Can we avoid introducing this overload for setupGrid? If we did this for every method that took double*, we'd have a lot of extra boilerplate code, and I don't want to force users to use std::vector.

This is a little more than meets the eye. The current setupGrid(size_t n, const double* z) is an outlier as it takes the array size as an additional argument rather than just the pointer. I currently don't have a rule to handle that outlier in the logic of CLibSourceGenerator and am hesitant to add it as it's different from what we do elsewhere. We have the alternative of introducing a setupGrid(const double* z) and deprecate the current version to be consistent with the remainder of our API. (Edit: that won't work as different to other cases the size of the array isn't known a priori). I opted for the inherently safer version of a std::vector.

PPS: Fwiw, the signature setupGrid(size_t n, const double* z) with size followed by array pointer is consistent with how we handle things in CLib. This is about the decision whether we allow for an overload that can be handled by standard logic or we add an edge case.