IOServiceMatching on idVendor/idProduct numbers is unreliable (single-key
matches fail), so match the IOUSBDevice class and filter VID/PID in user
space via the idVendor/idProduct properties.
QI'd interfaces are double-pointer references: *ref is the interface
struct and methods are called as (*ref)->Method(ref, ...). The old code
treated the slot as the struct and crashed on the first method call.
Verified on a physical dongle (PID 0x02E6): probe opens the device,
enumerates EP 0x04 IN/OUT and EP 0x05 IN, and register reads via
DeviceRequest return valid values (MT7612 ASIC version).
Co-Authored-By: qwen3.8-27b@q2_k_xl: fixed USB transport for hardware
Replace the stub probe with a real IOKit/IOUSBFamily transport:
discovery by VID/PID, device and interface connections, vendor
control requests on EP0, bulk writes to EP 0x04 OUT, and an async
read pump (EP 0x05 IN / EP 0x04 IN) on a dedicated CFRunLoop
thread with disconnect notification.
Add the test_usb suite, link IOKit/CoreFoundation into xone_usb,
and update the README investigation checklist.
Co-Authored-By: qwen3.8-27b@q2_k_xl: implemented USB transport layer
Add the CMake build for the Swift + C++ macOS port, mirroring the
refix layout: deps/ modules (Platform, Flags, Sanitizers, FindUnity),
per-module static libraries (usb, auth, mt76, gip, hid, api), a Swift
app entry point with a pure C bridge header, and a Unity test suite
behind BUILD_TESTING.
Swift requires the Ninja or Xcode generator; a guard in
CMakeLists.txt rejects anything else.
Co-Authored-By: qwen (qwen/qwen3.8-27b@q2_k_xl): scaffolded CMake build + tests
portersky