Reading HID Joysticks (hid.dll; formerly winmm)

7DTD uses Unity's legacy Input Manager, so Input.GetAxis only sees axes baked into the shipped InputManager.asset — a mod cannot add axes. To read a DirectInput/HID joystick (e.g. a RadioMaster transmitter in "USB Joystick (HID)" mode), bypass Unity and read the device directly. Working implementation: FPV/src/FPVGamepad.cs (direct hid.dll, all 8 axes).

Current approach: setupapi + hid.dll + overlapped ReadFile

Parse HID input reports yourself; winmm is a 6-axis dead end (see below). Recipe, all plain P/Invoke with no extra DLLs shipped:

1. Enumerate: HidD_GetHidGuid → SetupDiGetClassDevs(DIGCF_PRESENT | DIGCF_DEVICEINTERFACE) → SetupDiEnumDeviceInterfaces → SetupDiGetDeviceInterfaceDetail for the device path. The detail struct's cbSize is the fixed part only: 8 on x64 (the game is 64-bit only).
2. Filter: CreateFile(GENERIC_READ, share RW, FILE_FLAG_OVERLAPPED) — keyboards/mice refuse GENERIC_READ, which filters them for free. Then HidD_GetPreparsedData + HidP_GetCaps; keep devices with UsagePage 0x01 and Usage 0x04 (joystick) / 0x05 (gamepad) / 0x08 (multi-axis). If several qualify, prefer the one with the most axes.
3. Friendly name: HidD_GetProductString returns the same name the Game Controllers dialog shows — this replaces the winmm-era registry OEMName crawl entirely.
4. Axes: HidP_GetValueCaps (mind IsRange) → the 8 analog usages are 0x30..0x37 = X, Y, Z, Rx, Ry, Rz, Slider, Dial. Record LogicalMin/Max/BitSize per usage; if LogicalMin < 0 the raw value from HidP_GetUsageValue is two's-complement in BitSize bits — sign-extend before normalizing.
5. Read loop (no background thread needed): keep one overlapped ReadFile pending; each frame drain completed reports via GetOverlappedResult(wait:false) until ERROR_IO_INCOMPLETE, re-issuing the read after each report. Allocate the OVERLAPPED and report buffer in unmanaged memory (Marshal.AllocHGlobal); on x64 hEvent sits at OVERLAPPED offset 24. Any other read error = device unplugged.
6. Parse: HidP_GetUsageValue per axis usage and HidP_GetUsages (usage page 0x09; button usages are 1-based) per report. HIDP_STATUS_SUCCESS is 0x00110000; a non-success status usually means the usage lives in a different report ID — keep the previous value, don't zero it. Report buffer length passed to HidP_* is Caps.InputReportByteLength (includes the report-ID byte).
7. Teardown: if a read is pending, CancelIo + GetOverlappedResult(wait: true) before freeing the buffers, or the driver can write into freed memory.

Channel order still varies per radio/firmware, so bind axes by detection ("move the stick"), not a fixed mapping. Slots the device doesn't report should read as centered (0.5) so a stale bind can't peg a channel.

Why not winmm (joyGetPosEx): it silently drops 2 of 8 axes

winmm (joyGetNumDevs / joyGetPosEx / joyGetDevCapsW) has only six axis slots and the HID→slot mapping is not what the field names suggest. Measured with EdgeTX USB joystick "classic" mode (CH1-8 → HID X, Y, Z, Rx, Ry, Rz, Slider, Dial):

EdgeTX channel	HID axis	JOYINFOEX field
CH1	X	dwXpos
CH2	Y	dwYpos
CH3	Z	dwZpos
CH4	Rx	dwVpos (not dwRpos!)
CH5	Ry ("Y Rotation")	not visible
CH6	Rz	dwRpos
CH7	Slider	dwUpos
CH8	Dial	not visible
CH9-32	buttons	dwButtons

So a switch mixed onto CH5 (or CH8) moves in the Windows Game Controllers dialog (DirectInput sees all axes) but is invisible to winmm — no movement in any polled axis, no bind capture. This bit the FPV mod for real; the direct-HID rewrite above was the fix. Radio-side workarounds (if stuck with winmm): move the mix to CH6/CH7, or to CH9+ where it reports as a plain HID button; EdgeTX's USB joystick "Advanced" mode can also set a channel's mode to Btn explicitly. Source: https://manual.edgetx.org/edgetx-how-to/joystick-mapping-information-for-game-developers

Also, JOYCAPS.szPname is the driver string ("Microsoft PC-joystick driver"), not the product name; getting the real name under winmm requires a registry crawl through MediaResources\Joystick\<szRegKey> → MediaProperties\...\Joystick\OEM\<VID_xxxx&PID_xxxx>\OEMName (HKCU first, then HKLM). With hid.dll, HidD_GetProductString makes all of that obsolete.

Gotcha: transmitter switches can report as axes, not buttons

Depending on the radio's channel/mixer setup, a 2/3-position aux switch may be mapped onto an analog axis instead of a button bit — a press-to-bind button capture will never see it. Treat "axis as switch": at bind time record the axis index plus its rest position (normalized 0..1 value when capture started), then consider the action held while |value - rest| > 0.25. The rest-relative delta self-calibrates regardless of which way the switch throws or whether it idles at 0, 0.5, or 1. Exclude axes already bound to flight channels from this capture so brushing a gimbal doesn't steal the bind. Implementation: TryGetCapturedActionAxis in FPV/src/FPVControls.cs.

Gotcha: throttle channel direction varies per radio — a reversed one reads "stuck"

A transmitter's throttle channel can be reversed (radio-side channel reverse, mixer weight -100, or just how the model was set up), so stick-bottom reports raw 1.0 and stick-top 0.0. A non-centering throttle read as raw 0..1 then fails in a deceptively quiet way that presents as "the craft won't take off" (note: a real FPV-mod stuck-on-ground bug turned out to be resting-contact sweep freeze instead — see Unity Sweep Casts & Resting Contact — so confirm with the live axis readout before blaming the radio):

• Stick at idle reads high → any "drop throttle to idle" arming safety never clears (or clears only at full stick).
• Once armed, pushing the stick up commands ~0% thrust, so the craft never lifts; the motor audio still responds to stick movement (mirrored), which makes the input path look healthy.

Diagnosis trick: if motor audio pitch is driven by commanded throttle (FPV's loop sweeps pitch 0.52→1.85 over throttle 0→1), the periodic audio dbg log lines are a readout of what the flight model actually received — pitch pinned near the bottom of the sweep while the pilot holds full stick proves the command is ~0 without any extra instrumentation. Fix: per-binding Invert flag (controls.json → Throttle.Invert, or Options → Controls), or reverse the channel on the radio — not both. The mod now also logs the live stick % while the throttle safety holds, and prints it in the "Disarmed!" HUD prompt.