shard_accumulator only keeps a weak_ptr to the stream, but setup() passed
it a temporary from shared_from_this(). setup() runs on the network
thread, so if the main thread drops its reference meanwhile (e.g. while
shutting down after XR_ERROR_INSTANCE_LOST), that temporary is the last
one and ~stream runs on the network thread. It then joins
network_thread from itself: std::system_error (EDEADLK) and
std::terminate.
Pass weak_from_this() instead so no strong reference is created.
Mirror the gamepad forwarded from the headset to a uinput device with the standard Xbox layout, so applications that do not read it through OpenXR, such as SDL and Steam Input, pick it up. This extends the existing hid-forwarding uinput path, which already covered keyboard and mouse.
The headset chooses which devices to forward (keyboard, mouse, gamepad), off by default. The hid-forwarding server option gates the uinput mirror only; the OpenXR gamepad is always exposed. The mirror needs the uinput module and write access to /dev/uinput.
Gamepad rumble written to the uinput device is read back and forwarded to the headset as OpenXR haptics.
Forward a gamepad paired to the headset to the server. Useful when the controller is in range of the headset but not the server, such as a DualSense paired to a Quest over Bluetooth.
The headset reads the gamepad through the OpenXR microsoft/xbox_controller profile and forwards it on the existing input rails: gamepad device ids in the inputs packet, plus a gamepad entry in the interaction profile array that signals connection and disconnection from the tracking thread.
The server exposes a native OpenXR gamepad at /user/gamepad, connected while the headset reports a gamepad interaction profile. This needs no permission and is always available.
Rumble is sent back to the headset and applied through OpenXR haptics.
From pure trial and error, is seems that swapchain
management on the quest platform is broken. After a
swapchain is destroyed, there is a _chance_ the
next created swapchain will be unusable.
This seems to happen no matter what, where or when
the client decides to clean up its resources and the
safest bet is to not recreate swapchains much.