Hessian sparsity pattern of DerivativesCppad

[omersahintas]

Thanks for developing this great toolbox! Besides its control functionalities, by utilizing eigen, cppad and cppadcg, the library serves as a perfect NLP framework!

I use the toolbox to solve an NLP. So far, I have inherited the DiscreteCostEvaluatorBase and DiscreteConstraintBase classes and filled the required values by calling DerivativesCppadJIT. As the next step, I wanted to do the same with DerivativesCppad and later by using generated C++ code.

I decided to start with autodiff and could fill eval(), evalGradient() and sparseHessianValues() methods. However, I didn't find any getSparsityPatternHessian() method for DerivativesCppad.

I checked cppad and apparently there is the ForSparseHes() function that can return the sparsity pattern. However, I couldn't understand arguments' corresponding variables in CT (please see this brach in my fork). I would appreciate if you could guide me on that.

Do you know if this function could fit? Filling the getSparsityPatternHessian() implementation is mandatory, as leaving it blank causes the solver to fail; my naive assumption on a dense Hessian matrix by default (i.e. all elements are filled) was wrong.