labwrap.py is a script for automatically generating wrappers for C/C++ code so that it can be called in common numerical packages such as Matlab or Python/Numpy. At the moment it generates only Matlab mex files. The code is a rough proof of concept coded at the CW15 Hack Day. More target platforms will come.
The general idea is to simplify the problem by focusing on the data types commonly used in such environments, and map them those of C++ libraries that provide similar functionality. At the moment only Eigen is supported.
The generation process starts by parsing the function prototypes from a header file. For the moment only plain C functions (with C++ types) are supported. It is unlikely that full object orientation will ever be added, so you are expected to provide a C interface. In addition, a mapping file for all used types needs to be provided. The initial example includes some Eigen classes. From the function specification and the mapping, a jinja template for the target platform (e.g a template of the MEX file) is populated.
Provided that you have a .h file with your function definitions and a python distribution with the jinja package, the following should work:
./labwrap.py path_to_header.h out_dir path_to_mapping.csv
To run the example:
./labwrap.py examples/exampleFunctions.h mex examples/types_eigen.csv
The script assumes that your output directory is at the same level than the directory where the code lives, so that the mex files will ../include your header path.
After generating the mex files, you can compile them on the command line or in matlab:
mex CXXFLAGS="-fPIC -std=c++0x -O3" matrixDouble.cpp ../examples/exampleFunctions.cpp
And call them in Matlab...
>> out= matrixDoubleMex(ones(1,2),3)
out =
3 3 3 3 3 3