Hi,

I've just gotten started with autodiff, so this may just be me not having learned the ropes yet: I'm finding it surprisingly difficult to convert the result of things I've computed to a double. The reason I would like to do this is largely because I'm interfacing with other libraries and/or want to provide an interface to my own library that at most relies on stl-containers, while I want to use autodiff internally for computations that rely on various derivatives.

For a dummy example:

double func(double x_, double y_){
    dual2nd x = x_;
    dual2nd y = y_;
    double returnval = (x * y).val; // Compilation error (struct <...> has no member val)
    double returnval = static_cast<double>(x * y); // Compilation error (invalid static cast)
    double returnval = static_cast<double>(static_cast<dual>(x * y)); // OK
    return returnval;

My question is: Is there a "canonical" way to convert the results of large expressions to double, or will I have to nest static_casts as appropriate wherever I need to do this?

Edit:
After some more playing around I've found that it seems like I can get around this by doing

double func(double x_, double y_){
   dual2nd x = x_;
   dual2nd y = y_;
   dual2nd d2returnval = (x * y);
   double returnval = static_cast<double>(d2returnval);
   return returnval;

which makes it seem a bit strange to me that static_cast<double>(x * y) would throw a compilation error? It definitely feels like there's something I'm not understanding about how things are handled internally here, as I'm getting a lot of error: cannot convert 'autodiff::detail::Dual<double, double>' to 'double' in assignment and similar compilation errors.