[Cbc] Using SOS1 getting malloc: pointer being freed was not allocated

[John Forrest]

The bug was a bit more subtle this time.  The small B&B found a solution 
with all variable integer apart from one at 0.99999999215227509 and one 
at 9.3880305156922887e-07.  Later on it found the correct optimum - but 
threw
it away as it thought it was not as good.  At the end the incorrect 
optimistic solution was thrown away but that part of tree was not 
re-explored.

Initially I got an assert in a cut generator as an element value after 
preprocessing was > int_max! - so have added test in CglZeroHalf

I am changing code to stop doing small B&B at the first sign of trouble.
On 02/01/18 17:11, John Forrest wrote:
> Ryan,
>
> a) still bug with depth - will find and fix.
> b) I don't read the ?documentation? - if you do cbc and then depth?? 
> you will see that with no depth set the problem is thought to be too 
> big (> 500 variables).  It says to switch off then use depth -999.
>
> So my advice to set depth -100 was bad - it tried the small B&B at a 
> depth >200 - should be able to find bug.
>
> John
> On 02/01/18 13:59, Rhavar wrote:
>> I think you're right. There is probably some bugs in my code causing 
>> the models to be slightly different.
>>
>> But try this:  `cbc no-SOS.lp solve solution $`
>>
>> Result - Optimal solution found
>>
>> Objective value:81001582.00000000
>>
>>
>> But trying with `cbc no-SOS.lp depth -100 solve solution $`
>>
>> Result - Optimal solution found
>>
>> Objective value:81009832.00000000
>>
>>
>> Giving a worse solution. I wonder if that's related to the previous 
>> bug you fixed?
>>
>>
>> -Ryan
>>

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://list.coin-or.org/pipermail/cbc/attachments/20180105/80e958c6/attachment.html>