[Cbc] bug while solving MIPLIB lectsched-4-obj

[John Forrest]

Haroldo,

I have fixed lectsched-4-obj - I hope others are OK now.

The bug was in CglKnapsackCover.  I thought I had fixed a bug a month or 
so ago, but I had just moved it a bit.

Knapsack cuts are on constraints where all variables are integer, but 
you can relax continuous variables before trying.  More interestingly - 
if there is VUB constraint such as x +a*binary <= rhs, then you can 
replace x in the knapsack constraint.  You can also do the same thing if 
there is a VLB constraint on x.  The bug was when there were both 
constraints on x, when it could choose LB or UB but use some information 
from other test.  My first fix was incomplete, but I hope I have got it 
right now.  As it is uncommon to have both constraints, the bug was 
fairly rare - and often just meant a weak cut.

John
On 14/02/15 13:41, Haroldo Gambini Santos wrote:
> Hi,
>
> I have experienced some bugs with the latest CBC svn revision 2136.
>
> The one which appears to be most easily reproducible is while solving 
> the (feasible)  miplib instance   lectsched-4-obj
> which is declared infeasible by CBC.
>
> http://miplib.zib.de/miplib2010/lectsched-4-obj.php
>
> PS: the bug appears in more instances,  but it takes a lot of time to 
> appears.
>
> Cheers,
>
> Haroldo
>