Wow, I didn't realize that programming Python using the C interface was that tedious and verbose. I have attached a work-in-progress patch. It is not complete yet, but maybe somebody could already take a look to make sure that I'm not heading completely in the wrong direction.

Regarding StringIO.set_encoding(): I thought about this a bit, but I couldn't come up with a convincing scenario where having a no-op implementation would clearly help or or clearly hurt. Personally I would therefore to not provide it for now. If it turns out to be a problem, it can easily be added. But if we add it now and it turns out to have been a bad choice, it's probably much harder to remove it again.