Skip to content

Linux/Wayland: selecting external display leaves Omarchy bar unresponsive until shell restart #4119

Open

Description

@ginodaloisio

Description

Selecting the external display for capture in Cap makes the entire Omarchy top bar on that display unresponsive. The laptop display's bar and tray remain usable, so recording can be started and stopped from there. Quitting Cap completely does not restore the external bar; running omarchy restart shell does.

Expected: selecting a capture display should leave both monitors' desktop bars interactive, and ending capture or quitting Cap should release any capture-related UI state.

The trigger and recovery were observed by the user. The underlying cause has not been established: this may be an interaction between Cap's picker/clean-capture UI, Hyprland and Omarchy's Quickshell bar. No claim that an invisible overlay or input grab is the cause has been verified.

Reproduction

  1. Run Cap on Omarchy/Hyprland with an integrated laptop display and an external display.
  2. Choose Studio mode and select the external display for recording.
  3. The external display's entire top bar stops responding to interaction.
  4. Cap presents the clean-capture prompt asking to activate its Stop tray icon before recording starts.
  5. Use the integrated display's tray icon to start and stop recording. Recording works, but the external bar remains unresponsive.
  6. Quit Cap completely: the external bar still does not respond.
  7. Run omarchy restart shell: the external bar responds again.

Additional Context

Installed versions, read from pacman:

  • Cap: cap-bin 1:0.6.3-1 (official Arch download)
  • Omarchy: 4.5.0-3
  • Hyprland: 0.56.2-4
  • Quickshell: 0.3.2.r7.g2b4501a-1
  • xdg-desktop-portal: 1.22.1-2
  • xdg-desktop-portal-hyprland: 1.4.1-2
  • Linux x86_64, Wayland
  • GPUs: NVIDIA GeForce RTX 3050 Mobile and AMD Radeon 680M; which GPU drives each output was not verified.

Configured outputs:

hl.monitor({ output = "eDP-1", mode = "1920x1080@144", position = "0x0", scale = 1.6 })
hl.monitor({ output = "HDMI-A-1", mode = "preferred", position = "-1920x0", scale = 1 })

Relevant Cap log excerpts (timestamps omitted):

INFO cap_gpui::target_overlay: overlay start pressed target="Display"
INFO cap_gpui::hotkeys: Using the recording Stop tray instead of GlobalShortcuts error=The portal did not describe the granted Stop shortcut
INFO cap_gpui::hotkeys: recording Stop control ready label=Activate the Cap Stop tray icon to start. Activate the same icon again to stop. Recording will not start until you use that control.

Screenshot attempts in the same two-display setup also fail with:

ERROR cap_gpui::screenshot: screenshot capture failed: capturing the screenshot: Wayland screenshot portal cannot safely isolate the selected display

It is not established whether the screenshot error and stuck bar share a cause. Related tray-label report #3959 does not cover the persistent external-bar failure described here.

Desktop behavior above is user-observed; package versions, configuration and log excerpts were inspected locally. No recording attachment or independent automated desktop reproduction is available.

Activity

  1. linear commented on Nov 12, 2026

    @linear
  2. samanyudas commented on Nov 14, 2026

    @samanyudas
    Contributor

    Note

    🤖 Claude Opus 5.5 responding on behalf of samanyudas

    The stuck bar is a Quickshell bug, tracked upstream as quickshell-mirror/quickshell#1268. Cap's picker is a normal trigger for it, not the cause.

    Mechanism.

    1. On Wayland, Cap's display picker is a fullscreen xdg_toplevel on the selected output, which maps and closes within seconds.
    2. Omarchy 4.5.0's bar (shell/plugins/bar/Bar.qml) calls Hyprland.refreshToplevels() on every fullscreen event. It sets its input mask to an empty region while any toplevel on the monitor's active workspace reports fullscreen & 2 (BarModel.fullscreenOn). The bar stays visible but takes no clicks.
    3. If a window closes while that j/clients request is in flight, and Quickshell handles closewindow before the response, refreshToplevels() creates a new toplevel from the stale response. No later closewindow arrives for it, so the toplevel stays in Hyprland.toplevels with fullscreen: 2 until the shell restarts.
    4. The bar on that monitor then keeps an empty input mask. Quitting Cap doesn't help because the ghost lives in Quickshell. omarchy restart shell clears it.

    Reproduction (Hyprland 0.56.2, Quickshell master 9b0f339).

    • A QML probe runs the bar's exact fullscreenOn check from Omarchy v4.5.0 BarModel.js. It busy-waits 40 ms per fullscreen event to stand in for a busy shell.
    • A script opens a fullscreen window and kills it as soon as fullscreen>>1 arrives.
    • Over 30 rounds, stock Quickshell created 29 toplevels from j/clients responses for windows it had already closed. Ghosts with fullscreen=2 were left at the end, and the probe's underFullscreen stayed true with no fullscreen window open.
    • Cap's own display picker (CAP_GPUI_AUTO_OVERLAY=display) maps a fullscreen: 2 toplevel, which turns the bar's check true while the picker is open. In my single-display Cap runs, the picker stayed up about 1 s before closing, and I did not see Cap itself leave a ghost.
    • I couldn't reproduce the exact dual-monitor selection here, because this machine has one display.

    Fix. A 14-line Quickshell change fixes it locally. It records addresses closed while a j/clients request is in flight and skips them in the response. Under the same 30-round race it skipped 22 stale entries and left 0 ghosts. Quickshell's contribution rules require a human submitter, so the patch hasn't been sent upstream yet.

    Why no Cap change. Cap can't close its picker in a way that avoids the race, because the timing is inside Quickshell. Replacing the picker with a layer-shell surface would sidestep the bar's fullscreen logic, but that is a large change for a bug that any short-lived fullscreen window triggers.

    Workaround until a fixed Quickshell ships in Omarchy. Run omarchy restart shell.

    Leaving this open to track the upstream fix.

  3. samanyudas commented on Nov 14, 2026

    @samanyudas
    Contributor

    Note

    🤖 Claude Opus 5.5 responding on behalf of samanyudas

    The Quickshell fix is now up for review as quickshell-mirror/quickshell#1490. It skips windows that closed while a j/clients reply was pending. A deterministic repro showed a ghost on every run before the fix and none after.

    This issue stays open until a Quickshell release with the fix reaches Omarchy. Until then, omarchy restart shell clears a stuck bar.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions