[Coin-osi-devel] FW: [Cgl] cgl, cpx, mir, error

[Matthew Galati]

> -----Original Message-----
> From: cgl-bounces at list.coin-or.org
> [mailto:cgl-bounces at list.coin-or.org] On Behalf Of Matthew Galati
> Sent: Sunday, April 15, 2007 7:54 PM
> To: Matthew Saltzman; cgl at list.coin-or.org
> Subject: RE: [Cgl] cgl, cpx, mir, error
> 
> 
> > >> I'll put together a basic empty OSI class this weekend and
> > see how it
> > >> works out with CGL.  I need it for my work anyway.  
> Then, you can
> > >> choose to include it or no with the current OSI distribution.
> > >
> > > 	Thanks, Matt! This is an idea that deserves a try-out.
> > 
> 
> 
> Here's a first draft. 
>    
> https://coral.ie.lehigh.edu/projects/DECOMP/browser/trunk/Temp
> Fix/OsiNull
> 
> As expected, this brings up a lot of design questions. This probably 
> needs more thought. In particular, do we want this object to manage 
> its own memory?
> 
> As this is just a "patch" until OSI2 is out, I took the quickest path, 
> which was to use Francois's idea of a data object that just carries 
> const pointers (i.e., OsiData is for the most part, just like 
> CglData). That means, the user is responsible for all memory 
> allocation and freeing. Also, the user must manipulate the data 
> themselves and only pass in pointers (which are const). This means, 
> that many of the Osi methods had to be disabled - like those involving 
> changing the model dynamically. The only non-const members are those 
> needed to make the Cgl generators work.
> 
> I did not derive all of the pure virtual methods -- just enough to get 
> some of Cgl to work. Methods that "should not be called" from OsiNull 
> will throw an exception/assertion using CoinAssert.
> 
> I have not tested it much - only in the context of my framework. It 
> seems to run fine with CGLs: Knapsack, Flow Cover, Clique and MIRs. It 
> throws an error with Gomory (as expected).
> 
> For now, it is good enough for my needs -- but, I am not sure how 
> useful it will be to others. Have a look, let me know what you think.
> 
> Matt G

FYI. If you are interested -- this has moved to:
	https://coral.ie.lehigh.edu/projects/DECOMP/browser/trunk/Decomp/src/OsiNullSolverInterface.hpp
      https://coral.ie.lehigh.edu/projects/DECOMP/browser/trunk/Decomp/src/OsiData.hpp