Comment by noobermin

I think the only way I've avoided any of this is my academic work primarily deals with "write the performant, simulation codes in C++ or Fortran, write analysis in Python" (although others use matlab et al.) so everything parsing related has been through python, not C++, which ofc just has the "float()" type constructor. Generally, like most academics, I own my whole shop so fortunately no one uses my code other than one or two colleagues occasionally, so when bottlenecks arise I know where and what it is.