.. _pde:

Building in-house libraries: C++
===================================================

.. contents::
   :depth: 3
   :local:
   :backlinks: entry


Input/output
----------------

The data structure that best fits our purposes is the Python dictionary.

The main idea is that Python dictionary is a very rich data structure, containing not only
the inputs, but also the *logic* required for setup and pricing.
The C++ library should be unaware of financial concepts.

This is a very flexible and powerful design, as the C++ library can be thin
(development  focused on the mathematical and numerical aspects),
require seldom updates, and focus on the "heavy lifting" - perform heavy 
computations.


See :ref:`squareCPP` for an example of communication between C++ and Python
via a simple Python dictionary.

The main question is how to handle the Python dictionary inside the C++ library.
We already mentioned two off-the-shelf solution.
It also makes sense to build a custom solution in-house, as we do not need the
most general interface. Two possible approaches are:

- Create a dictionary structure in C++ by building on top of the STL map.
  While no catch-it-all solution seem to exist (akin to Boost for smart pointers),
  there are plenty of tips and ideas on the internet (Stack Overflow, etc).

- Dissasemble the Python dictionary in the wrapper, like we did in :ref:`squareCPP`.
  This can be done at *run* time by automatically generating the hadcoded wrapper
  cpp file and finish building the library (by spawning a sub-process).

  

Flow chart
------------

This flow is part of the framework chart :ref:`frmchart`,
with extra details.

.. _pdechart:

.. figure:: pythonExtension.png
   :scale: 100%
   :align: center
   :alt: Pricing of Bermudan swaption

   Pricing of Bermudan swaption