On trying to avoid making C++ code too obfuscated, and building a new high-performance virtual machine.

I am doing some hard-core C++ coding again after a long period of absence, where I was mostly using other languages like C# and Scheme. As frustrating as C++ programming can be, at least it is great for pedal to the metal coding.

The one thing I forget is how quickly complexity can overtake a few simple classes in C++. My lesson today was a reminder that my classes should do only one thing at a time: don't try to save a few lines of code and overload the responsibilities of your classes. To make sure I really learn my lesson I am paying the piper by refactoring my code. So far so good, but boy is it boring.

My current project is a bytecode interpreter for an intermediate language based on Cat http://www.ddj.com/architect/207200779 called CVML (Cat Virtual Machine Language). The CVML is intended as a proof of concept of how effective functional stack-based languages (concatenative languages) can be an intermediate language, and its feasibility as a high-performance intermediate language.

CVML is similar to a typical stack-based bytecode interpreter (e.g. a Java virtual machine) except locals and arguments are stored on a single shared stack, rather than a separate array of values. The other feature of CVML is that there are no labelled gotos, control flow is achieved through the use of higher-order functions, just like Cat. I am not sure what kinds of preconcieved ideas people have about this approach, so I would be interested in hearing some first impressions.

The whole reason for pursuing this is to demonstrate that this kind of form can be more easily optimized and compressed than your typical bytecode. I am also working on a paper for an upcoming conference. More on the progress in the coming weeks, but let me know if you have any thoughts or questions.