Add wheel scroll settings

[klovinad]

Summary

This proposes two related mouse-control improvements:

• Change the default horizontal gesture-button swipes to switch macOS desktops/spaces:
  • Left → Previous Desktop
  • Right → Next Desktop
• Add app-wide scroll wheel preferences:
  • invert wheel direction
  • scroll strength multiplier
  • tactility/chunking

Implementation notes

• Scroll preferences are persisted in AppSettings and mirrored into the hook runtime through a shared ScrollSettings value.
• The hook transforms captured scroll events only when non-default scroll settings are active; default settings pass through natively.
• Synthetic transformed scroll events are posted at the session tap to avoid being re-captured by OpenLogi's own HID event tap.
• Added unit coverage for scroll axis ordering, inversion, and chunking.

Verification

Ran locally on macOS:

cargo fmt --all
cargo check -p openlogi-gui
cargo test -p openlogi-gui -p openlogi-core

Results:

• openlogi-core: 34 passed
• openlogi-gui: 14 passed

[pullfrog]

ℹ️ Minor suggestions inline — one observation about the scroll-settings default check.

Reviewed changes — Adds three scroll-preference controls (invert, strength, tactility) to the Settings window and changes default horizontal gesture swipes to desktop/Space switching.

• Default gesture direction update — PrevTab/NextTab replaced with PreviousDesktop/NextDesktop.
• ScrollSettings shared state — Arc<RwLock<ScrollSettings>> mirrors the persisted config to the hook runtime.
• Scroll event transformation — transform_scroll/quantize_scroll apply inversion, strength, and chunking to captured wheel events.
• Session-tap re-injection — Transformed scroll events are posted at CGEventTapLocation::Session to avoid re-capture by OpenLogi's HID tap.
• Settings UI — Invert switch, strength slider (1–10), and tactility slider (0–10) in a new "Scroll" group box.
• Generalized setting_row — Signature changed from control: Switch to control: impl IntoElement to accept sliders alongside switches.
• Expanded native-click passthrough — Back→BrowserBack and Forward→BrowserForward added.

  ^{｜ Fix all ➔ ｜ Fix 👍s ➔ ｜ View workflow run ｜ Using Big Pickle (free) (credentials for Anthropic not configured) ｜ 𝕏}

[pullfrog]

✅ Prior feedback addressed — no new issues found.

Reviewed changes — Fixed the ScrollSettings::default() mismatch with AppSettings defaults by replacing the derived Default with a manual impl whose strength: 1 matches the app config.

• Manual Default impl for ScrollSettings — Replaced #[derive(Default)] with a manual impl Default where strength: 1 aligns with AppSettings::wheel_strength. This restores the identity fast-path in hook_runtime.rs.

  ^{｜ View workflow run ｜ Using Big Pickle (free) (credentials for Anthropic not configured) ｜ 𝕏}

[averypelle]

This should close #126

[phuocleoceo]

I'm really interested in the "invert wheel direction" feature. Thanks for bringing it up in this Pull Request, I hope it gets merged into the software soon!

[shawnflanagan]

Let's get these conflicts resolved so we can merge this. Invert wheel direction is sorely needed in the app

[averypelle]

for anyone looking for an interim fix, https://pilotmoon.com/scrollreverser/ is compatible with this app

[greptile-apps]

Greptile Summary

This PR adds user-configurable wheel scroll settings (inversion, strength multiplier, tactility/chunking) across all three platforms, routing discrete mouse wheel events through a ScrollSettings transform while bypassing the transform for continuous trackpad events via the new is_continuous field.

• hook_runtime.rs / binding.rs: ScrollSettings and SharedScrollSettings thread-safely share config with the hook callback; macOS mutates the CGEvent in-place (TransformScroll) while Windows/Linux suppress the original and re-inject a synthetic event — injection loops are correctly prevented by LLMHF_INJECTED on Windows and separate uinput device nodes on Linux.
• macos.rs: transform_scroll_event handles all three CGEvent scroll field families (double, point, fixed-point); transform_integer_scroll_field always receives tactility=0 from its call sites, leaving the quantize branch dead (P2).
• config.rs / state.rs / settings.rs: New fields use #[serde(default)] for backwards compatibility; GUI sliders clamp strength to 1–10 and tactility to 0–10, though the bounds are duplicated between settings.rs constants and state.rs magic numbers rather than sourced from a single definition (P2).

Confidence Score: 4/5

Safe to merge; all findings are P2 quality/polish items with no correctness or data-loss risk.

No P0 or P1 issues found. The three P2 comments cover dead code in transform_integer_scroll_field, duplicated bound constants, and a theoretical async race in post_symbolic_hotkey. Injection-loop guards on Windows (LLMHF_INJECTED) and macOS (SYNTHETIC_EVENT_USER_DATA) are both confirmed in place. Score is 4 rather than 5 to acknowledge the minor dead-code and duplication issues worth tidying before release.

crates/openlogi-core/src/binding.rs (symbolic hotkey async race), crates/openlogi-hook/src/macos.rs (dead tactility branch), crates/openlogi-gui/src/windows/settings.rs + crates/openlogi-gui/src/state.rs (duplicated wheel bound constants)

Important Files Changed

Filename	Overview
crates/openlogi-agent-core/src/hook_runtime.rs	Core scroll transform logic added: ScrollSettings struct, SharedScrollSettings Arc, transform_scroll/quantize_scroll helpers, and per-platform dispatch (TransformScroll on macOS, Suppress+post on others).
crates/openlogi-agent-core/src/orchestrator.rs	Wires SharedScrollSettings into SharedRuntime; initialises from config and updates via apply_config. Straightforward plumbing with no logic issues.
crates/openlogi-agent/src/main.rs	One-line change to forward scroll_settings clone to hook_runtime::start(). No issues.
crates/openlogi-core/src/binding.rs	Large addition moving Action::execute() and platform scroll-injection helpers into openlogi-core. post_symbolic_hotkey has a potential async race (P2). Injection-loop risk on macOS is correctly prevented by the SYNTHETIC_EVENT_USER_DATA tag check.
crates/openlogi-core/src/config.rs	Adds wheel_inverted, wheel_strength (default 1), and wheel_tactility (default 0) to AppSettings with proper serde defaults and Default impl. No issues.
crates/openlogi-gui/src/state.rs	Adds set_wheel_inverted/strength/tactility mutators with in-state clamping before persist_and_reload. Clamp bounds are magic numbers duplicated from settings.rs (P2).
crates/openlogi-gui/src/windows/settings.rs	New wheel inversion switch and two sliders added to the settings UI. Wheel bound constants redeclared locally rather than sourced from config.rs (P2 duplication).
crates/openlogi-hook/src/lib.rs	MouseEvent::Scroll gains is_continuous: bool; new EventDisposition::TransformScroll(ScrollTransform) variant added. Additive and non-breaking.
crates/openlogi-hook/src/linux.rs	scroll() sets is_continuous: false; device_thread disposition broadened to !matches!(Suppress) so TransformScroll passes through correctly on Linux. Tests updated.
crates/openlogi-hook/src/macos.rs	transform_scroll_event modifies CGEvent fields in-place. transform_integer_scroll_field always receives tactility=0, leaving the quantize branch dead code (P2).
crates/openlogi-hook/src/tests.rs	Scroll event literal updated to include is_continuous: false. Trivial maintenance change.
crates/openlogi-hook/src/windows.rs	Both scroll constructors gain is_continuous: false. LLMHF_INJECTED guard confirmed, preventing synthetic scroll re-capture.

Sequence Diagram

%%{init: {'theme': 'neutral'}}%%
sequenceDiagram
    participant HW as Physical Mouse
    participant OS as OS Hook (WH_MOUSE_LL / CGEventTap / evdev)
    participant CB as hook callback (mouse_proc / tap_callback / device_thread)
    participant HR as hook_runtime (EventDisposition)
    participant macOS as macos::transform_scroll_event
    participant Post as binding::post_scroll_delta

    HW->>OS: Raw scroll event
    OS->>CB: Invoke hook callback
    CB->>CB: "Translate to MouseEvent::Scroll{delta_x, delta_y, is_continuous}"
    CB->>HR: invoke user callback(event)
    
    alt is_continuous (trackpad) OR default settings
        HR-->>CB: EventDisposition::PassThrough
        CB->>OS: CallNextHook / Keep / forward
    else macOS discrete wheel
        HR-->>CB: "EventDisposition::TransformScroll(ScrollTransform{inverted, strength, tactility})"
        CB->>macOS: transform_scroll_event(cg_event, transform)
        macOS->>macOS: modify DELTA_AXIS fields in-place
        macOS-->>CB: modified CGEvent
        CB->>OS: CallbackResult::Keep (OS delivers modified event)
    else non-macOS (Windows / Linux)
        HR->>HR: transform_scroll(delta_x, delta_y, settings) → (i32, i32)
        HR-->>CB: EventDisposition::Suppress
        CB->>OS: Block original event
        HR->>Post: post_scroll_delta(v, h)
        Post->>OS: Inject new synthetic scroll (LLMHF_INJECTED / uinput write)
        OS->>CB: Re-deliver synthetic event
        CB->>CB: LLMHF_INJECTED flag / different device node → return None
        CB->>OS: PassThrough (synthetic event flows to apps)
    end

Loading

%%{init: {'theme': 'base', 'themeVariables': {"darkMode": true, "background": "#0d1117", "primaryColor": "#21262d", "primaryTextColor": "#e6edf3", "primaryBorderColor": "#8b949e", "lineColor": "#8b949e", "textColor": "#e6edf3", "edgeLabelBackground": "#161b22", "actorBkg": "#21262d", "actorBorder": "#8b949e", "actorTextColor": "#e6edf3", "actorLineColor": "#8b949e", "signalColor": "#8b949e", "signalTextColor": "#e6edf3", "noteBkgColor": "#373320", "noteBorderColor": "#d4a72c", "noteTextColor": "#f0e6c0", "labelBoxBkgColor": "#21262d", "labelBoxBorderColor": "#8b949e", "labelTextColor": "#e6edf3", "loopTextColor": "#e6edf3", "activationBkgColor": "#30363d", "activationBorderColor": "#8b949e"}}}%%
sequenceDiagram
    participant HW as Physical Mouse
    participant OS as OS Hook (WH_MOUSE_LL / CGEventTap / evdev)
    participant CB as hook callback (mouse_proc / tap_callback / device_thread)
    participant HR as hook_runtime (EventDisposition)
    participant macOS as macos::transform_scroll_event
    participant Post as binding::post_scroll_delta

    HW->>OS: Raw scroll event
    OS->>CB: Invoke hook callback
    CB->>CB: "Translate to MouseEvent::Scroll{delta_x, delta_y, is_continuous}"
    CB->>HR: invoke user callback(event)
    
    alt is_continuous (trackpad) OR default settings
        HR-->>CB: EventDisposition::PassThrough
        CB->>OS: CallNextHook / Keep / forward
    else macOS discrete wheel
        HR-->>CB: "EventDisposition::TransformScroll(ScrollTransform{inverted, strength, tactility})"
        CB->>macOS: transform_scroll_event(cg_event, transform)
        macOS->>macOS: modify DELTA_AXIS fields in-place
        macOS-->>CB: modified CGEvent
        CB->>OS: CallbackResult::Keep (OS delivers modified event)
    else non-macOS (Windows / Linux)
        HR->>HR: transform_scroll(delta_x, delta_y, settings) → (i32, i32)
        HR-->>CB: EventDisposition::Suppress
        CB->>OS: Block original event
        HR->>Post: post_scroll_delta(v, h)
        Post->>OS: Inject new synthetic scroll (LLMHF_INJECTED / uinput write)
        OS->>CB: Re-deliver synthetic event
        CB->>CB: LLMHF_INJECTED flag / different device node → return None
        CB->>OS: PassThrough (synthetic event flows to apps)
    end

Loading

_{Reviews (9): Last reviewed commit: "Merge branch 'master' into proposal/whee..." | Re-trigger Greptile}

[AprilNEA]

Thanks for this — invert-scroll is clearly in demand (#126 plus the comments here). Before merging I want to flag some concerns with the current approach, because the suppress-and-reinject design has a few real problems on macOS, and one of them is that it doesn't actually solve #126 as written.

1. It doesn't solve #126 (per-device inversion)

#126 specifically asks to keep the trackpad on natural scrolling and invert only the mouse. The hook taps at CGEventTapLocation::HID (crates/openlogi-hook/src/macos.rs:257) and captures every ScrollWheel event regardless of source, and transform_scroll never looks at kCGScrollWheelEventIsContinuous. So enabling invert flips the trackpad too — exactly the native-setting behavior #126 wants to avoid. We shouldn't close #126 with this as-is.

2. Modifier flags are dropped

post_scroll_delta builds a fresh CGEvent::new_scroll_event(...) and never copies the original event's flags. Modifier+scroll gestures that apps read off the event flags (zoom, etc.) will stop working while non-default settings are active. (System-level Ctrl+scroll screen zoom reads live key state and may be unaffected — worth verifying on-device.)

3. Momentum / pixel precision are lost

We suppress the original and re-emit ScrollEventUnit::LINE from the rounded AXIS_1/AXIS_2 line deltas. That drops scroll-phase/momentum and ignores the pixel POINT_DELTA_* fields, so continuous input (trackpad, MX free-spin high-res wheel) becomes coarse. Micro-deltas round to 0 → PassThrough, so on a free-spinning wheel small movements aren't inverted while fast ones are — inconsistent. Given our primary devices are MX-class, this matters.

4. Per-event cost on the hottest path

The macOS tap must stay lock-light or it stalls the whole input stream (see the comments in macos.rs). This adds an RwLock read per scroll event and, when active, a fresh CGEventSource::new + CGEvent allocation per event. On a high-rate free-spin stream that risks TapDisabledByTimeout / scroll stutter.

Suggested approach

Instead of suppress + reinject, mutate the event in place and return Keep. We already do in-place field writes elsewhere (crates/openlogi-core/src/binding.rs:1312,1327), and the same set_*_value_field works on the callback's &CGEvent. Negate/scale AXIS_1/AXIS_2 plus the POINT_DELTA_* and fixed-point fields and keep the event — that preserves momentum, pixel precision, flags, and continuity, with zero allocation and no re-entry concern. For #126, branch on kCGScrollWheelEventIsContinuous to apply inversion to discrete (mouse) scroll only. This is essentially how Scroll Reverser does it.

Unrelated change

The default gesture remap (PrevTab/NextTab → PreviousDesktop/NextDesktop) is a separate UX decision — please split it into its own PR so each can be reviewed/reverted independently.

The config plumbing, clamping, and default fast-path are clean; the concern is purely the macOS transform path. Happy to help iterate on the in-place version.