feat: Switch-layout button remap for PDP wired Switch pads (Standard quirk + Direct button order)

[emir-hasanbegovic]

Problem

A PDP Faceoff Wired Pro Controller for Nintendo Switch (USB 0e6f:0180, hardware in hand) streams with its A button dead in Standard mode. Tracing the pad (live USB descriptor read plus the Android input chain) showed the whole button row is shifted; A is just the loudest casualty.

These pads declare 14 buttons in the Switch usage order: Y B A X L R ZL ZR Minus Plus L3 R3 Home Capture.

Standard: the kernel maps button usages sequentially from BTN_SOUTH, and Generic.kl turns that into BUTTON_A..BUTTON_THUMBL. Physical A therefore arrives as KEYCODE_BUTTON_C and physical R as KEYCODE_BUTTON_Z; keycodeToXusb() had no case for either, so both key gates dropped the events. Everything else landed one label off: Y acted as A, L as Y, ZL/ZR as bumpers, Minus/Plus as full-press triggers, stick clicks as Back/Start, Capture as L3, and Home vanished.

Direct: the pad classifies as GENERIC_HID_GAMEPAD and its descriptor parses fine, but buttonBit() assumes the western A B X Y order, which scrambles the pad differently (A acted as X, ZL/ZR as Back/Start, R3/Home/Capture dropped). So Direct was not a workaround either.

Fix

One Switch-order layout applied on both paths, positional to match decodeSwitchProUsb: south to A, east to B, west to X, north to Y, L/R to the bumpers, ZL/ZR to the triggers, Minus/Plus to Back/Start, L3/R3 to the stick clicks, Home to Guide. Capture stays unmapped because XUSB has no equivalent bit.

Standard path

• New QUIRK_SWITCH_LAYOUT bit (0x04) in gamepad_input.h and GamepadQuirks.kt; resolveGamepadQuirk() now takes VID and PID and the registry pushes both.
• applyKey() branches on the quirk into switchLayoutKeycodeToXusb(); ZL/ZR (arriving as BUTTON_L1/R1) drive the triggers through the existing ltFromKey/rtFromKey latch so a zero axis sample cannot clear a held trigger.
• Both key gates (gamepadKeyFilter and processGamepadKeyEvent) now look the device quirk up under g_devicesMtx before deciding whether a keycode is consumable, so BUTTON_C/BUTTON_Z/BUTTON_MODE are consumed only for Switch-layout devices. Non-quirked devices keep the exact previous gate behavior, including the pre-IME gate not special-casing BUTTON_7/8.

Direct path

• KnownDevice and Classification carry a ButtonOrder (default WESTERN).
• The attach path stamps HidLayout.switchOrderButtons after fetchHidLayout (the parse resets the struct, pinned by a test).
• decodeFromLayout() uses switchOrderButtonBit() when flagged, with button indices 6/7 folding into bLT/bRT.

Model catalog

Five PDP models join kImported with Parser::GENERIC_HID_GAMEPAD and ButtonOrder::SWITCH: 0180 Faceoff Wired Pro, 0181 Faceoff Deluxe, 0184 Faceoff Deluxe+ Audio, 0185 Wired Fight Pad Pro, 0187 Rock Candy. They get card names and stay off the verified fast lane, so nothing auto-claims into Direct.

0e6f:0186 (Afterglow Wireless) is excluded on purpose: SDL classifies it as a Switch Pro protocol controller whose USB port is charge-only, so the wired Switch-order remap would be wrong for it. Source: SDL controller_list.h.

The model set intentionally lives in two places (Kotlin quirk table for the framework path, kImported for the direct path); each side pins its five PIDs in tests, and the exclusion of 0186 is tested on both sides too.

Tests

Current-state pins (documenting the bug and the assumptions the fix rests on):

• keycodeToXusb drops BUTTON_C/BUTTON_Z/BUTTON_MODE; applyKey without the quirk ignores them.
• A byte-accurate Faceoff descriptor fixture parses to the expected field map (buttons at bit 0, count 14, hat at bit 16, X/Y/Z/Rz bytes), and western decode of that shape reproduces the scramble (A lands on X, ZL on Back, upper row vanishes).
• Classification fallbacks and PDP's Xbox pads stay WESTERN; PDP Switch pads are not verified fast lane.

New behavior:

• Full truth tables for switchLayoutKeycodeToXusb and switchOrderButtonBit (face row, bumpers, Minus/Plus, stick clicks, Home, dpad, Capture unmapped).
• Trigger latching: ZL/ZR set and clear bLT/bRT on both paths, survive zero axis samples while held, and Minus/Plus do not touch the triggers.
• Gate predicate coverage (switchLayoutConsumesKey), quirk-bit cross-language pin (0x04 asserted in both gtest and JUnit), stray swap bits do not double-remap.
• End-to-end decodeReport with a Switch-order layout through ParserState.
• Kotlin: the five PIDs resolve to QUIRK_SWITCH_LAYOUT, 0186 and PDP Xbox PIDs resolve to none, Nintendo vendor behavior unchanged.

Verification

• Native host tests: 288/288 pass (:app:nativeTest).
• :app:ktlintCheck, :app:detekt, GamepadQuirksTest green; full unit-test source set compiles.
• clang-format 22.1.4 (CI pin) clean over app/src/main/cpp.
• :app:externalNativeBuildDebug green on arm64-v8a and x86_64.
• Hardware: the 0180 unit is on hand. Draft until the on-device pass is done: Standard (A/B/X/Y positional, ZL/ZR as triggers, Minus/Plus as Back/Start, Home as Guide) and Direct (same table through the descriptor path), plus a regression check on an Xbox pad and a Switch Pro.

Notes for review

• Positional mapping (not label) was chosen to match the repo's existing Nintendo convention (resolveGamepadQuirk comment, decodeSwitchProUsb). If label mapping is preferred for these pads, it is a two-table change.
• The key gates now take g_devicesMtx slightly earlier (to read the quirk) and hold it across the same work as before; no new lock ordering.
• Capture is consumed nowhere and mapped nowhere; if the wire ever grows a Share bit it slots into both tables.