There were some tough decisions to make. In my C# wrapper, I was sub-classing algorithm for handling messages from server (unzipping message content into data structure). Within this implementation, I made a decision to prefer straightforward simplicity, even it would mean a bit of data duplication. Personally, when using this kind of functionality, I would prefer to have just one single class (one header, one implementation), instead of having several classes.
Another big issue was concerning data output coming from Bloomberg server back to a client : how to pack all requested data into data structure to be easily used by a client ? In C# implementation we had dynamic data type, which enables us to create arrays of dynamic data types. However, in C++ there are no heterogeneous data structures available, besides std::tuple. Initially, I was considering to use tuple as result data structure, but I decided to reject the idea due to material increase in general complexity. I decided to return all possible data types from Bloomberg server as std::string within result data structure. It would be then up to a client to convert these values "back to their origins", by using std::stod or some other conversion method.
For the sake of clarity, I have created type definitions for several nested data structures used in the program : Vector [std::vector<std::string>], Matrix for all types of reference data [std::vector<Vector>] and Cube for historical time-series data [std::vector<Matrix>].
Finally, in order to "make it happen", a set of header files and libraries has to be configured into Visual Studio project. Extremely well-written and complete step-by-step tutorial is available in here.
Finally, thanks a lot for reading my blog.