generating-idyntree-matlab-bindings.md

Generating Matlab bindings for iDynTree Classes

As per the documentation of iDynTree for modifications of Matlab/Octave bindings, the prerequisite is an experimental version of Swig with Matlab support. This can be found in the repository https://github.com/robotology-dependencies/swig/ (branch matlab).

Installing experimental version of Swig

• Clone the repository from https://github.com/robotology-dependencies/swig/
• Install the required dependencies for SWIG listed in the Required Tools section of the Swig page.
• Follow the installation instructions in http://www.swig.org/Doc3.0/SWIGDocumentation.html#Preface_unix_installation to install Swig locally. By default SWIG installs itself in /usr/local, in Unix environment. However, it is preferable to install Swig in a different location so as to avoid conflicts with other installed versions of SWIG.
• Proper installation of SWIG will generate a swig executable in <swig_install_dir>/bin.

This executable will be used to generate the iDynTree Matlab/Octave bindings for custom classes.

Including the custom classes for bindings generations

In this section, as a toy example, let us consider a custom class DiscreteKalmanFilterHelper for which the bindings need to be generated. Let's assume the class DiscreteKalmanFilterHelper exists in the header file iDynTree/Estimation/KalmanFilter.h.

• Take a glance at bindings/iDynTree.i. It follows a particular structure, adding the necessary header files twice for the desired classes, one using the prefix # and another using the prefix %.

  • It is also important to notice the order in which these headers are added, which ensures that the dependent header file is compiled after the dependee header file.
  • There are also sections for adding library specific data-structure; sensors-related, estimation-related, etc.
  • Templatized classes are explicitly specialized, so that the bindings are generated for the template specializations.

• Coming back to our toy example, we will add the header file in the section related to the Estimation related classes, in two different places

#include "iDynTree/Estimation/KalmanFilter.h"

and

%include "iDynTree/Estimation/KalmanFilter.h"

Now, all is left is to compile and generate the bindings.

Generating the bindings

• In the build folder of iDynTree, run ccmake to open the CMake Gui.
  • Set the CMake option SWIG_EXECUTABLE to <swig_install_dir>/bin/swig.
  • Set the CMake option IDYNTREE_GENERATE_MATLAB to ON.
  • Set the CMake option IDYNTREE_USES_MATLAB or IDYNTREE_USES_OCTAVE to ON.
• Having configured the CMake, run make and make install to install the bindings in the mex directory of the iDynTree install location.

NOTE: It is possible that you might face some compiler warnings or errors which needs to be handled properly for successful generation.

• Now, it is possible to use the iDynTree.DiscreteKalmanFilterHelper object in Matlab environment, if Matlab is configured to use the Matlab bindings as described in iDynTree/Bindings section.

• Additionally, you will notice in the source folder that the bindings for other classes (not only related to DiscreteKalmanFilterHelper) have been modified in iDynTree/bindings/matlab/autogenerated/+iDynTree/. This is because the generation of the bindings, modifies a certain function numbering in all of these files, and the class related methods in Matlab are called depending on this specific numbering.

Adding the generated bindings to the iDynTree Library

Usually, the generation of bindings is only allowed by the developers. So in order to create a pull request to add the custom generated bindings to iDynTree, we will follow some contribution instructions.

• Along with the modified bindings/iDynTree.i file, commit all the autogenerated binding files, not only the changes related specifically to your class. The latter might break the complete usage of the bindings due to the dependence on certain function numbering followed by SWIG.
• Keep only the bindings related changes in a single commit with the commit message [bindings] update matlab bindings, if you're generating matlab bindings. Similar message applies for python or octave bindings.
• Open a Pull request and wait for review!

Ending note - IMPORTANT

In this tutorial, we addressed a basic feature of generating bindings for a simple class. It must be noted that there are certain additional rules that need to be followed to generate bindings for templatized classes. Additionally, there are certain features that the SWIG does not support, for example, bindings cannot be generated for nested classes. In general, it is useful to check out the SWIG documentation for C++.

It might also be helpful to take a look into the files iDynTree/bindings/ignore.i, iDynTree/bindings/sensors.i, etc. to handle specific use cases. In particular, these are used to ignore certain methods for which the bindings need not be generated or to extend the functionality of certain classes using methods that needs to be explicitly generated as bindings only.