Make XpressSolver struct mutable for Alpine compatibility.

[hellemo]

I''m working on support for Xpress as mip solver for Alpine, and this change seems to be necessary given the way Alpine sets the settings for the subsolvers. Any reason to keep it the way it is?

[joaquimg]

I don´t see any specific reason. If cplex and gurobi are mutable, then xpress should also be, did you check them?

[mlubin]

Modifying internal fields of the solver structs is not really what was intended. The RawParameter proposal would provide a much cleaner way to set parameters on a solver that's already constructed (jump-dev/MathOptInterface.jl#387 (comment)).

[hellemo]

Thanks, I guess the RawParameter way is cleaner, but in the short/immediate term perhaps we could do like Cplex and Gurobi and make it mutable?

https://github.com/JuliaOpt/CPLEX.jl/search?q=AbstractMathProgSolver&unscoped_q=AbstractMathProgSolver
https://github.com/JuliaOpt/Gurobi.jl/search?q=abstractMathProgSolver&unscoped_q=abstractMathProgSolver

[blegat]

Let's make RawParameter part of MOI v0.9: jump-dev/MathOptInterface.jl#733 so that it can be added in the short term :)
It is already available in MosekTools, it will just be replaced by MOI.RawParameter: https://github.com/JuliaOpt/MosekTools.jl/blob/38f17977484ecaf381622ba32241377cde90c62c/src/MosekTools.jl#L166-L168

[hellemo]

That's great, but Alpine still uses MPB. Maybe I'm confused, but that's why I propose to do like Cplex and Gurobi for now, and do it the proper way with MOI later.

[mlubin]

I don't have an objection to the proposed change to the MPB interface.