[Os-project-managers] Fwd: Re: Fwd: [OS] OSrL Problem

[R. Kipp Martin]

-------- Original Message --------
Subject: Re: Fwd: [OS] OSrL Problem
Date: Tue, 11 Oct 2011 23:34:38 -0300
From: Horand Gassmann <Horand.Gassmann at DAL.CA>
To: R. Kipp Martin <kmartin at chicagobooth.edu>

"R. Kipp Martin" <kmartin at chicagobooth.edu> wrote:

> Hi Gus:
>
> This is really frustrating. I could swear we fixed this. I have to
> drive back to Chicago today. I know you are working like crazy on
> Visual Studio stuff, but if you get a chance could you look at it.

Hi Kipp,

hope you made it back safely.

I have looked at the problem and can at least diagnose it.

We send the problem to Clp and get back a solution of sorts, so we
enter writeResult(). However, since solver-solver->isProvenOptimal is
false, we never set cbasis in line OSCoinSolver.cpp:1072. But we have
defined it previously (line 1036), so cbasis now has garbage in it.
When we next access it, in line 1154, and switch on it in line 1158,
we fall straight through to the default case. This immediately throws
an error, which is caught in line 1505.

***Note that we do *not* delete rbasis in line 1516, nor cbasis in
line 1530, but we do NULL them both. Hence there is a memory leak
here.***

In line 1543 we set an honest osrl (that even has an objective value,
in my version of 142.85714285714286, set way back in line 1143, before
any signs of trouble).  The entire osrl string is then thrown as an
error in line 1544.
This error gets caught in line 904, where the repackaging takes place.

The short-term, band-aid solution is to use code there like in
OSSolverService.cpp:1225, but I think there are bigger issues that we
need to address more carefully. In the meantime I have committed the
changes as revision 4329.

Cheers

gus





> -------- Original Message --------
> Subject: [OS] OSrL Problem
> Date: Tue, 11 Oct 2011 16:47:30 +0200
> From: Mike Steglich <mike.steglich at berlin.de>
> To: <os at list.coin-or.org>
> CC: "Prof. Dr. Christian Müller" <christian.mueller at th-wildau.de>
>
> Hi,
>
> It seems there is a problem with the OSrL files if a linear problem
> is infeasible.
>
> I've changed one of the upper bounds of the
> <COIN-OS-HOME>/OS/data/osil/parincLinear.osil file to generate an
> infeasible problem.  (Row 24 -> <con name="sewing"  ub="0"/>)
>
>
>



-------------------------------------------------------

Horand I. Gassmann, Professor

School of Business Administration, Dalhousie University
6100 University Avenue
Halifax, Nova Scotia, Canada , B3H 3J5
ph. (902) 494-1844
fax (902) 494-1107

http://myweb.dal.ca/gassmann/