This is such a chronic problem with compiling any existing C or C++ code. I'm paranoid about touching any new 3rd party software in those languages because of the inevitable frustrating fragility of it all.
Python programs are many orders of magnitude more fragile than C/C++ programs, because I have seen plenty of such programs that require a certain Python 3.x version and they refuse to run with Python 3.(x-1) or 3.(x+1), making it difficult to install simultaneously many such programs, each desiring a different value of x.
Yes, you're correct that neither Python not C++ get this right, it's pretty sad that your conclusion is that therefore it's impossible.
One of the perverse things about human nature is that many people would rather believe nothing better is possible than go to any effort at all to improve things.
My experience with C/C+=/Fortran problems is not just versions of dependencies but version of compilers, platforms (they're non very portable), and mostly the extreme unhelpfulness of the error messages. It's always some undefined symbol that seems like it should be fine but turns out you're supposed to use all sorts of special command line parameters on the compiler and linker to make it work just right.