Possibly they are entirely working, but they are hard to debug because doing the numerical derivatives correctly is very tricky, especially shifting the mole fractions while keeping molar density constant. A new method might be needed for testing use only.
Fixes the compilation problems, I hope
Python part still needs to be fixed, almost there, compiles in clang again
Add some forceeval and type double where possible
Thought it would maybe help with compilation
But not clear why this changes anything...
The key point is to speed up compilation, and make it easier to develop, by using the packaged C++ interface more extensively