Ive been thinking about this stuff quite a bit for the last year or 2. I worked alot with C++ in the 90′s, developed dynamic type systems and multiple inheritance hierarchies (learned my lesson on that one), all kinds of things. after a while i thought it was too bulky and switched to python, and i was very happy since i felt i could program in 1 page what used to take me 10+ pages in c++. but python has the PEP and it’s rapidly enlarging also and i thought i could just take the python style back to c++ and get the speed, so that’s what i did on my most recent coding project. i thought, why not just forget the c++ object model and all that, it’s a tool to be used in certain appropriate situations, not in general. there is a bit of c++ overhead, but i think it worked mostly. so part of my conclusion is that the problem with c++ is not that it’s too big or inelegant – the compiler doesn’t force us to do anything more than c – but that we approach it with the c++ object mindset which only matches a few situations.
i think knuth had the right idea when he said that TeX would stop increasing and focus on reliability and bug fixes instead of enhancements; i believe he even had version numbers go as 3.1, 3.14, 3.141 etc as the program got closer to perfection. i’m also trying to learn to use his literate programming stuff since i think that was probably a much more effective direction than oop.
another link i found interesting by rob pike, who i think must be the old bell labs pike. i think in those days that was actually the right approach.
