[Coin-discuss] Re: Coin-discuss digest, Vol 1 #39 - 1 msg

[Vivian De Smedt]

about your question about getInfinity my understanding is that it's an easy 
way to specify infinity as double wherever infinity is needed (instead of 
introducing more functions in the interface or a class that will model the 
closure of R). My guess is based on the CPLEX implementation of the 
OsiSolverInterface and on the helper functions already implemented that 
translates one part of the interface to an another part.

about your remark about documentation I totally agree because I faced 
exactly the same problem and maybe the OsiSolverInterface header could 
contains that information.

I have also work on an OSI interface to GLPK and maybe part of it could be 
useful to you or even to the a set of people that want to implement OSI 
interfaces for miscellaneous solver. Indeed many functions of the interface 
could be implemented in term of the others and I was wondering if we 
couldn't code a base classes that contains this kind code. In particular the
- writeMps  but also
- assignProblem, loadProblem, addCols, setContinuous, setColSetBounds, ...

at this stage the interface I have work on allows:

- the setup of the problem,
- its solving,
- the access to the solution and
- the input output from and to the .mps format
are implemented

but
- the hot start,
- the column or row cut and
- the branch and bound
aren't yet implemented.

the all isn't fully tested yet.

Tell me if you are interested by my code or part of it.

Vivian De Smedt.