libvhal-client is a library written in C++17 for Touch, Joystick, GPS, Camera, Audio and Sensor modules. Currently only Camera is supported. Using this library, a client can interact with Camera Vhal without worrying about socket connection, data structure, command details. Instead, the library exposes simple and powerful API that simplifies the life of VHAL client modules such as CG-Proxy, Streamer, etc. Any internal changes in the Camera VHAL and the library is abstracted and hence it won’t affect client code. New features or improvements can be provided via APIs rather than new socket commands, and other details.
shakthi@mypc build (master) mkdir build
shakthi@mypc build (master) cd build
shakthi@mypc build (master) $ cmake ..
-- The CXX compiler identification is GNU 7.5.0
-- Check for working CXX compiler: /usr/bin/c++
-- Check for working CXX compiler: /usr/bin/c++ -- works
-- Detecting CXX compiler ABI info
-- Detecting CXX compiler ABI info - done
-- Detecting CXX compile features
-- Detecting CXX compile features - done
-- Project name: vhal-client
-- Looking for C++ include pthread.h
-- Looking for C++ include pthread.h - found
-- Looking for pthread_create
-- Looking for pthread_create - not found
-- Looking for pthread_create in pthreads
-- Looking for pthread_create in pthreads - not found
-- Looking for pthread_create in pthread
-- Looking for pthread_create in pthread - found
-- Found Threads: TRUE
-- Configuring done
-- Generating done
-- Build files have been written to: /home/shakthi/gitlab/libvhal-client/build
shakthi@Lenovo-Yoga-C740-14IML build (master) $ cmake --build .
Scanning dependencies of target vhal-client
[ 14%] Building CXX object source/CMakeFiles/vhal-client.dir/unix_stream_socket_client.cc.o
[ 28%] Building CXX object source/CMakeFiles/vhal-client.dir/vhal_video_sink.cc.o
[ 42%] Linking CXX shared library ../../libs/linux/libvhal-client.so
[ 42%] Built target vhal-client
Scanning dependencies of target camera_socket_client
[ 57%] Building CXX object examples/CMakeFiles/camera_socket_client.dir/camera_socket_client.cc.o
[ 71%] Linking CXX executable ../../bins/linux/camera_socket_client
[ 71%] Built target camera_socket_client
Scanning dependencies of target camera_client
[ 85%] Building CXX object examples/CMakeFiles/camera_client.dir/camera_client.cc.o
[100%] Linking CXX executable ../../bins/linux/camera_client
[100%] Built target camera_client
### To install/copy libvhal-clinet.so in to destination directory
cmake -DCMAKE_INSTALL_PREFIX=<Absolute path to install>
-DCMAKE_INSTALL_LIBDIR=<Relative directory name to install libs>
-DCMAKE_INSTALL_BINDIR=<Relative directory name to install bins> ..
Example:
libvhal-client$mkdir build
libvhal-client$cd build
cmake -DCMAKE_INSTALL_PREFIX=$(pwd)/..
-DCMAKE_INSTALL_LIBDIR=lib ..
libvhal-client$make
libvhal-client$make install
input-pipe<x>-<y> are used for touch and joystick. The touch supports fingers number from 1 to 10. The finger id should start from 0 in every session. Sample code is here:
examples/virtual_input_receiver_sample.cc
- On a normal mobile phone tablet, the touch size corresponds to the screen size one by one. For example, the screen is X * Y = 1024 * 600, and the touch size is also x * y = 1024 * 600.
- When initially implementing Remote Input, the screen size X * Y = 720 * 1280 required at the time. We selected the touch size x * y = 720 * 1280. One feature that was planned to be done at that time was to dynamically set the screen size according to the surfaceFlinger display.
- Touch size may not be consistent with display size Just do the appropriate mapping. The accuracy has changed slightly. For example, Touch is currently x * y = 720 * 1280. The display size is X * Y = 1024 * 600. Given Client X 'and Y', the touch input should be x '= X' * (720-1) / (1024-1). y '= Y' * (1280 -1) / (600-1). 720-1 and 1024-1 are because the maximum value of the abscissa is 719 and 1023. The ordinate is similar.
- Because the size of the client is variable, the latest solution does not use the dynamic setting of the screen size of the surfaceFlinger display. But in order to preserve the accuracy as much as possible, set the Touch x * y = 32768 * 32768 (starting from 0, signed 16-bit integer maximum + 1). The latest calculation method is: Given Client X 'and Y', then What should be input to touch is x '= X' * (32768-1) / (1024-1). Y '= Y' * (32768 -1) / (600-1).
- The screen size X * Y can be converted into a float value including 0 to 1. For example, the screen size X * Y = 720 * 1280, given Client X 'and Y', the conversion formula to [0, 1] is: x01 = X '/ (720 -1), y01 = Y' / (1280 -1). Given x01 'and y01', according to the latest version, what should be input to touch is x '= x01' * (32768-1). Y '= y01' * (32768 -1).
- Please confirm the version you are using (Touch is x * y = 720 * 1280, or Touch is x * y = 32768 * 32768), screen size, and convert the corresponding value.
Touch protocol follows OpenSTF minitouch protocol: https://github.com/openstf/minitouch
It is assumed that you now have an open connection to the minitouch socket or you're running minitouch in stdin/file mode. If not, follow the instructions above.
The minitouch protocol is based on LF-separated lines. Each line is a separate command, and each line begins with a single ASCII letter which specifies the command type. Space-separated command-specific arguments then follow.
When you first open a connection to the socket, you'll receive a header with metadata which you'll need to read from the socket. Other than that there will be no responses of any kind.
Example input: c
Commits the current set of changed touches, causing them to play out on the screen. Note that nothing visible will happen until you commit.
Commits are not required to list all active contacts. Changes from the previous state are enough.
Same goes for multi-contact touches. The contacts may move around in separate commits or even the same commit. If one contact moves, the others are not required to.
The order of touches in a single commit is not important either. For example you can list contact 5 before contact 0.
Note however that you cannot have more than one d, m or u for the same <contact> in one commit.
Example input: r
Attemps to reset the current set of touches by creating appropriate u events and then committing them. As an invalid sequence of events may cause the screen to freeze, you should call for a reset if you have any doubts about the integrity of your events. For example, two touchstart events for the same contact is very suspect and most likely means that you lost a touchend event somehow.
We try to discard obviously out-of-order events automatically, but sometimes it's not enough.
If the screen freezes you'll have to reboot the device. With careful use this will not happen.
Example input: d 0 10 10 50
Schedules a touch down on contact <contact> at <x>,<y> with <pressure> pressure for the next commit.
You cannot have more than one d, m or u for the same <contact> in one commit.
Example input: m 0 10 10 50
Schedules a touch move on contact <contact> at <x>,<y> with <pressure> pressure for the next commit.
You cannot have more than one d, m or u for the same <contact> in one commit.
Example input: u 0
Schedules a touch up on contact <contact>. If you need the contact to move first, use a combination of m and u separated by a commit.
You cannot have more than one d, m or u for the same <contact> in one commit.
Example input: w 50
Immediately waits for <ms> milliseconds. Will not commit the queue or do anything else.
Tap on (10, 10) with 50 pressure using a single contact.
d 0 10 10 50
c
u 0
c
Long tap on (10, 10) with 50 pressure using a single contact.
d 0 10 10 50
c
<wait in your own code>
u 0
c
Tap on (10, 10) and (20, 20) simultaneously with 50 pressure using two contacts.
d 0 10 10 50
d 1 20 20 50
c
u 0
u 1
c
Tap on (10, 10), keep it pressed, then after a while also tap on (20, 20), keep it pressed, then release the first contact and finally release the second contact.
d 0 10 10 50
c
<wait in your own code>
d 1 20 20 50
c
<wait in your own code>
u 0
c
<wait in your own code>
u 1
c
Swipe from (0, 0) to (100, 0) using a single contact. You'll need to wait between commits in your own code to slow it down.
d 0 0 0 50
c
m 0 20 0 50
c
m 0 40 0 50
c
m 0 60 0 50
c
m 0 80 0 50
c
m 0 100 0 50
c
u 0
c
Pinch with two contacts going from (0, 100) to (50, 50) and (100, 0) to (50, 50). You'll need to wait between commits in your own code to slow it down.
d 0 0 100 50
d 1 100 0 50
c
m 0 10 90 50
m 1 90 10 50
c
m 0 20 80 50
m 1 80 20 50
c
m 0 20 80 50
m 1 80 20 50
c
m 0 30 70 50
m 1 70 30 50
c
m 0 40 60 50
m 1 60 40 50
c
m 0 50 50 50
m 1 50 50 50
c
u 0
u 1
c
The same pinch but with more chaotic (or natural) ordering.
d 1 100 0 50
c
d 0 0 100 50
c
m 1 90 10 50
m 0 10 90 50
c
m 0 20 80 50
c
m 1 80 20 50
c
m 0 20 80 50
m 1 80 20 50
c
m 0 30 70 50
c
m 1 70 30 50
c
m 1 60 40 50
c
m 0 40 60 50
c
m 0 50 50 50
m 1 50 50 50
c
u 0
c
u 1
c
Link: https://www.win.tue.nl/~aeb/linux/kbd/scancodes-14.html
AIC could support multiple joysticks and accept a unified scan code depending on Generic.kl. For controller devices with different types or designed by different vendors, their scan code may be different, so customer needs to map the controller device's scan code to the AIC unified scan code when inject joystick's scan code.
Profile for a generic game controller
EV_KEY events scan code map
EV_ABS events axis code value map
If customer would develop your own client app in Android OS, the Android framework provides APIs for detecting and processing user input from game controllers, you could use these APIs to get the Android keycode. Though different joystick vendors' device may have different scan code, Android OS has masked the difference and you could get a unified Android keycode from the Android API. Then the Android keycode could be used to do the map work for joystick. Client app can inspect the key code by calling getKeyCode() or from key event callbacks such as onKeyDown().
Socket is used for GPS. Sample code is here:
examples/virtual_gps_receiver_sample.cc
Currently, it only supports Global Positioning System Fix Data(GPGGA).
Reference: http://aprs.gids.nl/nmea/#gga
Camera VHal runs socket server (UNIX, VSock are supported). VHAL Client library shall connect to socket server path or address/port endpoint.
- Create an object to libVHAL's
vhal::client::VideoSinkby passingvhal::client::UnixConnectionInfo. Then libVHAL will prepare and connect to Camera server using UNIX domain sockets.
UnixConnectionInfo conn_info = { socket_path, instance_id };
try {
video_sink = make_shared<VideoSink>(conn_info);
}- Register a callback with
vhal::client::VideoSinkfor Camera open/close commands.
video_sink->RegisterCallback([&](const VideoSink::CtrlMessage& ctrl_msg) {
// Callback Implementation here
}- Send the camera frames to libVHAL-client as below.
video_sink->SendDataPacket(inbuf.data(), inbuf_size);- For non-encoded data camera frames always of fixed size example I420 in such cases both vhal and libVHAL-client knows the data size exchange in such scenario SendRawPacket will used as below.
video_sink->SendRawPacket(inbuf.data(), inbuf_size);- To read vhal supported capability call GetCameraCapabilty after connection estabilish. On success, this api will return a shared_ptr of type struct camera_capability_t
std::shared_ptr<camera_capability_t> GetCameraCapabilty();- To set libvhal capability to vhal call SetCameraCapabilty
bool SetCameraCapabilty(std::vector<camera_info_t> camera_info);Example implementation to interact with libVHAL-client is present in examples/camera_client.cc
Audio VHAL runs socket server for INET domains. VHAL Client library(libvhal-client.so) shall connect to socket server path.
Here shown the E2E block diagram of audio
- Create an object to libVHAL's
vhal::client::AudioSinkby passingvhal::client::TcpConnectionInfo. Then libVHAL will prepare and connect to Audio server using TCP domain sockets.
TcpConnectionInfo conn_info = { ip_addr };
AudioSink audio_sink(conn_info);- Register a callback with
vhal::client::AudioSinkfor Audio open/close commands.
audio_sink->RegisterCallback([&](const AudioSink::CtrlMessage& ctrl_msg) {
// Callback Implementation here
}- Send the raw pcm audio data to libVHAL-client as below.
audio_sink->SendDataPacket(inbuf.data(), inbuf_size);Example implementation to interact with libVHAL-client is present in examples/audio_record_client.cc
Sensor VHAL runs socket server for UNIX and INET domains. VHAL Client library(libvhal-client.so) shall connect to socket server path.
Here shown the E2E block diagram of sensors and API sequence to communicate with libVHAL Client
- Create an object to libVHAL's SensorInterface class by passing vhal::client::UnixConnectionInfo. Then libVHAL will prepare and connect to Sensor server using UNIX domain sockets.
UnixConnectionInfo conn_info = { socket_path, instance_id };
try {
sensorHALIface = make_unique<SensorInterface>(conn_info);
}- Register a callback func to get sensor control packets upon received from Sensor VHAL server. CtrlPacket contains respective sensor's enable/disable info and sampling period.
sensorHALIface->RegisterCallback([&](const SensorInterface::CtrlPacket& ctrlPkt){});- Use below api to know list of sensors supported by libVHAL-client or AIC. This API returns supported sensor list in bitmap format.
uint64_t GetSupportedSensorList();- Fill the available sensor data into structure sensorDataPacket. Send the data to libVHAL-client like below
sensorHALIface->SendDataPacket(&event);Example implementation to interact with libVHAL-client is present in examples/sensor_client.cc