Getting started with libmapper and Python

Since libmapper uses GNU autoconf, getting started with the library is the same as any other library on Linux; use ./configure and then make to compile it. You'll need swig available if you want to compile the Python bindings. On Mac OS X, we provide a precompiled Framework bundle for 32- and 64-bit Intel platforms, so using it with XCode should be a matter of including it in your project.

Once you have libmapper installed, it can be imported into your program:

import mapper

Overview of the API organization

The libmapper API is is divided into the following sections:

For this tutorial, the only sections to pay attention to are Devices and Signals. Networks are reserved for providing custom networking configurations, but in general you don't need to worry about it. Databases, Maps and Slots are mostly used when building user interfaces for designing mapping configurations.


Creating a device

To create a libmapper device, it is necessary to provide a device name to the constructor. There is an initialization period after a device is created where a unique ordinal is chosen to append to the device name. This allows multiple devices with the same name to exist on the network.

If no other arguments are given, libmapper will randomly choose a port to use for exchanging signal data. If desired, a second argument setting a specific "starting port" can be given, but the allocation algorithm will possibly choose another port number close to it if the port is in use.

A third optional parameter of the constructor is a network object. It is not necessary to provide this, but can be used to specify different networking parameters, such as specifying the name of the network interface to use.

An example of creating a device:

dev = mapper.device("my_device")

Polling the device

The device lifecycle looks like this, in terrible ASCII diagram art:

creation --> poll --+--> destruction
              |     |

In other words, after a device is created, it must be continuously polled during its lifetime.

The polling is necessary for several reasons: to respond to requests on the admin bus; to check for incoming signals; to update outgoing signals. Therefore even a device that does not have signals must be polled. The user program must organize to have a timer or idle handler which can poll the device often enough. The polling interval is not extremely sensitive, but should be 100 ms or less. The more often it is polled, the faster it can handle incoming signals.

The poll function can be blocking or non-blocking, depending on how you want your application to behave. It takes a number of milliseconds during which it should do some work, or 0 if it should check for any immediate actions and then return without waiting:


An example of calling it with non-blocking behaviour:


If your polling is in the middle of a processing function or in response to a GUI event for example, non-blocking behaviour is desired. On the other hand if you put it in the middle of a loop which reads incoming data at intervals or steps through a simulation for example, you can use poll() as your "sleep" function, so that it will react to network activity while waiting.

It returns the number of messages handled, so optionally you could continue to call it until there are no more messages waiting. Of course, you should be careful doing that without limiting the time it will loop for, since if the incoming stream is fast enough you might never get anything else done!

Note that an important difference between blocking and non-blocking polling is that during the blocking period, messages will be handled immediately as they are received. On the other hand, if you use your own sleep, messages will be queued up until you can call poll(); stated differently, it will "time-quantize" the message handling. This is not necessarily bad, but you should be aware of this effect.

Since there is a delay before the device is completely initialized, it is sometimes useful to be able to determine this using ready(). Only when ready() returns non-zero is it valid to use the device's name.


Now that we know how to create a device and poll it, we only need to know how to add signals in order to give our program some input/output functionality. While libmapper enables arbitrary connections between any declared signals, we still find it helpful to distinguish between two type of signals: inputs and outputs.

This can become a bit confusing, since the "reverb" parameter of a sound synthesizer might be updated locally through user interaction with a GUI, however the normal use of this signal is as a destination for control data streams so it should be defined as an input signal. Note that this distinction is to help with GUI organization and user-understanding – libmapper enables connections from input signals and to output signals if desired.

Creating a signal

We'll start with creating a "sender", so we will first talk about how to update output signals. A signal requires a bit more information than a device, much of which is optional:

for input signals there is an additional argument:


sig_in = dev.add_input_signal("my_input", 1, 'f', "m/s", -10, 10, h)
sig_out = dev.add_output_signal("my_output", 4, 'i', None, 0, 1000)

The only required parameters here are the signal "length", its name, and data type. Signals are assumed to be vectors of values, so for usual single-valued signals, a length of 1 should be specified. Finally, supported types are currently 'i' or 'f' for int or float values, respectively.

The other parameters are not strictly required, but the more information you provide, the more the mapper can do some things automatically. For example, if minimum and maximum are provided, it will be possible to create linear-scaled connections very quickly. If unit is provided, the mapper will be able to similarly figure out a linear scaling based on unit conversion (centimeters to inches for example). Currently automatic unit-based scaling is not a supported feature, but will be added in the future. You can take advantage of this future development by simply providing unit information whenever it is available. It is also helpful documentation for users.

Lastly, it is usually necessary to be informed when input signal values change. This is done by providing a function to be called whenever its value is modified by an incoming message. It is passed in the handler parameter.

An example of creating a "barebones" int scalar output signal with no unit, minimum, or maximum information:

outA = dev.add_output_signal("outA", 1, 'i', None, None, None)

or omitting some arguments:

outA = dev.add_output_signal("outA", 1, 'i')

An example of a float signal where some more information is provided:

sensor1 = dev.add_output_signal("sensor1", 1, 'f', "V", 0.0, 5.0)

So far we know how to create a device and to specify an output signal for it. To recap, let's review the code so far:

import mapper

dev = mapper.device("test_sender")
sensor1 = dev.add_output_signal("sensor1", 1, 'f', "V", 0.0, 5.0)

while 1:
    ... do stuff ...
    ... update signals ...

It is possible to retrieve a device's inputs or outputs by name or by index at a later time using the functions get_signal_by_<name/index>.

Updating signals

We can imagine the above program getting sensor information in a loop. It could be running on an network-enabled ARM device and reading the ADC register directly, or it could be running on a computer and reading data from an Arduino over a USB serial port, or it could just be a mouse-controlled GUI slider. However it's getting the data, it must provide it to libmapper so that it will be sent to other devices if that signal is mapped.

This is accomplished by the update function:


So in the "sensor 1 voltage" example, assuming in do_stuff() we have some code which reads sensor 1's value into a float variable called v1, the loop becomes:

while 1:
    v1 = do_stuff()

This is about all that is needed to expose sensor 1's voltage to the network as a mappable parameter. The libmapper GUI can now be used to create a mapping between this value and a receiver, where it could control a synthesizer parameter or change the brightness of an LED, or whatever else you want to do.

Signal conditioning

Most synthesizers of course will not know what to do with "voltage"--it is an electrical property that has nothing to do with sound or music. This is where libmapper really becomes useful.

Scaling or other signal conditioning can be taken care of before exposing the signal, or it can be performed as part of the mapping. Since the end user can demand any mathematical operation be performed on the signal, he can perform whatever mappings between signals as he wishes.

As a developer, it is therefore your job to provide information that will be useful to the end user.

For example, if sensor 1 is a position sensor, instead of publishing "voltage", you could convert it to centimeters or meters based on the known dimensions of the sensor, and publish a "sensor1/position" signal instead, providing the unit information as well.

We call such signals "semantic", because they provide information with more meaning than a relatively uninformative value based on the electrical properties of the sensing technique. Some sensors can benefit from low-pass filtering or other measures to reduce noise. Some sensor data may need to be combined in order to derive physical meaning. What you choose to expose as outputs of your device is entirely application-dependent.

You can even publish both "sensor1/position" and "sensor1/voltage" if desired, in order to expose both processed and raw data. Keep in mind that these will not take up significant processing time, and zero network bandwidth, if they are not mapped.

Receiving signals

Now that we know how to create a sender, it would be useful to also know how to receive signals, so that we can create a sender-receiver pair to test out the provided mapping functionality.

As mentioned above, the add_input_signal() function takes an optional handler. This is a function that will be called whenever the value of that signal changes. To create a receiver for a synthesizer parameter "pulse width" (given as a ratio between 0 and 1), specify a handler when calling add_input_signal(). We'll imagine there is some python synthesizer implemented as a class synthesizer which has functions setPulseWidth() which sets the pulse width in a thread-safe manner, and startAudioInBackground() which sets up the audio thread.

Let's use a real-world example using the pyo DSP library for Python to create a simple synth consisting of one sine wave. For now, we will only worry about controlling one parameter: the frequency of the sine.

We need to create a handler function for libmapper to update the pyo synth:

def frequency_handler(sig, id, val, timetag):
        print 'exception'
        print sig, val

Then our program will look like this:

from pyo import *
import mapper

# Some pyo stuff
synth = Server().boot().start()
sine = Sine(freq=200, mul=0.5).out()

def freq_handler(sig, id, val, timetag):
        print 'exception'
        print sig, val

dev = mapper.device('pyo_example')
dev.add_input_signal('frequency', 1, 'f', 'Hz', 20, 2000, freq_handler)

while True:


Alternately, we can simplify our code by using a lambda function instead of a separate handler:

from pyo import *
import mapper

# Some pyo stuff
synth = Server().boot().start()
sine = Sine(freq=200, mul=0.5).out()

dev = mapper.device('pyo_example')
dev.add_input_signal('frequency', 1, 'f', "Hz", 20, 2000,
                     lambda s, i, f, t: sine.setFreq(f))

while True:


Working with timetags

libmapper uses the mapper_timetag_t data structure internally to store NTP timestamps, but this value is represented using the timetag type in the python bindings. For example, the handler function called when a signal update is received contains a timetag argument. This argument indicates the time at which the source signal was sampled (in the case of sensor signals) or generated (in the case of sequenced or algorithimically-generated signals).

The update() function for output signals is overloaded; calling the function without a timetag argument will automatically label the outgoing signal update with the current time. In cases where the update should more properly be labeled with another time, this can be accomplished by simply adding the timetag as a second argument. This timestamp should only be overridden if your program has access to a more accurate measurement of the real time associated with the signal update, for example if you are writing a driver for an outboard sensor system that provides the sampling time.

Creating a new timetag without arguments causes it to be initialized with the current system time:

now = mapper.timetag()

Working with signal instances

libmapper also provides support for signals with multiple instances, for example:

The important qualities of signal instances in libmapper are:

All signals possess one instance by default. If you would like to reserve more instances you can use:


After reserving instances you can update a specific instance:

<sig>.update_instance(id, value)


<sig>.instance_update(id, value, timetag)

All of the arguments except one should be familiar from the documentation of update() presented earlier. The instance_id argument does not have to be considered as an array index - it can be any integer that is convenient for labelling your instance. libmapper will internally create a map from your id label to one of the preallocated instance structures.

Receiving instances

You might have noticed earlier that the handler function called when a signal update is received has a argument called id. Here is the function prototype again:

def frequency_handler(signal, id, value, timetag):

Under normal usage, the id argument will have a value (0 <= n <= num_instances) and can be used as an array index. Remember that you will need to reserve instances for your input signal using <sig>.reserve_instances() if you want to receive instance updates.

Instance Stealing

For handling cases in which the sender signal has more instances than the receiver signal, the instance allocation mode can be set for an input signal to set an action to take in case all allocated instances are in use and a previously unseen instance id is received. Use the function:


The argument mode can have one of the following values:

If you want to use another method for determining which active instance to release (e.g. the sound with the lowest volume), you can create an instance_event_handler for the signal and write the method yourself:

def my_handler(sig, id, event, timetag):
    # user code chooses which instance to release
    id = choose_instance_to_release(sig)
    sig.instance_release(id, timetag)

For this function to be called when instance stealing is necessary, we need to register it for mapper.IN_OVERFLOW events:

<sig>.set_instance_event_callback(my_handler, mapper.IN_OVERFLOW)

Publishing metadata

Things like device names, signal units, and ranges, are examples of metadata--information about the data you are exposing on the network.

libmapper also provides the ability to specify arbitrary extra metadata in the form of name-value pairs. These are not interpreted by libmapper in any way, but can be retrieved over the network. This can be used for instance to label a device with its loation, or to perhaps give a signal some property like "reliability", or some category like "light", "motor", "shaker", etc.

Some GUI could then use this information to display information about the network in an intelligent manner.

Any time there may be extra knowledge about a signal or device, it is a good idea to represent it by adding such properties, which can be of any OSC-compatible type. (So, numbers and strings, etc.)

The property interface is through the functions,

set_property(key, value)

where the value can any OSC-compatible type. This function can be called for devices or signals.

For example, to store a float indicating the X position of a device dev, you can call it like this:

dev.set_property("x", 12.5)

To specify a string property of a signal:

sig.set_property("sensingMethod", "resistive")

Reserved keys

You can use any property name not already reserved by libmapper.

Reserved keys for devices

description, host, id, is_local, libversion, name, num_incoming_maps, num_outgoing_maps, num_inputs, num_outputs, port, synced, value, version, user_data

Reserved keys for signals

description, direction, id, is_local, length, max, maximum, min, minimum, name, num_incoming_maps, num_instances, num_outgoing_maps, rate, type, unit, user_data

Reserved keys for links

description, id, is_local, num_maps

Reserved keys for maps

description, expression, id, is_local, mode, muted, num_destinations, num_sources, process_location, ready, status

Reserved keys for map slots

bound_max, bound_min, calibrating, causes_update, direction, length, maximum, minimum, num_instances, use_as_instance, type