Sdwdate-gui systray is not appearing in Whonix-Gateway sysmaint session

nurmagoz · June 29, 2026, 10:31am

This is reproducible in Vbox and KVM.

Patrick · June 29, 2026, 11:06am

Cannot reproduce. Testers repository. Whonix-Gateway, Whonix-Workstation, both have sdwdate-gui systray.

nurmagoz · June 29, 2026, 11:19am

Fresh download of current available version
Import
Upgrade
systray disappear

Any command to run it from terminal and see some verbose log?

Patrick · June 29, 2026, 2:23pm

Might be a race condition.

Analyzed with AI.

github.com/org-ai-assisted/sdwdate-gui

Fix tray icon not appearing in the Whonix-Gateway sysmaint session (tray-host startup race)

master ← tray-icon-startup-race-fix

opened 01:23PM - 29 Jun 26 UTC

assisted-by-ai

+88 -4

Fixes the intermittently-missing sdwdate-gui tray icon reported for the Whonix-G…ateway sysmaint session: https://forums.whonix.org/t/sdwdate-gui-icon-is-not-appearing-in-whonix-gw-sysmaint/23342 ## Root cause `QSystemTrayIcon` binds its backend -- legacy XEmbed vs. the StatusNotifier (SNI) D-Bus protocol -- when it is **constructed**. `sdwdate-gui-server` constructed the tray icon and called `show()` unconditionally at startup. When the server starts before the panel's `StatusNotifierWatcher` has registered on the session bus -- a race, notably in the user-sysmaint-split sysmaint session where the tray host (trayer/waybar) and sdwdate-gui are launched together with no ordering -- Qt falls back to XEmbed, and an SNI-only panel (lxqt-panel, waybar) never shows the icon. Result: an intermittent, hard-to-reproduce missing tray icon. Re-showing or calling `show()` again after the host appears does not recover it; only deferring construction does. ## Fix (2 commits) 1. **Defer tray icon construction** until `QSystemTrayIcon.isSystemTrayAvailable()` is true, so Qt selects the SNI backend instead of falling back to XEmbed. 2. **Decouple the IPC listener from the deferral.** Deferring the whole `SdwdateTrayIcon` would also defer the `SdwdateGuiListener` it creates, delaying the server socket. On Qubes the gateway's `sdwdate-gui-qubes-proxy-helper` waits only a bounded ~20s for that socket before giving up and never linking `sdwdate-gui.Connect`. So the listener now comes up immediately and only the `QSystemTrayIcon` is deferred; clients that connect before the tray exists are buffered and replayed (a client completes its handshake over its own socket regardless). The install retry is idempotent (construct first -> commit + stop timer -> one-time signal handoff). ## Verification Reproduced the bug and confirmed the fix against **real SNI tray hosts** -- waybar (the actual sysmaint Wayland host) and xfce4-panel -- under a controlled control/bug/fix matrix: with a late host, the old blind `show()` never registers a StatusNotifierItem (icon missing), while the deferred construction does (icon shows). Also verified the socket comes up with no tray host at all, and that a pre-tray client is buffered and replayed. Regression tests live in the dist-ai test suite (`sdwdate-gui-tests`): an offscreen unit test for the deferral/decoupling/buffering invariants, plus an SNI late-host integration phase driving a real `org.kde.StatusNotifierWatcher`. Reviewed with Codex + CodeRabbit + an independent pass; their findings (the Qubes socket-coupling and the non-idempotent retry) are folded in. --- Generated with [Claude Code](https://claude.com/claude-code) Session: https://claude.ai/code/session_01CzEvBanrdqYJaEcZEMgt8L ## Summary by CodeRabbit * **New Features** * Improved tray icon startup reliability when the desktop system tray is not immediately available. * Early incoming connections are now queued and handled once the tray icon appears. * **Bug Fixes** * Prevents missed or duplicated client handling during startup. * Reduces failures caused by delayed tray availability.

Assigned to @arraybolt3.

Disappear meaning sdwdate-gui is visible for a while and then disappearing?

nurmagoz · June 29, 2026, 9:56pm

at first importing of the VM, then after upgrade, gone.

arraybolt3 · June 30, 2026, 3:43am

Was able to reproduce the issue. Looks like the result of a botched refactor on my part. Fixes:

Patrick · July 2, 2026, 12:03pm

Merged. This is now in the testers repository.

nurmagoz · July 2, 2026, 12:41pm

Same result from my test, sysmaint not appearing, user session appear.

This is reproducible on KS as a host.

Patrick · July 2, 2026, 2:36pm

github.com/org-ai-assisted/sdwdate-gui

Comment by assisted-by-ai - Fix tray icon not appearing in the Whonix-Gateway sysmaint session (tray-host startup race)

master ← tray-icon-startup-race-fix

@ArrayBolt3 You're right on the primary cause, and I was wrong to frame this PR …as *the* root cause of the reported symptom. Your investigation nails it: in the sysmaint session the server wasn't running at all -- `user-sysmaint-split` was autostarting a non-existent `sdwdate-gui-server.desktop` when the real launcher is `sdwdate-gui.desktop`. Your two fixes ([sdwdate-gui a9d64a4](https://github.com/ArrayBolt3/sdwdate-gui/commit/a9d64a4e8a736e32627c87b96d16c71074a70cfa) and [user-sysmaint-split fbfecd1](https://github.com/ArrayBolt3/user-sysmaint-split/commit/fbfecd1d6dec77dbbbfa5e1b11477663c95e6066)) are the correct and necessary fix, and they explain the reported "icon never appears" far better than a tray race does. That should land regardless of this PR. Where I'd push back is the narrower claim that Qt always recovers when the tray host appears late. On the [Qt-doc citation](https://doc.qt.io/archives/qt-5.15/qsystemtrayicon.html#isSystemTrayAvailable): the "becomes available later" wording covers Qt re-adding the entry on the backend it has *already* bound -- it does not cover switching an already-constructed icon from the legacy XEmbed backend over to StatusNotifier. The backend is chosen at **construction** (Qt queries for the `StatusNotifierWatcher` when the `QSystemTrayIcon` is created), and that choice is permanent for that icon. I wrote a minimal, self-contained reproducer to check this rather than argue from the docs -- pure Qt, no sdwdate-gui import, two arms that differ *only* in ordering: [**`tray_startup_race_reproducer.py`**](https://github.com/org-ai-assisted/dist-ai/blob/72cd0add57c0a9e30e70f4508b2807182e3aa351/usr/share/sdwdate-gui-tests/tray_startup_race_reproducer.py) | arm | order | result | |-----|-------|--------| | `late` | construct+`show()`, then bring the watcher up 1s later | **never registers via SNI** (stuck on XEmbed) | | `early` | bring the watcher up first, then construct+`show()` | registers via SNI | ``` $ xvfb-run -a python3 tray_startup_race_reproducer.py RESULT {"arm": "late", "registered_via_sni": false} RESULT {"arm": "early", "registered_via_sni": true} VERDICT {"expected": {"late": false, "early": true}, "got": {"late": false, "early": true}} PASS: race confirmed -- same construct+show(), the icon registers via SNI only when the watcher exists BEFORE construction. ``` Deterministic on Qt 5.15.15 / xcb -- 10/10 runs each way. Same construct+show in both arms; only whether the watcher exists *at construction time* changes the outcome. `isSystemTrayAvailable()` does correctly flip to `true` in the `late` arm -- but the already-constructed icon still never migrates to SNI. (The same check is now a CI regression guard: a [`blind` negative-control arm](https://github.com/org-ai-assisted/dist-ai/blob/72cd0add57c0a9e30e70f4508b2807182e3aa351/usr/share/sdwdate-gui-tests/integration/sni_late_host_harness.py) that asserts the pre-fix path does *not* register, alongside the `deferred` arm that asserts the fix does.) I think this reconciles your kill-Waybar/restart test too: there the `org.kde.StatusNotifierWatcher` name was almost certainly still owned by a persistent panel/DE component at the moment the server constructed its icon, so Qt bound SNI from the start. On a cold sysmaint boot nothing owns that name until the tray host launches, and since the sysmaint session starts the tray host and sdwdate-gui together with no ordering, a server that wins the race constructs before the watcher exists and lands in the `late` case. Concretely: after your desktop-file fix the server does start, but master still constructs the tray unconditionally -- https://github.com/org-ai-assisted/sdwdate-gui/blob/239f7e828a7543ba2323de909f241cacfac2fd35/usr/lib/python3/dist-packages/sdwdate_gui/sdwdate_gui_server.py#L1458-L1459 -- so this residual race can still hide the icon intermittently. So I'd like to reframe this PR from "root cause" to "defense-in-depth on top of your fix": deferring construction until `isSystemTrayAvailable()` removes the race entirely. If you'd rather not carry it, that's fine and your fix is the important one -- but the race is real and reproducible, so I wanted to show the evidence rather than just assert it.

arraybolt3 · July 3, 2026, 12:34am

Can you share the output of the following three commands:

grep -A15 'early_user_service_list=(' /usr/libexec/user-sysmaint-split/sysmaint-session-shared
journalctl --no-pager --user --boot -u sdwdate-gui.service
systemctl --user status sdwdate-gui.service

nurmagoz · July 3, 2026, 4:51am

Sysmaint:

[sysmaint ~]% grep -A15 'early_user_service_list=(' /usr/libexec/user-sysmaint-split/sysmaint-session-shared 
  early_user_service_list=(
     'polkit-mate-authentication-agent-1'
     'msgdispatcher'
     'nm-applet'
     'sdwdate-gui-client'
     'sdwdate-gui-server'
     'updatecheck'
     'spice-vdagent'
     'livecheck'
     'lxqt-notifications'
     'wlr-resize-watcher'
     'backlight-tool-dist-restore'
  )
  late_user_service_list=(
     'lxqt-powermanagement'
  )
[sysmaint ~]% journalctl --no-pager --user --boot -u sdwdate-gui.service
-- No entries --
[sysmaint ~]% systemctl --user status sdwdate-gui.service 
○ sdwdate-gui.service - Secure Distributed Web Date Graphical User Interface
     Loaded: loaded (/usr/lib/systemd/user/sdwdate-gui.service; static)
     Active: inactive (dead)
zsh: exit 3     systemctl --user status sdwdate-gui.service

User

[user ~]% grep -A15 'early_user_service_list=(' /usr/libexec/user-sysmaint-split/sysmaint-session-shared
  early_user_service_list=(
     'polkit-mate-authentication-agent-1'
     'msgdispatcher'
     'nm-applet'
     'sdwdate-gui-client'
     'sdwdate-gui-server'
     'updatecheck'
     'spice-vdagent'
     'livecheck'
     'lxqt-notifications'
     'wlr-resize-watcher'
     'backlight-tool-dist-restore'
  )
  late_user_service_list=(
     'lxqt-powermanagement'
  )
[user ~]% journalctl --no-pager --user --boot -u sdwdate-gui.service

Jul 03 02:39:46 localhost systemd[2903]: Starting sdwdate-gui.service - Secure Distributed Web Date Graphical User Interface...
Jul 03 02:39:46 localhost systemd[2903]: Finished sdwdate-gui.service - Secure Distributed Web Date Graphical User Interface.
[user ~]% systemctl --user status sdwdate-gui.service

● sdwdate-gui.service - Secure Distributed Web Date Graphical User Interface
     Loaded: loaded (/usr/lib/systemd/user/sdwdate-gui.service; static)
     Active: active (exited) since Fri 2026-07-03 02:39:46 UTC; 3min 16s ago
 Invocation: a47f792085d248f9ae31a88ac5563c87
    Process: 3784 ExecCondition=/usr/libexec/helper-scripts/has_netvm_on_qubes (code=exited, status=0/SUCCESS)
    Process: 3797 ExecStart=/usr/bin/sdwdate-gui (code=exited, status=0/SUCCESS)
   Main PID: 3797 (code=exited, status=0/SUCCESS)
      Tasks: 5 (limit: 57316)
     Memory: 36.4M (peak: 38.4M)
        CPU: 289ms
     CGroup: /user.slice/user-1000.slice/user@1000.service/app.slice/sdwdate-gui.service
             ├─3812 /usr/bin/python3 -su /usr/bin/sdwdate-gui-server
             └─3813 /usr/bin/python3 -su /usr/bin/sdwdate-gui-client

Jul 03 02:39:46 localhost systemd[2903]: Starting sdwdate-gui.service - Secure Distributed Web Date Graphical User Interface...
Jul 03 02:39:46 localhost systemd[2903]: Finished sdwdate-gui.service - Secure Distributed Web Date Graphical User Interface.

arraybolt3 · July 3, 2026, 5:01am

Your user-sysmaint-split isn’t up-to-date. Enable the testers repository (note that Sdwdate-gui systray is not appearing in Whonix-Gateway sysmaint session - #7 by Patrick says the update is in the testers repo, not the main repo yet) and install all updates, then reboot into sysmaint mode and see if it works any better.

nurmagoz · July 3, 2026, 9:52am

Im always on testers repo, problem either from the repo didnt push the changes, or the changes pushed but didnt resolve the issue.

So its between you and @Patrick to see where is the issue happening.

Current version installed: 3.12.5-1

Patrick · July 9, 2026, 3:35pm

My mistake. The fixed version wasn’t in the testers repository yet.

Additional sdwdate-gui fixes are scheduled soon.

Patrick · July 12, 2026, 6:00pm

Done. Now in the testers repository.

Could you please test if the issue has been resolved? @nurmagoz

nurmagoz · July 12, 2026, 10:50pm

Yes it did, great.