Wrapping: The Devil's Hobby

As Andrew posted yesterday, we have spent the better part of two weeks working on integrating our C block code into the python interface. The first step of this is supposed to be wrapping our C code so that it can be imported into python.

Now you'd expect that this process wouldn't be so bad since I hear its a common thing to do. I have no idea what kind of experience everyone else has with this, but I have worked out that only the wrapping code is what the devil has programmers do in hell.

I've managed to use SWIG (Simplified Wrapper and Interface Generator) to make wrapped files of some of my own C code. However, the point where I seem to fail is when I try to use a python setup file to compile a python module file that points to the wrapped code, enabling it to be called from python.

I've installed mingw and gcc in various ways (on their own, with ide's, etc.) and followed like 5 different example procedures and failed to get it working every time. I am now toying with the idea of getting visual studio which might be able to fix this problem. In the mean time there's a race between Andrew and I to find a way to make this work.

Until next time.

Wrapping for Python

Oi. I feel comfortable with the current functionality of the block parser, and have begun trying to wrap the C++ files for use in Python. This has not proved to be easy thus far. A quick once over reveals two libraries out there to accomplish this -- Boost and SWIG. I focused primarily on Boost while Roberto has been looking into SWIG.

Up unto this point I had been using Dev-C++ as my IDE of choice. Unfortunately, it hasn't been updated by Bloodshed since early 2005 and is based off of a, as of now, old build of MinGW and GCC. Trying to build Boost for MinGW with Jam has proved to be a less than pleasent experience. After several failed attempts, MinGW and GCC were updated to the latest stable releases with hopes that I would be able to successfully build the libraries without several hundred errors. After exploring with and fumbling around the build arguments, I finally managed to update all 8000+ files. (Of the several times that Boost has been built, it takes ~1.25 hours on a laptop or ~0.33 hours on a desktop). Despite now having a brand new set of header files and libraries even after proper linking, compiling a wrapper for a simple canned class still yields a plethora of errors. Boost looks like it would make everything so incredibly simple, if only it would work. Unfortunately, several other posts on the internet make note of it's less than intuitive setup necessities. I'm considering utilizing a very frequently updated distribution of MinGW that comes with several precompiled libraries (including Boost) that is available at www.nuwen.net. More to follow.