ScalarNonlinearFunction is coming to MathOptInterface/JuMP. Among other benefits, this makes it easier to model constraint programming type problems, which tend to be expressed in terms of many different primitives/atoms. As a proof of concept, Oscar and I have been prototyping MiniZinc.jl on top of the MathOptInterface ScalarNonlinearFunction PR. I'll describe some challenges and decisions along the way. I'll also demonstrate some constraint programming examples.