Currently, we have 5 different representations for a MIP/LP problem and different stages of the solver pipeline use different representation. This leads to conversion bugs/overhead and code duplication, while also making recursive sub-MIP and restarts very cumbersome to implement.
Currently, we have 5 different representations for a MIP/LP problem and different stages of the solver pipeline use different representation. This leads to conversion bugs/overhead and code duplication, while also making recursive sub-MIP and restarts very cumbersome to implement.