Generates Python Extension modules from Cython PXD files.

One important application of Cython is to wrap C++ classes for using them in Python. As Cythons syntax is quite similar to the syntax of Python writing a wrapper can be learned easily. Further Cython prevents you from many typical errors which might get in your way if you write such a wrapper in C++.

This wrapping process typically consist of four steps:

1. Rewrite parts of the header files of your C++ library in so called .pxd files. These give Cython information for calling the library and for error checking the code written in the following step.

2. Write Cython code which wraps the C++ library. This code resists in one or more .pyx files.

3. Translate these .pyx files to C++ code with Cython.

4. Use distutils to compile and link the C++ code to the final Python extension module.

Depending on the size of your library step 2 can be tedious and the code will consist of many similar code blocks with only minor differences.

This is where autowrap comes into play: autowrap replaces step 2 by analyzing the .pxd files with Cythons own parser and generating correct code for step 3. In order to steer and configure this process the .pxd files can be annotated using special formatted comments.

The main work which remains is writing the .pxd files. This is comparable to the declarations you have to provide if you use SIP or SWIG.

We assume that you installed autowrap already, so running

$ autowrap --help

does not fail.

Assuming you want to wrap the following C++ class

class IntHolder {
    public:
        int i_;
        IntHolder(int i): i_(i) { };
        IntHolder(const IntHolder & i): i_(i.i_) { };
        int add(const IntHolder & other) {
            return i_ + other.i_;
        }
};

you could create the following .pxd file

cdef extern from "int_holder.hpp":
    cdef cppclass IntHolder:
        int i_
        IntHolder(int i)
        IntHolder(IntHolder & i)
        int add(IntHolder o)

These files are already contained in the examples/ folder.

To create files .pyx and .cpp for wrapping the class IntHolder run

$ autowrap --out py_int_holder.pyx int_holder.pxd

inside the examplesfolder. This will generate files py_int_holder.pyx and py_int_holder.cpp which you can compile using the following file setup.py:

from distutils.core import setup, Extension
from Cython.Distutils import build_ext
import pkg_resources

data_dir = pkg_resources.resource_filename("autowrap", "data_files")

ext = Extension("py_int_holder",
                sources=['py_int_holder.cpp'],
                language="c++",
                include_dirs=[data_dir],
               )

setup(cmdclass={'build_ext':build_ext},
      name="py_int_holder",
      version="0.0.1",
      ext_modules=[ext]
     )

To get some insight how autowrap works, you can inspect files py_int_holder.pyx and py_int_holder.cpp.

Building the Python extension module now is done by running

$ python setup.py build_ext --inplace

This module can now be used as follows

>>> import py_int_holder
>>> ih = py_int_holder.IntHolder(42)
>>> print ih.i_
42
>>> print ih.add(ih)
84

Further docs can be found in docs/ folder.

• Wrapping of template classes with their public methods and attributes,enums, free functions and static methods.

• Included converters from Python data types to (many) STL containers and back. As this is version 0.2, not all STL containers are supported. We plan full support of nested STL containers.

• Manually written Cython code can be incorporated for wrapping code which autowrap can not handle (yet), and for enriching the API of the wrapped library. As this is done by writing Cython instead of C/C++ code, we get all benefits which Cython shows compared to C/C++.

  Writing a code generator for handling all thinkable APIs is hard, and results in a difficult and hard to understand code base. We prefer a maintainable code generator which handles 95% of all use cases, where the remaining 5% are still wrapped manually.

• For achieving a pythonic API, converters for library specific data types can be implemented easily. These converters are written in Python and Cython, not in C/C++ code using the C-API of CPython.

• autowrap relies on Cython, so we get automatic conversion of C++ exceptions to Python exceptions and wrapper code with correct reference counting. Using distutils we do not have to care to much about the build process on the targeted platform.

• Support for generating some special methods, as __getitem__, __copy__ and numerical comparison operators.

Many thanks go to:

• Hannes Roest, ETH Zürich, for contributing new ideas, patches, fruitful discussions and writing the first draft of this README.

• Lars Gustav Malmström, ETH Zürich, for getting the ball rolling.

• The developers of Cython for providing such a powerful and high quality tool.

• Thanks to https://github.com/hendrik-cliqz for implementing the "no-gil" annotation.