Python to C++

Why would you want to!
Often people coming from a higher level language write better C++ code (using boost and STL) rather than people coming from C/Fortran who write at too low a level.

 

If you do a lot of intense computations such as CFD, it does not hurt to start programming in C++ which gives you a lot more return for your CPU seconds.
Python and other high level languages make excellent prototyping tools.

 

If you have an existing CFD (or whatever) library it might be worth looking for a Python wrapper, or making some wrappers using boost::python.
Let C++ (or Fortran) do the heavy lifting an write your own program on top of this in Python - you get a lot better value for your programming hours and a reduction in all-night bug hunts.

 

For what little it's worth, I've often found Boost::Python to be perhaps the most severely lacking component in the Boost libraries. (The reasons are technical and I won't discuss them here.)

On the other hand, Swig is a great way of wrapping C/C++ code so that it can be used with Python (among many other languages). It's discussed in several places but a really useful introduction to it, as well as to the use of Python more generally in a scientific context, is Langtangen's https://www.amazon.com/Python-Scrip...=sr_1_1?ie=UTF8&s=books&qid=1234739638&sr=1-1. (Langtangen's book is wonderful for many reasons but it also does a really good job of illustrating how Python can be used to interface with Fortran and C/C++ code, something that's essential knowledge if you're going to be working with scientific libraries.)

 

Thanks for the book pointer,
I've noticed a couple of projects switching back to SWIG. When I last used it I remember it generated a very 'c' like interface, I've only used boost:python for a few trial projects.

 

I really can't recommend it highly enough. It takes a bit of effort, but there's an enormous amount of useful information in there, particularly regarding Numpy and wrapping Fortran or C/C++ code in Python. It's a gem of a book.

 

I have Langtangen and agree it's excellent. I'd disagree w/ the recommendation of Swig now, especially for those starting out. Ctypes IMO has eclipsed it as the best bet: far easier to use, learn, and debug, comes builtin w/ Python, and for some things out performs Swig substantially. Downside is Ctypes won't handle C++