tslib consists of the library libts and tools that help you calibrate and use it in your environment.

If you have problems, questions, ideas, or suggestions, please contact us by writing an email to tslib@lists.infradead.org, which is our mailing list.

Visit the tslib website for an overview of the project.

• setup and configure tslib
• filter modules
• the libts library
• building tslib
• hardware support
• contribute

tslib runs on various hardware architectures and operating systems, including GNU/Linux, FreeBSD, or Android/Linux. See building tslib for details. Apart from building the latest tarball release, running ./configure, make, and make install, tslib is available from distributors like Arch Linux / Arch Linux ARM, Buildroot, Debian / Ubuntu, Fedora, or OpenSUSE and their package management.

This is just an example /etc/ts.conf file. Touch samples flow from top to bottom. Each line specifies one module and its parameters. Modules are processed in order. Use one module_raw on top, that accesses your device, followed by any combination of filter modules.

module_raw input
module median depth=3
module dejitter delta=100
module linear

See the section below for available filters and their parameters. On Linux, your first commented-in line should always be module_raw input which offers one optional parameter: grab_events=1 if you want it to execute EVIOCGRAB on the device.

With this configuration file, we end up with the following data flow through the library:

driver --> raw read --> median  --> dejitter --> linear --> application (using `ts_read_mt()`)
           module       module      module       module

Calibration is applied by the linear plugin, which uses its own config file TSLIB_CALIBFILE. Don't edit this file manually. It is created by the ts_calibrate program:

# ts_calibrate

The calibration procedure simply requires you to touch a cross on the screen, where it appears, as accurately as possible.

You may quickly test the touch behaviour that results from the configured filters, using ts_test_mt:

# ts_test_mt

On the bottom left of the screen, you will see the available concurrent touch contacts supported, and whether it's because the driver says so, or ts_test_mt was started with the -j commandline option to overwrite it.

You may override the defaults. In most cases, though, you won't need to do so:

TSLIB_TSDEVICE          Touchscreen device file name.
                        Default:                automatic detection (on Linux)

TSLIB_CALIBFILE         Calibration file.
                        Default:                ${sysconfdir}/pointercal

TSLIB_CONFFILE          Config file.
                        Default:                ${sysconfdir}/ts.conf

TSLIB_PLUGINDIR         Plugin directory.
                        Default:                ${datadir}/plugins

TSLIB_CONSOLEDEVICE     Console device. (not needed when using --with-sdl2)
                        Default:                /dev/tty

TSLIB_FBDEVICE          Framebuffer device.
                        Default:                /dev/fb0

If you're using X.org graphical X server, things should be very easy. Install tslib and xf86-input-tslib, reboot, and you should instantly have your ts.conf filters running, without configuring anything else yourself.

TL;DR: Run tools/ts_uinput_start.sh during startup and use /dev/input/ts_uinput as your evdev input device.

tslib tries to automatically find your touchscreen device in /dev/input/event* on Linux. Now make ts_uinput use it, instead of your graphical environment directly:

# ts_uinput -d -v

-d makes the program return and run as a daemon in the background. -v makes it print the new /dev/input/eventX device node before returning.

Now make your graphical environment use that new input device, using evdev drivers.

• For Qt5 for example you'd probably set something like this:

  QT_QPA_EVDEV_TOUCHSCREEN_PARAMETERS=/dev/input/eventX:rotate=0

• For X11 you'd probably edit your xorg.conf Section "InputDevice" for your touchscreen to have

  Option "Device" "/dev/input/eventX"

Please consult your input system's documentation on how to use a specific evdev input device.

Let's recap the data flow here:

driver --> raw read --> filter(s) ... --> ts_uinput --> libevdev read  --> GUI app/toolkit
           module       module(s) ...     daemon        e.g. in libinput

/dev/input/event numbers are not persistent. In order to know in advance, what enumerated input device file is created by ts_uinput, you can use a symlink:

• use the included tools/ts_uinput_start.sh script that starts ts_uinput -d -v and creates the symlink called /dev/input/ts_uinput for you, or

• if you're using systemd, create the following udev rule, for example /etc/udev/rules.d/98-touchscreen.rules:

  SUBSYSTEM=="input", KERNEL=="event[0-9]*", ATTRS{name}=="NAME_OF_THE_TOUCH_CONTROLLER", SYMLINK+="input/ts", TAG+="systemd" ENV{SYSTEMD_WANTS}="ts_uinput.service"
  SUBSYSTEM=="input", KERNEL=="event[0-9]*", ATTRS{name}=="ts_uinput", SYMLINK+="input/ts_uinput"
  

where NAME_OF_THE_TOUCH_CONTROLLER the touchscreen found in your cat /proc/bus/input/devices | grep Name. The first rule is only needed, if tslib doesn't automatically choose the correct device for you.

In case you have to use non-default paths, create a file containing the environment for tslib, like /etc/ts.env

  TSLIB_CALIBFILE=/etc/pointercal
  TSLIB_CONFFILE=/etc/ts.conf
  TSLIB_PLUGINDIR=/usr/lib/ts

and create a systemd service file, such as /usr/lib/systemd/system/ts_uinput.service

  [Unit]
  Description=touchscreen input
  BindsTo=dev-input-ts.device
  After=dev-input-ts.device
  RequiresMountsFor=/etc/ts.env

  [Service]
  Type=forking
  EnvironmentFile=/etc/ts.env
  ExecStart=/usr/bin/ts_uinput -d

Adjust the paths. They could just as well be in /usr/local/, too.

There is no tool that we know of that reads tslib samples and uses the Windows touch injection API, for example (yet).

Linear scaling - calibration - module, primerily used for conversion of touch screen co-ordinates to screen co-ordinates. It applies the corrections as recorded and saved by the ts_calibrate tool.

Parameters (usually not needed):

• rot

  overwrite the rotation to apply. Clockwise: rot=1, upside down: rot=2, counter-clockwise: rot=3. Default: screen-rotation during ts_calibrate calibration.

• xyswap

  Interchange the X and Y co-ordinates -- no longer used or needed if the linear calibration utility ts_calibrate is used.

• pressure_offset

  Set the offset applied to the pressure value. Default: 0

• pressure_mul

  factor to multiply the pressure value by. Default: 1.

• pressure_div

  value to divide the pressure value by. Default: 1.

Example: module linear rot=0

Invert values in the X and/or Y direction around the given value. There are no default values. If specified, a value has to be set. If one axis is not specified, it's simply untouched.

Parameters:

• x0

  X-axis (horizontal) value around which to invert.

• y0

  Y-axis (vertical) value around which to invert.

Example: module invert y0=640 (Y-axis inverted for 640 screen height, X-axis untouched)

The median filter reduces noise in the samples' coordinate values. It is able to filter undesired single large jumps in the signal. For some theory, see Wikipedia

Parameters:

• depth

  Number of samples to apply the median filter to. Default: 3.

Example: module median depth=5

Pressure threshold filter. Given that a release is always pressure 0 and a press is always >= 1, this discards samples below / above the specified pressure threshold.

Parameters:

• pmin

  Minimum pressure value for a sample to be valid. Default: 1.

• pmax

  Maximum pressure value for a sample to be valid. Default: (INT_MAX).

Example: module pthres pmin=10

Infinite impulse response filter. This is a smoothing filter to remove low-level noise. There is a trade-off between noise removal (smoothing) and responsiveness. The parameters N and D specify the level of smoothing in the form of a fraction (N/D).

Wikipedia has some theory.

Parameters:

• N

  numerator of the smoothing fraction. Default: 0.

• D

  denominator of the smoothing fraction. Default: 1.

Example: module iir N=6 D=10

Removes jitter on the X and Y co-ordinates. This is achieved by applying a weighted smoothing filter. The latest samples have most weight; earlier samples have less weight. This allows one to achieve 1:1 input->output rate. See Wikipedia for some theory.

Parameters:

• delta

  Squared distance between two samples ((X2-X1)^2 + (Y2-Y1)^2) that defines the 'quick motion' threshold. If the pen moves quick, it is not feasible to smooth pen motion, besides quick motion is not precise anyway; so if quick motion is detected the module just discards the backlog and simply copies input to output. Default: 100.

Example: module dejitter delta=100

Simple debounce mechanism that drops input events for the specified time after a touch gesture stopped. Wikipedia has more theory.

Parameters:

• drop_threshold

  drop events up to this number of milliseconds after the last release event. Default: 0.

Example: module debounce drop_threshold=40

Skip nhead samples after press and ntail samples before release. This should help if the first or last samples are unreliable for the device.

Parameters:

• nhead

  Number of events to drop after pressure. Default: 1.

• ntail

  Number of events to drop before release. Default: 1.

Example: module skip nhead=2 ntail=1

Simple lowpass exponential averaging filtering module.

Parameters:

• factor

  floating point value between 0 and 1; for example 0.2 for more smoothing or 0.8 for less. Default: 0.4.

• threshold

  x or y minimum distance between two samples to start applying the filter. Default: 2.

Example: module lowpass factor=0.5 threshold=1

After "pen/finger down", N number of input samples need to be delivered by the driver before they are considered valid and passed on to the user (application). If "pen/finger up" occurs before N samples are being read from the device driver, tslib will drop the "tap".

This filter can be used to avoid touches that, for example, result from electromagnetic interference. These are known to be shorter than one a real user would create.

In contrast to the skip filter, the evthres filter will not cut out events that are part of a real touch input. It will only cut out one whole "tap", if short enough.

Compared to the debounce filter, this filter will act on every occurrence of "pen/finger down", including the first one, not only starting with the second. Also, debounce is time-based, not events-based.

Parameters:

• N

  number of events between "down" and "up" that must be provided by the device driver for the touch to be considered real and passed on.

Example: module evthres N=5

The goal of this filter is to drop input touch events that lie outside of the Min/Max values for the x/y event codes, i.e. touch points that might (due to applied filters) lie ouside of the visible framebuffer.

It sends "pen/finger up" for a given slot when said values are outside of the framebuffer.

It read the framebuffer dimensions from TSLIB_CALIBFILE which is generated and saved by the ts_calibrate tool.

Example: module crop

Variance filter. Tries to do its best in order to filter out random noise coming from touchscreen ADC's. This is achieved by limiting the sample movement speed to some value (e.g. the pen is not supposed to move quicker than some threshold).

There is no multitouch support for this filter (yet). ts_read_mt() will limit your input to one slot when this filter is used. Try using the median filter instead.

Parameters:

• delta

  Set the squared distance in touchscreen units between previous and current pen position (e.g. (X2-X1)^2 + (Y2-Y1)^2). This defines the criteria for determining whenever two samples are 'near' or 'far' to each other.

  Now if the distance between previous and current sample is 'far', the sample is marked as 'potential noise'. This doesn't mean yet that it will be discarded; if the next reading will be close to it, this will be considered just a regular 'quick motion' event, and it will sneak to the next layer. Also, if the sample after the 'potential noise' is 'far' from both previously discussed samples, this is also considered a 'quick motion' event and the sample sneaks into the output stream.

The following example setup

       |--------|       |-----|      |--------------|
x ---> | median | ----> | IIR | ---> |              | ---> x'
       |--------|    -> |-----|      |    screen    |
                    |                |  transform   |
                    |                | (calibrate)  |
       |--------|   |   |-----|      |              |
y ---> | median | ----> | IIR | ---> |              | ---> y'
       |--------|   |-> |-----|      |--------------|
                    |
                    |
             |----------|
p ---------> | debounce | -------------------------------> p'
             |----------|

would be achieved by the following ts.conf:

module_raw input
module debounce drop_threshold=40
module median depth=5
module iir N=6 D=10
module linear

while you are free to play with the parameter values.

The graphical tools support rotating the screen, see ts_calibrate --help or the man pages for the details. ts_calibrate will (in the background) ignore screen rotation but will save the current rotation state in the "calibration file" TSLIB_CALIBFILE. The linear filter module will pick that up and apply the given rotation. It can be overwritten by using the rot module parameter: module linear rot=0 (if your toolkit or higher level application does rotation).

The API is documented in our man pages in the doc directory. Check out our tests directory for examples how to use it.

tslib_version()
ts_libversion()
ts_open()
ts_config()
ts_setup()
ts_close()
ts_reconfig()
ts_option()
ts_fd()
ts_load_module()
ts_read()
ts_read_raw()
ts_read_mt()
ts_read_raw_mt()
int (*ts_error_fn)(const char *fmt, va_list ap)
int (*ts_open_restricted)(const char *path, int flags, void *user_data)
void (*ts_close_restricted)(int fd, void *user_data)
ts_get_eventpath()
ts_print_ascii_logo()
ts_conf_get()
ts_conf_set()

To use the library from C or C++, include the following preprocessor directive in your source files:

#include <tslib.h>

To link with the library, specify -lts as an argument to the linker.

On UNIX systems, you can use pkg-config to automatically select the appropriate compiler and linker switches for libts. The PKG_CHECK_MODULES m4 macro may be used to automatically set the appropriate Makefile variables:

PKG_CHECK_MODULES([TSLIB], [tslib >= 1.10],,
  AC_MSG_ERROR([libts 1.10 or newer not found.])
)

If you want to support tslib < 1.2, while still supporting multitouch and all recent versions of tslib, you'd do something like this:

#include <tslib.h>

#ifndef TSLIB_VERSION_MT
        /* ts_read() as before (due to old tslib) */
#else
        /* new ts_setup() and ret = ts_read_mt() */
        if (ret == -ENOSYS)
                /* ts_read() as before (due to user config) */
#endif

This is a complete example program, similar to ts_print_mt.c:

#include <stdio.h>
#include <stdint.h>
#include <stdlib.h>
#include <fcntl.h>
#include <sys/time.h>
#include <unistd.h>
#include <errno.h>

#include <tslib.h>

#define SLOTS 5
#define SAMPLES 1

int main(int argc, char **argv)
{
    struct tsdev *ts;
    char *tsdevice = NULL;
    struct ts_sample_mt **samp_mt = NULL;
    int ret, i, j;

    ts = ts_setup(tsdevice, 0);
    if (!ts) {
            perror("ts_setup");
            return -1;
    }

    samp_mt = malloc(SAMPLES * sizeof(struct ts_sample_mt *));
    if (!samp_mt) {
            ts_close(ts);
            return -ENOMEM;
    }
    for (i = 0; i < SAMPLES; i++) {
            samp_mt[i] = calloc(SLOTS, sizeof(struct ts_sample_mt));
            if (!samp_mt[i]) {
                    for (i--; i >= 0; i--)
                            free(samp_mt[i]);
                    free(samp_mt);
                    ts_close(ts);
                    return -ENOMEM;
            }
    }

    while (1) {
            ret = ts_read_mt(ts, samp_mt, SLOTS, SAMPLES);
            if (ret < 0) {
                    perror("ts_read_mt");
                    ts_close(ts);
                    exit(1);
            }

            for (j = 0; j < ret; j++) {
            	for (i = 0; i < SLOTS; i++) {
		#ifdef TSLIB_MT_VALID
			if (!(samp_mt[j][i].valid & TSLIB_MT_VALID))
				continue;
		#else
			if (samp_mt[j][i].valid < 1)
				continue;
		#endif

			printf("%ld.%06ld: (slot %d) %6d %6d %6d\n",
			       samp_mt[j][i].tv.tv_sec,
			       samp_mt[j][i].tv.tv_usec,
			       samp_mt[j][i].slot,
			       samp_mt[j][i].x,
			       samp_mt[j][i].y,
			       samp_mt[j][i].pressure);
                    }
            }
    }

    ts_close(ts);
}

If you know how many slots your device can handle, you could avoid malloc:

struct ts_sample_mt TouchScreenSamples[SAMPLES][SLOTS];

struct ts_sample_mt (*pTouchScreenSamples)[SLOTS] = TouchScreenSamples;
struct ts_sample_mt *ts_samp[SAMPLES];
for (i = 0; i < SAMPLES; i++)
        ts_samp[i] = pTouchScreenSamples[i];

and call ts_read_mt() like so

ts_read_mt(ts, ts_samp, SLOTS, SAMPLES);

Wikipedia has background information.

Usually, and every time until now, libts does not break the ABI and your application can continue using libts after upgrading. Specifically this is indicated by the libts library version's major number, which should always stay the same. According to our versioning scheme, the major number is incremented only if we break backwards compatibility. The second or third minor version will increase with releases. In the following example

libts.so -> libts.so.0.7.0
libts.so.0 -> libts.so.0.7.0
libts.so.0.7.0

use libts.so for using tslib unconditionally and libts.so.0 to make sure your current application never breaks.

If a release includes changes such as added features, the second number is incremented and the third is set to zero. If a release includes mostly just bugfixes, only the third number is incremented.

A tslib tarball version number doesn't tell you anything about its backwards compatibility.

• libc (with libdl only when building dynamically linked)
• libsdl2-dev (only when using --with-sdl2 for SDL2 graphical applications)

• libevdev - access wrapper for event device access, uinput too ("Linux API")
• libinput - handle input devices for Wayland (uses libevdev)
• xf86-input-evdev - evdev plugin for X (uses libevdev)

This lists the programs for the every day use of tslib, facing the outside world. For testing purposes there are tools like ts_test_mt too.

• ts_calibrate - graphical calibration tool. Configures the linear and crop filter modules.
• ts_uinput - userspace evdev driver for the tslib-filtered samples.

• xf86-input-tslib - direct tslib input driver for X11
• qtslib - direct Qt5 tslib input plugin
• enlightenment - A Window Manager (direct support in framebuffer mode, X11 via xf86-input-tslib)
• DirectFB - Graphics library on top of framebuffer

struct tslib_module_info
struct tslib_vars
struct tslib_ops
tslib_parse_vars(struct tslib_module_info *,const struct tslib_vars *, int, const char *);

tslib modules (filter or driver/raw module) in the plugins directory need to implement mod_init(). If the module takes parameters, it has to declare a const struct tslib_vars and pass that, its lengths and the params string that is passed to mod_init to tslib_parse_vars() during mod_init().

Furthermore a const struct tslib_ops has to be declared, with its members pointing to the module's implementation of module-operations like read_mt that get called in the chain of filters.

libts can be built to fit your needs. Use the configure script to enable only the modules you need. By default, libts is built as a shared library, with each module being a shared library object itself. You can, however, configure tslib to build libts statically linked, and the needed modules compiled inside of libts. Here's an example for this:

./configure --enable-static --disable-shared --enable-input=static --enable-linear=static --enable-iir=static

This should result in a libts.a of roughly 50 kilobytes, ready for using calibration (linear filter) and the infinite impulse response filter in ts.conf.

Alternatively, you can use CMake to build the project. To create building in the project tree:

mkdir build && cd build
cmake ../
# or adding configuration options: cmake -Denable-input-evdev=ON ../
cmake --build .
cmake -P cmake_install.cmake

By default, the core tslib is built as a shared library. In order to build it as static, add -DBUILD_SHARED_LIBS=OFF to the configure line.

Also, the plugins are by default built as shared. Add -Dstatic-<module>=ON to the configuration step to build plugins statically into the core tslib. To disable and enable modules, use flags: -Denable-<module>=ON/OFF.

The following is a minimal example of how to use tslib, built with CMake, in your client app. Adding tslib::tslib as a link target will add required dependencies and include directories generated build files.

cmake_minimum_required(VERSION 3.10)
project(tslib_client)
find_package(tslib 1.16)
add_executable(tslib_client main.c)
target_link_libraries(tslib_client PUBLIC tslib::tslib)

In case you cannot draw directly on the framebuffer, there is an experimental implementation of the necessary graphical tools using SDL2. They are more portable, but require more resources to run. To use them, make sure you have SDL2 and the development headers installed and use ./configure --with-sdl2.

tslib is cross-platform; you should be able to build it on a large variety of operating systems.

This is the hardware independent core part: libts and all filter modules as shared libraries, build on the following operating systems and probably more.

• GNU / Linux
• Android / Linux
• FreeBSD
• GNU / Hurd
• Haiku
• Windows
• Mac OS X

This makes the thing usable in the real world because it accesses your device. See hardware support for the currently possible configuration for your platform.

The libts default configuration currently has the following input modules disabled:

• cy8mrln-palmpre
• dmc_dus3000
• galax
• arctic2
• corgi
• collie
• dmc
• h3600
• mk712
• ucb1x00
• tatung

Please note that this list may grow over time. If you rely on a particular input plugin, you should enable it explicitly. On Linux, you should only need input though.

• GNU / Linux - all (most importantly input)
  • ./configure
• Android / Linux - all (most importantly input)
  • ./configure
• FreeBSD - almost all (most importantly input)
  • ./configure --disable-waveshare
• GNU / Hurd - some, see hardware support
  • ./configure --disable-input --disable-waveshare
• Haiku - some, see hardware support
  • ./configure --disable-input --disable-touchkit --disable-waveshare
• Windows - no tslib module for the Windows touchscreen API (yet)
  • ./configure --with-sdl2 --disable-input --disable-touchkit --disable-waveshare

Writing your own plugin is quite easy, in case an existing one doesn't fit.

• GNU / Linux - all
• Android / Linux - all (?)
• FreeBSD - all
• GNU / Hurd - ts_print_mt, ts_print, ts_print_raw, ts_finddev
• Haiku - ts_print_mt, ts_print, ts_print_raw, ts_finddev
• Windows - ts_print.exe, ts_print_raw.exe ts_print_mt.exe ts_test_mt.exe ts_calibrate.exe

For GNU/Linux all architectures are very well covered, thanks to Debian, Arch Linux, and others.

Please help porting missing programs!

TL;DR: On Linux, use module_raw input

For mostly historical reasons, tslib includes device specific module_raw userspace drivers. The ts.conf man page has details on the available module_raw drivers; not all of them are listed in the default etc/ts.conf config file. Those are to be considered workarounds and may get disabled in the default configuration in the future. If you use one of those, please ./configure --enable-... it explicitly.

It is strongly recommended to have a real device driver for your system and use a generic access module_raw of tslib. For Linux (evdev) this is called input. There is an equivalent experimental module that needs libevdev installed: module_raw input_evdev.

This project exists [thanks] to all the people who contribute. [Contribute].

In case you like our project, use it in your professional environment, and want it to stay maintained, please consider sponsoring the current maintainer via Github (or ask for a btc address if that's what you do). Maintenance costs time and money. thank you very much.

• NLNet supports improvements and maintenance 2024 / 2025

You can become a sponsor and be listed here.