fix(linux): send scroll to the mouse device that last moved

[GabeDuarteM]

Description

In Moonlight's "Optimize mouse for remote desktop" mode, scrolling on a Linux host works while the pointer is still, but breaks once it's moving.

The mouse is two uinput devices. Absolute motion goes to the absolute one, but scroll always went to the relative one, and the absolute device had no wheel axes. Buttons already follow whichever device moved last, so this does the same for scroll:

 configure_evdev_mouse(kind):
   enable buttons
+  enable EV_REL (both devices)
   if absolute:
     enable INPUT_PROP_DIRECT, ABS_X, ABS_Y
-    return
-  enable REL_X, REL_Y
+  else:
+    enable REL_X, REL_Y
   enable REL_WHEEL, REL_HWHEEL (+ hi-res codes)

 submit_scroll():
-  emit on the relative device
+  emit on last_motion_device_

The absolute device still has no REL_X or REL_Y, so libinput keeps treating it as an absolute pointer.

inputtino fixed the same bug this way in games-on-whales/inputtino#38, which Sunshine picked up in LizardByte/Sunshine#5135. It came back when Sunshine moved to libvirtualhid in LizardByte/Sunshine#5368, which I noticed after my last Sunshine update.

Screenshot

Moonlight in remote desktop mode, spinning the wheel non-stop while moving the pointer. The wheel counter stalls on the stock Copr build and keeps counting with the fix:

scroll-fix-before-after.mp4

Issues Fixed or Closed

• Related to Mouse scrolling is unreliable in desktop mode moonlight-stream/moonlight-qt#1020 (the cause is traced in Mouse scrolling is unreliable in desktop mode moonlight-stream/moonlight-qt#1020 (comment))

Roadmap Issues

Testing

New and updated tests:

• LinuxBackendTest.PipeBackedUinputMouseScrollsOnLastMotionDevice sends absolute motion, both scroll directions, relative motion and another scroll, then checks which device each wheel event came out of.
• The absolute mouse capability checks now expect the wheel axes, and still no REL_X or REL_Y.
• LinuxConsumerTest.LibinputSeesSplitUinputMouseMotionAndButtons now scrolls after the absolute moves and expects libinput to report LIBINPUT_EVENT_POINTER_SCROLL_WHEEL (v120 of -120) from the absolute device.

I ran the full suite with no filter in a Fedora 43 VM (kernel 6.17.1-300.fc43, real /dev/uinput and /dev/uhid), on top of 1e41c36:

• Before (master plus the new tests): 159 passed. The three mouse tests above fail, and the libinput test times out waiting for a scroll event from the absolute device.
• After (with the fix): 162 passed. The only failures are SdlSeesXboxOneCanonicalButtons and SdlSeesXboxSeriesCanonicalButtons (SDL sees 11 buttons instead of 15), and they fail the same way on unmodified master in that VM, so I don't think they're related.

The tests that don't need device nodes also pass when built with gcc 14 in a fedora:43 container, and git clang-format reports no changes.

Manual: I built the Sunshine Copr RPM for 2026.914.233613 with this fix applied to its pinned libvirtualhid (53e1a94, same fix lines) and installed it on Fedora 44. Scrolling while moving the mouse in Moonlight's remote desktop mode works now.

FreeBSD: uhid_backend.cpp builds there too, so I also tested in a FreeBSD 15.1 VM (base clang 19.1.7, uinput loaded). The FreeBSD suite passes on both master and this branch (111 passed, 2 Linux-only tests skipped). I also ran a small throwaway probe (not part of this PR) that creates a mouse, moves it absolutely, scrolls, and reads the absolute /dev/input/event* node with libevdev. On master that node has no wheel axes and gets no scroll events. With this change it gets REL_WHEEL 1 and REL_WHEEL_HI_RES 120. I didn't test libinput or a desktop on FreeBSD.

Type of Change

• feat: New feature (non-breaking change which adds functionality)
• fix: Bug fix (non-breaking change which fixes an issue)
• docs: Documentation only changes
• style: Changes that do not affect the meaning of the code (white-space, formatting, missing semicolons, etc.)
• refactor: Code change that neither fixes a bug nor adds a feature
• perf: Code change that improves performance
• test: Adding missing tests or correcting existing tests
• build: Changes that affect the build system or external dependencies
• ci: Changes to CI configuration files and scripts
• chore: Other changes that don't modify src or test files
• revert: Reverts a previous commit
• BREAKING CHANGE: Introduces a breaking change (can be combined with any type above)

Checklist

• Code follows the style guidelines of this project
• Code has been self-reviewed
• Code has been commented, particularly in hard-to-understand areas
• Code docstring/documentation-blocks for new or existing methods/components have been added or updated
• Tests have been added or updated for new or modified functionality, including regression tests for bug fixes

Let me know if theres any comment needed in here, its a small fix so I thought it unnecessary

AI Usage

See our AI usage policy.

• AI tools were used while preparing this PR

Coded with the help of Opus 5.5

[CLAassistant]

All committers have signed the CLA.

[ReenigneArcher]

Reviewed b316f9f366ffae94f1d1d87e8f2df65afb34d7f0 against current master (1e41c36). I found no functional defect in the scroll-routing change; one required backend Doxygen update remains, detailed inline.

Validation: the GoogleTest target built with GCC 15.3.1 in a Fedora 43 container, and cmake-build-review/tests/test_libvirtualhid --gtest_filter='LinuxBackendTest.*' passed all 37 tests, including the new routing test and updated capability assertions. git diff --check 1e41c36...HEAD passed. Real-device libinput validation could not run locally because /dev/uinput and /dev/uhid were unavailable.

Merge/check status: GitHub reports no merge conflicts. SonarCloud and the primary documentation build passed. Actions await maintainer approval; the GH-Pages documentation log timed out waiting for check runs. That workflow limitation is separate from the documentation finding.

_{Review completed by Codex | Model: gpt-6.1-sol | Thinking: xhigh}

[sonarqubecloud]

Quality Gate passed

Issues
0 New issues
0 Accepted issues

Measures
0 Security Hotspots
0.0% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarQube Cloud

[codecov]

Codecov Report

❌ Patch coverage is 36.36364% with 7 lines in your changes missing coverage. Please review.
✅ Project coverage is 73.73%. Comparing base (1e41c36) to head (6212871).
⚠️ Report is 1 commits behind head on master.
✅ All tests successful. No failed tests found.

Files with missing lines	Patch %	Lines
src/platform/linux/uhid_backend.cpp	36.36%	2 Missing and 5 partials ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##           master     #195      +/-   ##
==========================================
- Coverage   73.77%   73.73%   -0.04%     
==========================================
  Files          43       43              
  Lines       10702    10704       +2     
  Branches     3695     3696       +1     
==========================================
- Hits         7895     7893       -2     
- Misses       2196     2198       +2     
- Partials      611      613       +2     

Flag	Coverage Δ
FreeBSD	53.84% <18.18%> (-0.01%)	⬇️
Linux-Clang	68.40% <27.27%> (-0.01%)	⬇️
Linux-GCC	80.90% <30.00%> (-0.08%)	⬇️
PowerShell	75.60% <ø> (ø)
Windows-MSVC	74.17% <ø> (+0.01%)	⬆️
Windows-MinGW-UCRT64	66.09% <ø> (ø)
macOS	50.96% <ø> (ø)

Flags with carried forward coverage won't be shown. Click here to find out more.

Files with missing lines	Coverage Δ
src/platform/linux/uhid_backend.cpp	81.38% <36.36%> (-0.19%)	⬇️

---

Continue to review full report in Codecov by Harness.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 1e41c36...6212871. Read the comment docs.