#117 Try out boost.python (High priority)

[Ulf Lorenz]

Using boost.python, it seems to be reasonably easy to produce a wrapper to import C++ code into python. On the plus side, this would hugely simplify the usability for end users: Instead of all the CMake workarounds, you simply build a python module and use python.

However, the documentation is lacking in many respects, so a few things need to be tried out. In particular, things that I can think of:

• can the tensor lib be made to work with boost::python?
• how does the whole ecosystem work with shared_ptr and other complex types?
• Does every class need a factory function then for the shared_ptr or what?
• What to do with callbacks? The user may want to write functions in python that need to be handled as functors in C++.
• does unit conversion work?
• any other uncommon issues: class enums, templatized types, ...?

Also, have a look at SWIG as alternative: http://swig.org

---

Resolution

Used pybind11 (boost::python alternative). Not everything has been solved (see ticket [#130] for not yet implemented features), but there is a working prototype including conversion of tensors into python data types (numpy.array) and back.