[Scip] consistent "double free" segfault with fscip
Yuji Shinano
shinano at zib.de
Sat Oct 19 03:18:13 CEST 2013
Hi George,
I see, thank you very much for the good instance to debug.
Anyway, I have to know the bug and I am pleased to fix it.
FiberSCIP has a deterministic mode for debugging. I hope your instance
get the bug with the deterministic mode.
If I need your help, I will send you e-mail personally.
By the way, I think this top is not good for this mailing list,
this should be better to put bug-report.
I would like to recommend FiberSCIP and ParaSCIP usres
as follows in this opportunity:
1. I strongly recommend to use SCIP version 3.0.2 bug fix version released in this week.
Resetting of user interrupt flag was missing. This is a very small bug for SCIP users,
but it is a serious bug for FiberSCIP and ParaSCIP users. Current FiberSCIP and ParaSCIP
can interrupt a branch-and-bound node processing when the other solvers found a good feasible
solution. Therefore, if a sub-MIP, which leads to an optimal solution,
is assigned to such an interrupted solver, it may lose the optimal solution.
2 Please try to use SCIP first
FiberSCIP is not alway better even if it uses many cores.
You can see it in the following FiberSCIP paper.
http://opus4.kobv.de/opus4-zib/frontdoor/index/index/docId/4259
Best,
Yuji
(13/10/19 1:21), George Nychis wrote:
> Sure, sorry Yuji.
>
> On a 6 core machine with 16GB of memory, it often takes about 6 hours to segfault. Rarely does it
> segfault earlier than that. If you give me pointers on how to debug it, I could also try to build
> with some debug flags, etc.
>
> I will re-run it and try to capture the output. Previously, I was not saving this output.
>
> - George
>
>
> On Fri, Oct 18, 2013 at 4:02 PM, Yuji Shinano <shinano at zib.de <mailto:shinano at zib.de>> wrote:
>
> So, please send me your output until it causes the segmentation fault.
> How long did it take time to get it.
>
> I assumed that you compiled with opt mode. Then, I compiled with
> - Soplex: make PARASCIP=true
> - SCIP: make PARASCIP=true
> - UG: make
> Currently, it is running more than 12640 sec. without the segmentation fault.
> I need more information from you to debug it.
>
> Best,
> Yuji
>
>
> (13/10/18 22:40), George Nychis wrote:
>
> I definitely built with PARASCIP=true, and I've been using fscip for a while now with some
> different
> programs without an issue. This one in particular has been segfaulting for me, though. I
> believe
> that I did a standard make and built UG, SCIP, and Soplex will all of the options.
>
> Thanks for the help!
>
> - George
>
>
> On Fri, Oct 18, 2013 at 12:51 PM, Yuji Shinano <shinano at zib.de <mailto:shinano at zib.de>
> <mailto:shinano at zib.de <mailto:shinano at zib.de>>> wrote:
>
> Hi George,
>
>
> I've tried this on 5 different machines now, and my MIP keeps crashing with a
> segfault on a
> double
> free using fscip:
>
> *** glibc detected *** fscip: double free or corruption (!prev): 0x00007f84b57b2ae0 ***
>
>
> I've attached the MIP, you can run it on "obj_jainFairness.zpl"
>
>
> Did you use Soplex and SCIP compiled with "PARASCIP=true" option?
> Could you tell me how to build the binaries of Soplex, SCIP, UG (all options)?
>
> Best,
> Yuji
>
>
> ___________________________________________________
> Scip mailing list
> Scip at zib.de <mailto:Scip at zib.de> <mailto:Scip at zib.de <mailto:Scip at zib.de>>
> http://listserv.zib.de/____mailman/listinfo/scip
> <http://listserv.zib.de/__mailman/listinfo/scip>
> <http://listserv.zib.de/__mailman/listinfo/scip <http://listserv.zib.de/mailman/listinfo/scip>>
More information about the Scip
mailing list