[steven12138]

What this does

On niri, an optional per-bar Hide bar on fullscreen setting slides the top bar out of view when the active window on that output enters fullscreen. It slides back when fullscreen ends. Both transitions use DBar's existing auto-hide animation and the theme's animation duration (including disabled animations). Moving the pointer to the screen edge can reveal the bar while fullscreen.

The option requires Use overlay layer, defaults to off, and does not turn on global auto-hide or change the bar's layer. It preserves the configured exclusive zone, so other windows do not resize during the transition. Frame-hosted bars, islands, and other compositors are unaffected.

This revision ports the feature to the Dux 2.0 DBar layout and current master. It also clears the IPC reveal state when effective auto-hide ends, as requested in the latest review; otherwise a previous bar toggleReveal could prevent hiding on the next fullscreen. The earlier edge-hover and test-embedding review findings remain addressed.

The subsequent review found that fullscreen-only hiding released the bar's exclusive zone. The fix keeps the configured zone when only the fullscreen option is active, while preserving the existing global auto-hide behavior. A regression test covers both cases.

Niri IPC and Wayland toplevels have no shared window ID. Matching uses exact app ID/title on the output, with keyboard focus to disambiguate where possible. Ambiguous matches leave the bar visible. See docs/fullscreen-bar.md for usage and limits.

Validation

• node --test quickshell/tests/*.test.mjs: 398 passed.
• make test-qml: Qt 31 passed, lock 38 passed, logic passed.
• make build: passed; embedded shell excludes tests.
• git diff --check: passed.
• make lint-qml could not run in this fresh worktree because Quickshell's VFS .qmlls.ini had not been generated. No new lint result is claimed.
• Installed the matching patch in local dux-shell 2.0.0-3. A plain dms restart loads the embedded patch, and fullscreen entry hides the top bar while exit restores it in a live single-monitor niri session. The user also confirmed the fullscreen behavior works on their machine.
• Live pointer-edge behavior and physical multi-monitor behavior on Dux 2.0 have not been checked separately. Exclusive-zone preservation is covered by the new regression test; it has not been measured live.

Checklist

• Feature is opt-in and documented.
• Settings search index regenerated.
• Tests and build pass on current master.
• Live Dux 2.0 single-monitor fullscreen entry/exit validation.
• Live Dux 2.0 multi-monitor and pointer-edge validation.

[bbedward]

/claude review

[claude]

Claude review

Two issues: the documented edge-hover reveal is dead in the fullscreen-only path, and the new test files ship inside the dms binary.

• Edge-hover reveal never fires when only hideOnFullscreen is on — topBarMouseArea.hoverEnabled/enabled still gate on barConfig.autoHide instead of topBarCore.autoHide, quickshell/Modules/DankBar/DankBarBody.qml:1075
• New scripts/test-*.mjs are not in sync-shell's strip list, so they are embedded into the shipped binary, quickshell/scripts/test-niri-fullscreen.mjs:1

Checked: niri workspace/active-window state fixes (string-vs-numeric key lookups, stale active_window_id preservation, unfocused-output focus flags) all look correct; NiriFullscreen.js matching and its binding dependencies; exclusive-zone preservation; overlay-layer and frame-host gating; toggleReveal IPC guard; settings spec/defaults and regenerated search index; catalogs untouched. Model: claude-opus-5.

[steven12138]

Addressed the requested changes and rebased onto the current master:

• Moved the Node tests to quickshell/tests/*.test.mjs; the current sync-shell staging excludes the entire tests directory, and the embed exclusion test passes.
• Fullscreen-only hiding now uses the effective topBarCore.autoHide state for edge hover and click-through input, so the pointer edge can reveal the bar when global auto-hide is off.
• Updated the bar IPC regression fixture and added coverage for toggleReveal during fullscreen auto-hide.
• Resolved the upstream settings, bar, compositor, and niri service conflicts while preserving the current APIs.

Validation: 19 focused tests pass, the full logic suite passes, Qt tests pass (26/26), settings search index and i18n variant checks pass, git diff --check passes, embed staging contains no tests, and make build succeeds.

[bbedward]

/claude review

[claude]

The IPC reveal state can get stuck after fullscreen ends. With global auto-hide off, run bar toggleReveal while fullscreen and barIpcRevealStates[id] becomes true. When fullscreen ends, autoHide goes false, but nothing clears that state. SettingsData.updateBarConfig only clears it when autoHide: false is written to the config. The next time a window goes fullscreen, ipcReveal is still true, so the bar doesn't hide. Outside fullscreen, toggleReveal also returns BAR_AUTO_HIDE_DISABLED, so IPC can't reset it; only hovering the bar can.

Fix: clear the state when the effective auto-hide turns off.

Suggested change

	onAutoHideChanged: evaluateReveal()
	onAutoHideChanged: {
	if (!autoHide)
	SettingsData.setBarIpcReveal(barConfig?.id ?? "", false);
	evaluateReveal();
	}

[steven12138]

Fixed in 6f84847. The Dux 2.0 DBar now clears barIpcRevealStates when effective auto-hide turns off, before evaluating reveal again. Added a regression test for the fullscreen exit/re-entry transition; the full Node and QML suites pass.

[claude]

Claude review

Both earlier findings are fixed (edge hover now gates on topBarCore.autoHide, and the tests moved to quickshell/tests). One new issue:

• The IPC reveal state isn't cleared when fullscreen auto-hide ends, so the next fullscreen window doesn't hide the bar, quickshell/Modules/DankBar/DankBarBody.qml:659

Checked: NiriFullscreen.js matching and missing-data guards, NiriService active-window/snapshot fixes, the fullscreenAutoHide gating (overlay layer, frame, island, overview), exclusive-zone preservation, the toggleReveal IPC guard and dankBarItems lookup, the settings toggle and I18n strings, and the search index. Catalogs are untouched. Model: claude-opus-5-5.

[purian23]

/claude review

[claude]

With fullscreenAutoHide ORed into autoHide, surfaceExclusiveZone (line 805) drops to -1 whenever a fullscreen window shows up. It already does that for topBarCore.autoHide && !reserveExclusiveWhenAutoHidden, and that guard is only true for frame-hosted bars, which this feature excludes. So every fullscreen enter/exit releases and then re-reserves the bar's zone. The other windows on that output relayout both times. That contradicts the PR description and docs/fullscreen-bar.md ("preserves the configured exclusive zone, so other windows do not resize").

Fix: keep the zone while hiding for fullscreen. In the line 805 condition, use (barConfig?.autoHide ?? false) (the effective global auto-hide) instead of topBarCore.autoHide, or add && !barWindow.fullscreenAutoHide.

[steven12138]

Fixed in bd9f7a8. Fullscreen-only hiding now preserves the bar exclusive zone; the prior global auto-hide behavior is unchanged. Added a regression test for both cases. The matching Dux 2.0 package is installed locally, and fullscreen entry/exit works after a plain dms restart. Node: 398 passed; QML and package build passed.

[claude]

Claude review

The port to DBar fixes the earlier findings: IPC reveal is cleared, edge hover uses the effective auto-hide, and the tests are in quickshell/tests. One new issue:

• Fullscreen hiding drops the exclusive zone to -1, so windows on that output resize on every fullscreen enter and exit, against the PR's stated behavior, quickshell/Modules/DBar/DBarBody.qml:917

Checked: the fullscreenAutoHide gating (island, frame, overlay layer, overview), NiriFullscreen.js matching and its binding dependencies, NiriService active-window/focus changes, the toggleReveal IPC guard through dankBarItems, onAutoHideChanged IPC reset, the settings toggle and I18n strings, and the search index. Catalogs are untouched. Model: claude-opus-5-5.

[steven12138]

Superseded by #4000, which carries the Dux 2.0 DBar port and addresses the review findings, including the exclusive-zone regression. Continuing review there; keeping this PR closed for its review history.