Skip to content

Find the bar's monitor through the live monitor list - #14162

Open
Rishab-Agrawal wants to merge 1 commit into
omacom:quattrofrom
Rishab-Agrawal:fix/bar-under-fullscreen
Open

Rishab-Agrawal wants to merge 1 commit into
omacom:quattrofrom
Rishab-Agrawal:fix/bar-under-fullscreen

Conversation

@Rishab-Agrawal

@Rishab-Agrawal Rishab-Agrawal commented Oct 3, 2026 •

Copy link
Copy Markdown

Follow-up to #16991, for the un-mirroring path in #10323.

Problem

#16991 drops the bar's input while the active workspace on its screen has a real fullscreen window. It finds that screen's monitor with Hyprland.monitorFor(screen), which is a one-off call. Un-mirroring removes the output and adds it back as a new monitor object, and the binding keeps reading the old one. On the omarchybot worker (Hyprland 0.57.0), fullscreening a terminal, mirroring and un-mirroring a second output, and moving the workspace onto it left the bar taking clicks there: clicking where the clock sits opened the calendar over the fullscreen terminal.

Fix

underFullscreen finds the monitor in Hyprland.monitors.values by screen.name, the same live lookup OverlayWindow uses, so it follows the replacement monitor. Everything else in #16991 stays as it is, including the mode-bit-2 check, so maximized windows keep a working bar.

What changed in this PR

This PR first moved the bar to the bottom layer under fullscreen. #16991 has since fixed the click problem in #14079 another way, and the omarchybot review on Hyprland 0.57.0 found that the layer move left the bar missing after fullscreen ended. So that change is gone, and this PR is now only the lookup.

#10323 should stay open after this. In the worker run the bar also drew over the fullscreen window after un-mirroring, and this change only stops it taking the clicks there.

Testing

  • bash test/shell.d/bar-test.sh gives the same results with and without the change.
  • I haven't run it on a live Hyprland session. The worker steps above are the check: after un-mirroring and moving the fullscreen workspace, a click where the clock sits should reach the window.

@greptile-apps

greptile-apps Bot commented Oct 3, 2026 •

Copy link
Copy Markdown
Contributor

RetriggerView in GreptileConfidence Score: 5/5

[Medium impact] This PR appears safe to merge; no actionable issues were found in the monitor lookup change.

Summary

Updates underFullscreen in Bar.qml to find the screen's monitor through Hyprland.monitors.values.

  • The bar finds the current monitor before checking fullscreen.

The earlier, unnumbered test suggestion is resolved. The current tests check fullscreen behavior in fullscreenOn, rather than the removed layer change.

Reviews (4) · Last reviewed commit: "Find the bar's monitor through the live ..." · Reviewed by Greptile

Comment thread test/shell.d/bar-test.sh Resolved
@omarchybot

Copy link
Copy Markdown
Collaborator

Not reviewed yet. The second opinion (Codex Medium) is needed to settle what this pull request is, and it is over today's review limit, so it never ran. Without it this gets no label, no worker run and no verdict. It goes round again once reviews are available.

What I did check, by reading only: quattro still puts the bar on WlrLayer.Top unconditionally (shell/plugins/bar/Bar.qml), so #14079 isn't fixed there. The branch no longer merges cleanly with quattro: Bar.qml merges without trouble, but test/shell.d/bar-test.sh conflicts. #13776 (screensaver) is still open and changes the same WlrLayershell.layer binding.

Nothing was run, and this is waiting on the second opinion.

@scottjones

Copy link
Copy Markdown
Contributor

I hit this on an M2 Max (Hyprland 0.56.2, dwindle, bar on top): in fullscreen, clicking where the bar sits opened the audio, display and battery panels.

It isn't only a remap. I watched each layer's above_fullscreen with hyprctl eval while a terminal was fullscreen (Super+F). It read false for about 20 seconds, and then the notifications layer closed when a card timed out. At that moment every layer on the monitor flipped back to true with the terminal still fullscreen, and from then on the bar took the clicks. So any notification during a fullscreen video call is enough.

I ran this branch, merged with current quattro, as my live shell. Under fullscreen the bar moved to the bottom layer. After a notification came and went, clicks along the top, over the bar's icons and along the dock all went to the window. The bar worked again once fullscreen ended.

It needs a rebase: against current quattro it conflicts only in test/shell.d/bar-test.sh.

@manuaudio

Copy link
Copy Markdown
Contributor

@scottjones, thanks for testing the notification-remap path. That live result covers the compositor behavior the source-level test and Greptile review could not: the bar stays below fullscreen after the notification layer remaps, clicks reach the window, and the bar recovers when fullscreen ends. I linked this result back to #14079. The branch still needs its bar-test.sh conflict resolved against current quattro before maintainer review.

@Rishab-Agrawal
Rishab-Agrawal force-pushed the fix/bar-under-fullscreen branch from aa0cb24 to 7ec3b2d Compare October 28, 2026 14:43
@Rishab-Agrawal

Copy link
Copy Markdown
Author

Rebased onto current quattro. I dropped the source-text assertion in bar-test.sh, since #16334 removed that kind of check, so the PR is now just the Bar.qml binding and the conflict is gone.

@scottjones, thanks for running it live and for finding the notification path. I've added both to the description and the code comment.

@manuaudio

Copy link
Copy Markdown
Contributor

@Rishab-Agrawal, thanks for rebasing this and removing the stale source-text assertion. The conflict is resolved. One concrete worker finding from #10323 still applies to this head: Hyprland.monitorFor(screen) does not re-evaluate when an output is removed and added again, so after un-mirroring the destination bar stayed on Top over the fullscreen window. Looking the monitor up through Hyprland.monitors.values by screen.name fixed both the normal workspace move and the un-mirroring path in repeated worker runs. Could you use that reactive lookup here? That would let this PR close #10323 as well as #14079.

@Rishab-Agrawal

Copy link
Copy Markdown
Author

Switched the lookup to Hyprland.monitors.values by screen.name, in the same form OverlayWindow uses, so the binding follows an output that is removed and added back. Pushed as a new commit on top. The description now closes #10323 as well.

@manuaudio

Copy link
Copy Markdown
Contributor

@Rishab-Agrawal, thanks. I checked a1f852f: the bar now uses the same Hyprland.monitors.values lookup pattern as OverlayWindow, so the binding can follow a replacement monitor object after un-mirroring. That addresses the #10323 worker finding, and the current Greptile check passes.

@omarchybot omarchybot added the bug Something isn't working label Nov 2, 2026
@omarchybot

Copy link
Copy Markdown
Collaborator

Reviewed a1f852f on a worker built from the Omarchy ISO (Hyprland 0.57.0, quickshell 0.3.2), against quattro at 175e63b. Two problems block it as it stands.

1. Merged into current quattro, the shell no longer starts. #16991 (merged 2026-10-29) added its own readonly property bool underFullscreen to the same BarPanel. Git merges the two cleanly, so GitHub shows no conflict, but the result declares the property twice (Bar.qml lines 1811 and 1816 of the merge), and on the worker omarchy-shell restart-looped with Bar.qml[1816:28]: Duplicate property name.

2. After leaving fullscreen, the bar doesn't come back. On a single monitor I opened a terminal, pressed fullscreen and pressed it again. On this head (and on a port that resolves the duplicate property), hyprctl layers still has omarchy-bar on the bottom layer, and nothing is drawn in its reserved strip. It was still missing after a minute and after switching workspaces. On quattro the same steps bring the bar back.

quattro: after leaving fullscreen, the bar is back
This head: after leaving fullscreen, the bar's strip is empty

The second opinion found the mechanism, and it matches what the worker showed. Quickshell sends set_layer for a layer change but doesn't commit the surface, and the request only takes effect on the next commit. Hyprland, meanwhile, fades layers out under fullscreen and restores only top-layer surfaces when fullscreen ends, so a bar parked on the bottom layer keeps alpha 0. This may not show on 0.56.2, where @scottjones saw the bar come back, but 0.57.0 is what the ISO installs now.

What still stands on quattro. #16991 already drops the bar's input under a real fullscreen window, which covers the ordinary #14079 case. But it finds the monitor with Hyprland.monitorFor(screen), and the un-mirroring path in #10323 is still broken. On quattro I fullscreened a terminal, mirrored and then un-mirrored a second output, and moved the workspace onto it. Clicking where the clock sits then opened the calendar, and the bar drew over the fullscreen window:

quattro: after un-mirroring, a click over the fullscreen terminal opens the calendar and the bar draws on top

So the part of this PR that quattro still needs is the Hyprland.monitors.values lookup. Feeding it into #16991's existing underFullscreen keeps the mode-bit-2 test, so maximized windows keep a working bar. As written, hasFullscreen is also true for maximized windows, so this head would move the bar to the bottom layer whenever a window is maximized. Whether the layer move is still wanted on top of that, or whether the lookup alone is enough, is for the author and the maintainer to decide. If the layer move stays, it needs a way back to a visible top-layer bar after fullscreen ends.

Both models on the second-opinion panel, Codex Medium and muse-wonder via pi at high thinking, reached the same two blockers. Their independence isn't guaranteed, because they could read the worker notes from this review. The Quickshell and Hyprland mechanism above came from them. They also corrected one of my hypotheses: bar tooltips don't sink with the bar, because Hyprland draws layer popups after windows.

Nothing was pushed to the branch. This is waiting on @Rishab-Agrawal.

@Rishab-Agrawal
Rishab-Agrawal force-pushed the fix/bar-under-fullscreen branch from a1f852f to 74afc40 Compare November 4, 2026 21:06
@Rishab-Agrawal Rishab-Agrawal changed the title Step the bar below a fullscreen window so it stops taking its clicks Find the bar's monitor through the live monitor list Nov 4, 2026
@Rishab-Agrawal

Copy link
Copy Markdown
Author

Thanks for the review. I've dropped the layer move, since #16991 already covers #14079 and the bottom-layer bar didn't come back on 0.57.0. The branch is now one change on current quattro: #16991's underFullscreen finds its monitor through Hyprland.monitors.values instead of monitorFor, so the un-mirroring path in #10323 stops taking clicks. The bar still draws over the window in that case, so I've changed the description to stop closing #10323.

@manuaudio

Copy link
Copy Markdown
Contributor

@Rishab-Agrawal, thanks for narrowing this to the part that still applies. I checked 74afc409: the layer move is gone, underFullscreen still uses #16991's fullscreenOn path, and the monitor now comes from the live list so it can follow the replacement object after un-mirroring. The updated description also keeps the remaining draw-over behavior explicit instead of claiming to close #10323. Greptile is passing on this head.

@omarchybot omarchybot added the verified Omarchy Triage has verified that this issue is ready for final review label Nov 17, 2026
@omarchybot

Copy link
Copy Markdown
Collaborator

Re-reviewed 74afc40 merged into current quattro (3b36b9e), on a worker built from the Omarchy ISO (Hyprland 0.57.2). The fix works, and I found nothing wrong with it.

Reproduction. I fullscreened a terminal and mirrored a second output onto that display, then un-mirrored it. After doing that twice, I moved the fullscreen workspace onto the second output and clicked where the clock sits. On quattro the calendar opened over the terminal. That happened in three runs, two of them from a fresh boot. With a single mirror cycle it didn't happen, so the repeat matters for anyone retrying this. On this head, the same steps left the click with the terminal after two cycles and after four. With the change reverted, the calendar opened again.

quattro: after un-mirroring twice, a click over the fullscreen terminal opens the calendar
This head: the same click stays with the terminal

Nothing else moved. On a single monitor, a fullscreen window still keeps the bar from taking clicks, a maximized window still leaves the bar clickable, and the bar comes back once fullscreen ends. All of test/shell.d/bar-*-test.sh and ./test/cli pass. ./test/shell has 5 failures (ascii, branding-about-animation, first-run, provision-owner-keyboard, runtime-smoke), and each one fails the same way on quattro on the same worker.

About #10323's drawing half. In these runs the bar was never drawn over the terminal before the click, on either branch. On quattro it showed up together with the calendar. I didn't see the bar draw over the window by itself, so these runs can't say whether that part still needs a fix. Leaving the issue open, as the description does, is right.

Overlap. #18381 and #19037 both move this same underFullscreen lookup into a helper that still calls Hyprland.monitorFor(screen). Whichever of them lands after this one would need to keep the monitor-list lookup.

The second opinion came from Codex Medium and muse-wonder via pi at high thinking, and neither found a defect. Both judged the change proportionate and the binding reactive to an output being removed and added back, through Hyprland.monitors.values and the matched monitor's activeWorkspace. They were given the worker results above, so their agreement isn't independent.

Nothing was pushed to the branch. The ready label was refused only because the full shell suite isn't green on the worker, and those 5 failures are already on quattro. This is waiting on the maintainer.

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

Labels

bug Something isn't working verified Omarchy Triage has verified that this issue is ready for final review

Projects

None yet

Development

Successfully merging this pull request may close these issues.

4 participants