multiple coredumps/day) on the unlock following a suspend/resume. Backtrace: `app.quit()` → `gtk_window_destroy` → gtk4-layer-shell → `g_signal_emit` → NULL deref (`0x278`, rax=0). Root cause: `connect_monitor` is add-only — it creates one window per monitor and pushes to `all_handles`, but nothing ever removes a window when a monitor powers off on suspend. gtk4-session-lock unmaps + drops *its* ref to that window on monitor removal (per `gtk4-session-lock` 0.4 docs), but the GtkApplication (`ApplicationWindow::builder().application(app)`) and `all_handles` still hold refs. The orphaned window survives until unlock, where destroying it dereferences its now-NULL monitor association. A diagnostic confirmed the windows are GTK-valid but accumulate (3 windows for 1-2 monitors) with a NULL associated object.
+- **Tradeoffs**: An earlier attempt deferred `unlock()`/`quit()` via `glib::idle_add_local_once` on a reentrancy theory — proven wrong by the logs (crash happens *inside* the idle trampoline). Reverted. The library doc explicitly says monitor removal is detected via "GTK APIs"; we follow that rather than fighting the library. We release our refs (do **not** call `destroy` — the lib already unmapped+dereffed the window). Diagnostic UNLOCK logging is kept for now, to be removed once a suspend/resume validation cycle confirms the fix.
+- **How**: `LockscreenHandles` gains `monitor: Option<gdk::Monitor>`, set in the `connect_monitor` handler. `activate_with_session_lock` watches `display.monitors()` via `connect_items_changed`; on any change it retains only handles whose monitor `is_valid()`, calling `app.remove_window()` on the pruned ones to drop the application's ref. With both refs released, the orphaned window is gone before unlock.
+
 ## 2026-06-02 – Align power-confirm to moonset's ActionDef pattern (v0.6.14)
 
 - **Who**: ClaudeCode, Dom

src/main.rs

@@ -75,9 +75,12 @@ fn activate_with_session_lock(
     // Shared unlock callback — unlocks session and quits.
     // Guard prevents double-unlock if PAM and fingerprint succeed simultaneously.
     let lock_clone = lock.clone();
-    let app_clone = app.clone();
     let already_unlocked = Rc::new(Cell::new(false));
     let au = already_unlocked.clone();
+    // unlock() only. The library destroys the lock windows itself when the lock ends,
+    // and the ::unlocked handler quits the app afterwards. Calling app.quit() here too
+    // double-destroyed the windows (their surface was already gone) and segfaulted
+    // gtk_window_destroy. Matches the upstream gtk4-session-lock example.
     let unlock_callback: Rc<dyn Fn()> = Rc::new(move || {
         if au.get() {
             log::debug!("Unlock already triggered, ignoring duplicate");
@@ -85,7 +88,6 @@ fn activate_with_session_lock(
         }
         au.set(true);
         lock_clone.unlock();
-        app_clone.quit();
     });
 
     // Shared caches for multi-monitor — first monitor renders, rest reuse
@@ -147,6 +149,16 @@ fn activate_with_session_lock(
         }
     ));
 
+    // Quit only after the library finishes unlocking (::unlocked fires after the lock
+    // ends). gtk4-session-lock destroys the lock windows itself at lock-end; quitting
+    // earlier — or destroying windows ourselves — races that teardown and segfaults
+    // gtk_window_destroy on an already-gone surface. Mirrors the upstream example.
+    lock.connect_unlocked(glib::clone!(
+        #[weak]
+        app,
+        move |_| app.quit()
+    ));
+
     lock.lock();
 
     // Async fprintd initialization — runs after windows are visible