Closed kibaekkim closed 7 years ago
I agree with @kibaekkim
IMHO StructJuMP
would benefit from staying more at the modelling level. In the REQUIRE, it is said to need MPI and in the README, it is said to need PIPS and Ipopt while in StochasticDualDynamicProgramming, I use StructJuMP without needing any of MPI, PIPS or Ipopt.
I do not think that it is a big issue to require MPI (I may require it in https://github.com/blegat/StochasticDualDynamicProgramming
in the future anyway) but it may be clearer to move a part of the README to StructJuMPSolverInterface
if it is specific to that project and just mention in the README that StructJuMP is a modeling interface that is used by several project such as StructJuMPSolverInterface
and StochasticDualDynamicProgramming
and could also be used for many other different purposes (any modeling that need parent-children structure).
Of course, I might be misinterpreting the whole project so correct me if I am wrong.
I agree that the requirements are overly strict. I will try to take a stab this weekend at cleaning this up (along with some of the other open issues).
@joehuchette I have already removed a quite on my local. I can create a new branch and put my code there, if you give me an access to create a branch.
@kibaekkim, you don't need write permission for this repo, you can just make a fork
@mlubin I already have a folk that keeps a old version of StochJuMP that can still run without PIPS. And Github does not allow me to have two folks for the same project.
@kibaekkim, you should be able to push a new branch to your fork without disrupting what you already have there
@kibaekkim I believe you already have write access, can you test this out?
@mlubin Ah! that should work. @joehuchette Yes, now I can create a branch. Thanks!
I see you have create the remove-pips branch. Can it be merged ? I like the fact that MPI is removed from the REQUIRE file :)
@kibaekkim This can be closed now :)
@blegat Thanks!
StructJuMP
requires to install bothPIPS-NLP
andIpopt
. To me this is quite strict, becauseStructJuMP
potentially could support any solver. The solvers are used in StructJuMPSolverInterface, even where they seem not necessary. Moreover,MPI.jl
could also be optional inStructJuMP
.Can we relax these requirements (PIPS-NLP, Ipopt, and MPI.jl)?