Repository navigation
Fix empty-desktop bar reposition by Top-only expand (#11915) - #12646
paulogeyer wants to merge 2 commits into
Conversation
|
No production defect was identified in the source review. The empty-desktop drag fix still needs independent live validation. Verified by source inspection: only the moving panel expands, remains on the Top layer, and keeps its strip content at the original edge. The gesture area grows with the panel; release/cancel clears the move. The Overlay previews remain input-transparent. Optional test improvement: the four-anchor assertion can match
Constrain the match to the Review informationTest scope: Head Community review: Independent automated community review, unaffiliated with the Omarchy team, intended to help prepare PRs for their review. Automated AI review: Astra Medium initial inspection, independent Opus 5.5 High and GPT 6 Sol Xhigh assessments, followed by coordinator verification of the cited source and assertion results. |
|
Reviewed at 8aeea9e by Claude Opus 5.5, with Codex Medium as a second opinion, and run on a worker VM built from the Omarchy ISO (Hyprland 0.56.2, one output, a virtual mouse that presses, holds, drags and releases). The fix works for the reported case. On Three things are still open:
The exclusion change is a visible trade-off, as the description says. While the bar is expanded it reserves no space, so tiled windows resize at the start of every move, including on occupied workspaces and cancelled drags, and again when it ends. Compared with #12184, both reviews prefer this approach. It keeps the original press on one surface rather than handing the gesture to an overlay that can't see the held button. The choice between them is still the maintainer's. This now waits on @paulogeyer for the notification and peek issues, then on the maintainer. |
|
This no longer merges into What conflicts. In What a resolution keeps. Keep Still open after the merge. Crossing the notification cards on an empty desktop still ends the drag, because their Overlay input mask is unchanged ( This waits on @paulogeyer to bring the branch up to date and address the two findings. After that it goes to the maintainer, who still has to choose this approach. |
|
Fixes #11915 (bar reposition/drag dies on an empty desktop) via Top-only expand while moving. Approach makes sense. Currently CONFLICTING on Bar.qml/BarModel.js — needs a rebase before review. Happy to re-check the gesture edge cases (multi-monitor, vertical bar) once it’s green. |
8aeea9e to
b7aba24
Compare
|
Thanks for the detailed reviews. I rebased onto current
I've live-soaked it on my laptop, dragging the bar between all edges with and without windows, and had no issues. The description is updated. Thanks @jandrusk for the review, and for offering to re-check multi-monitor and vertical bar layouts. It's green and ready for that whenever you have time. |
|
Reviewed again at b7aba24 against The fix works, and both earlier findings are fixed. On Two things are still open:
Codex Medium also checked the peek bookkeeping, the bottom/right coordinate offset during the expand, This waits on @paulogeyer to trim the tests, and either to look into the wake case or to name it in the description. After that it goes to the maintainer. |
|
This no longer merges into What conflicts. All four conflicts are in What breaks even where it merges cleanly.
What a resolution keeps. From this PR: the owning panel's expand to all four anchors, Still broken on This waits on @paulogeyer to bring it onto current |
b7aba24 to
f3ab44a
Compare
|
Rebuilt on The test additions are trimmed to the four checks you named, once each: the per-edge anchor check, the strip offsets, the Virtual mouse on an empty workspace on my laptop: I couldn't reproduce the wake case on my laptop: the first drag after a |
[Medium impact] The PR appears safe to merge; the previous cloned-toast issue is fixed. SummaryThe PR expands the moving bar across its screen on the Top layer and disables toast input during the move.
|
|
Reviewed again at f3ab44a against Both points from the last review are addressed. The test additions are down to the four checks, and the rebuild onto #15517 keeps one panel-level gesture area and settles What works. On What still fails. With the press held 0.4 s, half the empty-desktop moves still drop (6/12). I added logging to see why. Press-and-hold starts the move at 200 ms, and the panel then takes 113–178 ms to reach the full output (once it still hadn't after 188 ms). Every failed move ended in Codex Medium added two points that follow from this:
It goes to the maintainer now, to decide whether to take it with the race as it stands or ask for the earlier start. Either way, the description should state the limitation as above. Codex Medium agreed with the rest of this. It can read this session, so that agreement isn't guaranteed to be independent. |
|
Reworded the Known limitation section: any move that leaves the strip before the expanded bar takes effect can drop, it depends on draw speed, and quattro has the same race. Left expand-on-press out of this PR. |
|
Read again at f3ab44a by Claude Opus 5.5, after the description update. The head hasn't moved, so the worker results from the last review still stand. The Known limitation section now states the race the way the last review asked. It says any move that leaves the strip before the expanded surface takes effect can drop, that how often depends on draw speed, and it gives the 0.4 s figures, which cover the press-and-hold case. One wording point, not a blocker: expanding on press would narrow the race, not close it. The compositor still has to configure and draw the larger surface before the pointer leaves the strip, and expanding on press only gives it the hold time to do that. No second opinion ran this time. Codex Medium answers, but today's review budget is spent, so this read is mine alone. This waits on the maintainer to decide whether to take it with the race as it stands or ask for the earlier start. Before it can be marked |
While a bar is being moved, its panel fills its output on the Top layer, so the one panel-level gesture area that took the press keeps getting moves and the release after the pointer leaves the strip (omacom#11915). The strip stays on its own edge, exclusion is Ignore for the move, and the inner gesture area in CenterModules is gone. Toast cards take no input during a move, so a drag can cross them. A move that relocates the bar or ends off the strip settles the moving panel's hover in barHoverCount; a press-and-hold released in place keeps the peek.
f3ab44a to
2709fa7
Compare
|
Reviewed again at b325725 against The fix works on the reworked notification deck, and nothing is open. With the press held 1.5 s before moving, an empty-desktop move fails every time on Greptile's point about cloned notifications plugins holds, and b325725 fixes it: those plugins get a At a 0.4 s hold, 6 of 8 moves worked. Both misses were top→bottom, released at y=15 just as the pointer left the strip. That's the race the Known limitation section describes, and the description states it accurately. Codex Medium read the whole diff at this head and found no defects. It judged the change proportionate to the bug and agreed the Known limitation wording matches the code. It can read this session, so its agreement isn't guaranteed to be independent. This now waits on the maintainer. |
Summary
BarPanelto fill its output onWlrLayer.Top, so the sameCenterGestureAreakeeps receiving moves and the release. One panel-level gesture area per orientation; the strip (sized by the bar's notch-aware thickness) stays pinned to the original edge, and the move-edge ghosts stay Overlay and input-transparent.Relation to #12184
#12184 is still open and takes a different handoff/overlay approach. Live soak on my laptop: that PR still fails on empty desktop. This PR is an intentional competing fix that passed empty + windowed press-hold-drag-release soak on my laptop (ASUS Vivobook, Intel Arc).
Updated for current
quattroRe-done on
quattro246f33a after #15517 rather than rebased. The strip wraps #15517's widget pools andCenterSections/CenterBoxunchanged, the panel-levelCenterGestureAreais the only one (the one insideCenterModulesis gone), andclearBarMoveno longer touches the removedcenterSectionHovered. Instead, when the bar relocates or the move ends off the strip, the moving panel'sHoverHandlergives back the hover it counted inbarHoverCount, and it tracks what it counted so a late leave never subtracts twice. A press-and-hold released in place keeps the peek.Trade-off
While expanded, exclusion is
Ignore, so tiled clients reflow into the strip area for the drag (including on occupied workspaces and cancelled drags), and again on release. I prefer this to the Overlay re-layer blink.Known limitation: moves that leave the strip before the expanded bar takes effect
The fix relies on the expanded bar surface being in place before the pointer leaves the bar strip. Any move that leaves the strip before that happens can drop, and how often depends on how quickly the machine draws the expanded surface. That covers quick flicks, and a slower first draw such as right after a display wake. In review, a 1.5 s hold moved the bar 8/8, while a 0.4 s hold dropped 6 of 12 moves. On my laptop (Hyprland 0.56.2, one output), with a 0.3–0.4 s hold, the first drag after a wake worked 19/19, and a flick moving within about 20 ms of the press always failed.
quattrohas the same race (0/3 on a quick flick with a window open, 3/3 with a normal hold), so this change doesn't introduce it. Expanding on press would narrow the race, giving the compositor the hold time to draw the larger surface, but not close it, and it reflows tiled windows on every click on the bar, so I've left that out of this PR.What this deliberately avoids
HyprlandFocusGrab/ seat grabpressedButtonsoverlay handoffTop→Overlayduring the move (visible blink)Test plan
bar-test.sh269 ok,bar-notch-test.sh99,notifications-test.sh150,clock-test.sh167,bar-drag-test.sh,bar-click-clipping-test.sh,bar-text-color-test.shpass. Added checks: each anchor inside theBarPanelanchors block, the bottom/right strip offsets, apointInBarStriptable and the toast input mask, once each. Onquattrowith these tests, bar, notch and notifications fail.quattro, 0/8 across top→left→bottom→right→top twice. With this PR, 8/8, and 3/3 across toast cards.Service.qmlreverted. Glassy background: 4/4 drags (the blur itself not checked).Fixes #11915
Fixes #7923