Python as scripting plug-in for large C++ project

[Daniel Dittmar]

Bjorn Pettersen wrote:

>  - We extensively (exclusively) use our own basic datatypes,
>    e.g. NInteger which in addition to keeping a value also
>    stores whether it is NULL (think databases). How much work
>    is it going to be to expose these to embedded Python?


When mapping from/to Python, translate NULL values to Py_None. Creating 
multiple full fledged numeric types is probably more trouble than it is 
worth.

 
>  - Our object/relational mapping tool (written in Python :-)
>    outputs ~100KLocs of C++ that defines an object tree that 
>    ends in timeseries data ("vectors") or records. The code
>    is highly optimized and makes sure we keep data integrity. 
>    Would it be simpler to use SWIG/SIP/boost to expose this, 
>    or would it be just as easy to have the mapping tool 
>    generate the Python glue directly.


Just a shot in the dark: having both C++ and Python representation of 
the same data in memory will probably make data integrity quite 
difficult, so expose your current data structures via SWIG etc. for 
production use. You might also think about generating the Python glue 
directly. This isn't really that difficult. It might actually be easier 
to generate this with your knowledge of the data structures rather than 
trying to sqeeze it into the SWIG etc. input.

Generating Python bindings directly will give you a great tool to play 
with new ideas, so I would probably (time permitting) add this one as well.


Daniel