Blacklist XR_EXT_hand_interaction for HTC so the runtime can't switch
/user/hand/* to ext/hand_interaction_ext mid-session, which drops
controller button input whenever that profile is briefly active.
Point the Focus3 squeeze/touch binding at XRT_INPUT_FRAME_SQUEEZE_TOUCH,
the name the device actually exposes for that slot, instead of a name
that's never set active.
interpolate(before, after, t) expects t = 0 at the before sample and
t = 1 at the after sample (see hand_joints_list.cpp: radius = a.radius
* (1 - t) + b.radius * t, and monado's m_space_relation_interpolate
which lerps from the predecessor with the elapsed fraction).
The factor was computed as (after - now) / (after - before), which is
1 - t: querying at the before timestamp returned the after sample,
querying at the after timestamp returned the before sample, and
queries in between returned the pose mirrored about the segment
midpoint.
Compute t from the before sample so hand and body skeleton queries
between two tracking samples interpolate along the motion instead of
against it.
In case of discarded frames (no end frame or no layers) we don't get
called on mark_gpu_done, so times don't get updated. If the game
continues to call waitFrame, we advance the predicted display time, but
may enter the immediate return, accumulating delay.
The SPS allowed coding blocks down to 8x8. On NVIDIA that costs 4x the
encode time: 13.4 ms per 1920x1472 frame against 4.4 ms for the same
GPU's NVENC path, so the encoder sustained 36.5 fps against a 90 Hz
target. At 16x16 it is 3.2 ms and holds 90 Hz.
The angular velocity was worked out with the wrong formula. It used
dq * q, but it should be dq * q^-1 (the inverse of q, which is just
its conjugate here because q is normalized).
Using q instead of its inverse pointed the angular velocity the wrong
way. Games read this value when you let go of a controller, so thrown
objects came out spinning wrong.
With the pinned SDK gone, CI builds against distro headers, so raise the
server requirement to 1.4.304 to match what is actually tested and
supported: Ubuntu 26.04 has 1.4.341, the Ubuntu package build (plucky)
has 1.4.304, as does Debian 13. This subsumes the separate 1.3.283 check
for Vulkan video encode, and makes VK_HEADER_VERSION >= 304 always true
for the server. The client conditionals stay, it is built for Android
against the 1.3.268 headers fetched in common/CMakeLists.txt.
copy_mutex duplicated the pthread_mutex_t of vk.queue.mutex into
main_queue->mutex by value, which creates an independent mutex: monado
submissions on IPC client threads (comp_swapchain image transitions,
vkDeviceWaitIdle in image_cleanup) did not exclude wivrn's layer_commit
and encoder submissions on the same VkQueue. This intermittently
corrupts driver state when a client creates or destroys swapchains
while streaming, observed as SIGSEGV inside vkQueueSubmit2 and as
uncaught vk::SystemError from layer_commit on NVIDIA.
Initialize monado's queue mutexes and redirect vk.queue.mutex to the
main queue os_mutex so every submitter takes the same lock.