Description

Author: James Pangia

usage: ./js2mouse [deviceName] [L]

deviceName: the name of the joystick device to read; expects a js* device name
            If no device is specified, /dev/input/js0 is used.
L: specify that the left joystick should move the cursor (default uses right)

Description: reads the inputs from the specified joystick device and uses the joystick for keyboard/mouse inputs

Notes:

Designed for use with an XBox 360 controller; behavior for other joystick devices
is undefined.

used as example, but not directly copied: https://gist.github.com/jasonwhite/c5b2048c15993d285130

pretty sure that event type 129 is button init and 130 is axis init
also, when opening a js device, it runs through an init cycle that tests all the joystick elements

js2mouse_v1.c is an old version

Dependencies:

xdotool
    standalone in Debian-based systems; installed with `sudo apt install xdotool`
    standalone for Arch-based as well; installed with `sudo pacman -S[yu] xdotool`

Known Bugs

1

read() call returning 2^64 bytes when reading from the joystick file. even though there is a max parameter that seems like it should give the maximum number of bytes to read

2

memory overflow crash when reading from stdin. Occurs for both using getline and getchar

TODO:

- re-compile with debug info and check valgrind output. see if solving the "address is 0 bytes after a block of size 32 is alloc'd" error fixes the crash - translate usage of system() into exec family of functions (see man 3 execl) - using execl will also allow for checking if xdotool is installed - look into option to use wayland-based equivalent of xdotool (there exists ydotool, which is intended to be a drop-in replacement for xdotool) - security risk: config file could inject malicious code into system() calls but if the program reads the file expecting an int, then the program should crash at worst, so I think it's ok - research chardevice files to learn more about js0 - /!\ look into using access again; looks like it can return 0 on an empty device file

check file before reading in the loop record time since last input; prompt user to quit if left for too long port the config values to a struct that gets populated by a config file and pass the struct as a const pointer in each function that uses config values

Long-Term TODO:

- port to C++, making the joystick utilities being part of a class - investigate porting to Rust. If Rust supports an analog of the C joystick header, then that will probably get around the weird read bug. Might be able to use the C header directly, depending on how cross-language support works

dev notes:

xorg xlib docs: https://x.org/releases/current/doc/libX11/libX11/libX11.html

pointed me to xlib: https://forums.justlinux.com/showthread.php?104013-Controlling-the-mouse-from-bash-scripts

links to resources related to moving the mouse: https://tronche.com/gui/x/xlib/input/XWarpPointer.html https://www.x.org/releases/current/doc/man/man3/XSendEvent.3.xhtml

/!\example of using XLib: https://www.linuxquestions.org/questions/programming-9/simulating-a-mouse-click-594576/

Realized that the tool xdotool exists. Simple bash command that does all the stuff I need. Yes, I'm calling shell commands in a C program.

solution to getting both axis values from one event

code slippet from: https://archives.seul.org/linuxgames/Aug-1999/msg00107.html

while( 1 ) 	/* infinite loop */
{
	/* read the joystick state */
	read(joy_fd, &js, sizeof(struct js_event));

		/* see what to do with the event */
	switch (js.type & ~JS_EVENT_INIT)
	{
		case JS_EVENT_AXIS:
			axis   [ js.number ] = js.value;
			break;
		case JS_EVENT_BUTTON:
			button [ js.number ] = js.value;
			break;
	}

		/* print the results */
	printf( "X: %6d  Y: %6d  ", axis[0], axis[1] );

	if( num_of_axis > 2 )
		printf("Z: %6d  ", axis[2] );

	if( num_of_axis > 3 )
		printf("R: %6d  ", axis[3] );

	for( x=0 ; x<num_of_buttons ; ++x )
		printf("B%d: %d  ", x, button[x] );

	printf("  \r");
	fflush(stdout);
}

note that every event sets values in an axis[] array. This way, a running book of the axis states is kept, updated by every caught event

using supertuxkart to see how velocity is maintained

stk code base: https://github.com/supertuxkart/stk-code/tree/master/src

in src/karts/abstract_kart.cpp:241: body->setLinearVelocity(Vec3(0.0f)); body is of type btRigidBody.

This is from the Bullet physics library: https://pybullet.org/Bullet/BulletFull/classbtRigidBody.html

Note on read mode for sockets

in https://docs.python.org/3/howto/sockets.html#non-blocking-sockets: it mentions that O_NONBLOCK is a BSD flavor flag. O_NDELAY is a POSIX flavor flag. Perhaps using POSIX would solve the weird reading behavior

A note on checking for data in the file:

in https://unix.stackexchange.com/a/25607

they say to use poll() or select() to check for incoming data