[enaboapps]

Closes #1222

What

Switch interfaces in gamepad mode and game controllers now work as switches. Buttons 1 to 8 on any connected controller are switch keys named Gamepad1 to Gamepad8, shown as Gamepad 1 to Gamepad 8, learned from the Switches page like keys and usable for every press and hold action.

• Windows: the capture worker registers the HID joystick and gamepad usages with Raw Input (RIDEV_INPUTSINK), so presses arrive while another app is in front, and reads button state with HidP_GetUsages. Button edges ride the existing edge queue at virtual-key codes Windows leaves unassigned (0x97 to 0x9E); the keyboard hook ignores that range so a stray keyboard code can never pose as a button. Controllers are recorded on arrival and at startup so unplugging one is never mistaken for a keyboard removal. Buttons are never suppressed, a held button blocks re-enabling until released, a removed controller releases its buttons, and controller registration is best effort so keyboard capture never depends on it.
• macOS: an IOHIDManager scheduled on the existing capture run loop, matching the same usages and only button elements. GameController's background flag is not used (it broke silently in macOS 15.4). A button held while capture was off blocks re-enabling, like a key.
• Settings: the switch settings file stays schema 1 (the key is a string). The synced settings schema moves from 5 to 6 with a no-op upgrade. Consequence to note in the release notes: once any computer on 1.3.0 saves, a Switchify PC 1.2.x on the same account shows "update to keep syncing" instead of silently rejecting a profile with a gamepad key.
• Docs and help: Add a switch, Troubleshooting, the in-app "Switch not recognized" advice and the interface demonstration now say gamepad mode is fine and mouse-click mode is the one that cannot be learned.

Out of scope, as in the issue: sticks, triggers, the D-pad, Xbox layout names, buttons 9 and up, and telling two controllers apart.

Validation

• npm run prediction-model, npm run lint, npm test (312 passed), npm run build
• cargo fmt --check, cargo clippy --all-targets -D warnings on Windows and cross-checked for aarch64-apple-darwin, cargo test (849 passed)
• site: npm run build
• New tests: key naming and display names, per-device button merging, Raw Input edge handling in learning and active modes, keyboard removal while a button is held, the keyboard hook rejecting the controller code range, a held button blocking startup, switch settings validation, and the version 5 to 6 sync upgrade with gamepad keys.
• Two independent review rounds; every actionable finding is fixed in the second and third commits.

Not yet done, needs hardware: a controller or switch box in gamepad mode on Windows and on a Mac, learned while another app is in front. The Xbox Adaptive Controller should be checked over USB and Bluetooth since its two transports expose different descriptors, and a composite box such as the Pretorian USB Switch should be checked on a Mac in case Input Monitoring is required. Known limits accepted for now: a button held across a worker restart on Windows is not known until its next report, and a controller that splits its buttons across several HID report IDs would need per-report tracking.

🤖 Generated with Claude Code

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated
site	Ready	Preview	Oct 28, 2026 9:37am UTC