As a physicist and engineer, I mainly built tools for data analysis and Monte Carlo simulation. Some of them eventually reached the point where you start to wonder whether the stuff should be called tool, software or framework. C++ was an obvious choice for its performance and the libraries (Geant4 and Root where for instance mandatory at some points). I also wrote a lot of Perl for data management and some Python for what I pompously call "meta-analysis", which is only gluing together the results of long C++ analysis made on a distant farm in a reasonable number of plots.