Potential Contribution: Interface to CONOPT

[stephenjmaher]

Hi CasADi team and community,

I am interested in developing a formal interface for the CONOPT solver within CasADi. CONOPT is widely used in the GAMS ecosystem for large-scale non-linear programming, and I believe having a dedicated CasADi plugin would be a valuable addition for users handling sparse, high-dimensional NLPs. Especially since CONOPT provides an alternative solution approach, the generalized reduced gradient algorithm, to the interior point algorithms common in NLP solvers.

I have been digging into the source code for the IpoptInterface and SnoptInterface to understand the plugin architecture. My plan is to implement this by inheriting from the Nlpsol base class and ensuring compatibility with CasADi’s sparse matrix formats.

A few specific questions before I dive in:

1. Duplicate Work: Is anyone currently working on a CONOPT interface, or is there an existing (perhaps internal or older) branch I should be aware of to avoid duplicating effort?
2. JIT/Code Generation: I’ve noticed the importance of the CodeGenerator in the IPOPT interface for performance. Is it recommended to prioritize JIT support in the initial PR, or should I focus on a stable VM-based interface first?
3. Solver Binaries: Since CONOPT is a proprietary solver, I intend to follow the model used for SNOPT or KNITRO, where CasADi provides the interface/bridge and the user provides the shared library and license at runtime. Does this align with your current distribution strategy?
4. Implementation Tips: Are there any specific "gotchas" regarding CONOPT’s C-API or memory management that you’ve encountered with similar solvers?

I'm happy to share my progress via a draft Pull Request once the basic structure is in place. Looking forward to your thoughts and any guidance you can provide!