Fix several bugs: regex flags, implicit globals, crashes, ngettext, undefined icon

[KAMI911]

Summary

This PR contains 8 small, independent bug fixes found during a code review. Each fix is a separate commit.

---

1. util: fixupPCIDescription: fix missing global flag in regex replacements

Two regex replacements in fixupPCIDescription() were missing the /g flag, causing only the first occurrence to be replaced:

• /[_,]/ → /[_,]/g — replace all underscores/commas with spaces
• /\([\s\S][^\(\)]{2,}\)/ → same with /g — strip all parenthesized segments, not just the first

---

2. keyboardManager: fix implicit global variable declarations

Five constants at the top of the file (DESKTOP_INPUT_SOURCES_SCHEMA, KEY_INPUT_SOURCES, KEY_KEYBOARD_OPTIONS, KEY_PER_WINDOW, KEY_SOURCE_LAYOUTS) were assigned without var, making them implicit globals that pollute the global scope. Added var to match the style used elsewhere in the file.

---

3. notificationDaemon: add default case for unknown urgency in icon switch

The switch (hints.urgency) in _iconForNotificationData() had no default branch, leaving stockIcon as undefined for any urgency value outside LOW/NORMAL/CRITICAL. This silently created a broken St.Icon with icon_name: undefined. Added a default case falling back to the information icon.

---

4. notificationDaemon: fix crash in _expireNotification when notification is null

NotifyAsync() adds ndata to _expireNotifications and starts the expire timer before the async GetConnectionUnixProcessID D-Bus call completes. ndata.notification is only assigned later inside _notifyForSource(). If the timer fires during that window, calling ndata.notification.destroy() on undefined throws a TypeError. Added a null guard; if the notification object doesn't exist yet, the orphaned entry is removed and the timer is restarted for the next item.

---

5. windowManager: fix ngettext missing count argument in display-change dialog

ngettext(singular, plural) requires a third argument — the count used to select the correct plural form. Without it the function always returns the singular string (e.g. "Reverting in 5 second." instead of "Reverting in 5 seconds."). Fixed to match the pattern used in endSessionDialog.js.

---

6. cinnamonDBus: fix crash in activateCallback when xlet is not found

_getXletObject() returns null when no applet/desklet/extension matches the given uuid/instance_id. The next line called null[callback].bind(null), throwing a TypeError. This happens when cinnamon-settings calls a callback for an xlet that was removed without a shell restart. Added a null guard with a warning log.

---

7. cinnamonDBus: fix crash in updateSetting when instance_id is not found

updateSetting() already guarded against an unknown uuid, but not against an unknown instance_id within a valid uuid. Accessing uuids[uuid][instance_id].remoteUpdate() when instance_id is absent throws "Cannot read property 'remoteUpdate' of undefined". Added a second guard with a warning log.

---

8. cinnamonDBus: remove leftover debug comment in GetInputSources

Removed a stray // global.log(source.preferences); debug line accidentally left in GetInputSources().

---

Test plan

• PCI device description cleanup in network manager shows all underscores/commas replaced and all parenthesised segments stripped
• Keyboard layout switching works correctly (no regressions from var fix)
• Notifications with non-standard urgency values display the information icon
• No crash when a short-expiry notification fires before its source PID is resolved
• Display-change countdown dialog shows correct plural form ("5 seconds", "1 second")
• No crash in cinnamon-settings when calling callbacks or updating settings for a removed xlet
• GetInputSources D-Bus method works correctly

🤖 Generated with Claude Code

[mtwebster]

.... uuid] Unable to find instance 'instance_id' from SettingsManager' is enough, keep the message short.

[mtwebster]

I'm ok with everything but expired notification changes -

You're removing the ndata from the expire queue, but it's still (theoretically) in flight with GetConnectionUnixProcessIDRemote - now once that completes this notification may never expire. Why not consider just restarting the timer?

It's all very unlikely in any event - this can only happen if a notification sets a custom timeout of under 1 second, and the dbus lookup takes longer than that (extremely unlikely for a local call like that).

Considering the complexity handling this case would add, I would just leave this code alone, unless you have a real-world scenario where this occurs.

The remainder of the PR looks fine (other than the overly-long log warning).