Skip to content

signal 11 with infeasible LPs #1166

Closed

Description

@keithbriggs

I am getting a python crash with

Command terminated by signal 11

whenever I solve an infeasible LP problem. Actually, the solution appears to complete correctly, but the crash occurs when I try to return from my function. I tried lots of things, like deleting my model with

del scip

but that just triggers the crash anyway. So it must be caused by some internal memory mismanagement within pyscipopt.

The exact same code works perfectly when the LP is feasible.

Activity

  1. Joao-Dionisio commented on Jan 21, 2026

    @Joao-Dionisio
    Member

    Hey @keithbriggs ! Can you please share a minimal reproducible example?

  2. keithbriggs commented on Jan 21, 2026

    @keithbriggs
    Author

    Not easy - it's in the middle of a large code.

  3. Joao-Dionisio commented on Jan 21, 2026

    @Joao-Dionisio
    Member

    Can you then please tell me what you're trying to do with the LP you get, and potentially also share the LP?

  4. keithbriggs commented on Jan 21, 2026

    @keithbriggs
    Author

    I just want to determine feasibility, That is correctly being determined, but I can't return from my function. i should have said that I am using vtype='B'.

  5. Joao-Dionisio commented on Jan 21, 2026

    @Joao-Dionisio
    Member

    Alright. Maybe the issue is the same that PR #1161 is fixing, of variables and constraints not being properly freed upon model deletion. Assuming this is indeed the fix, it will only be properly available via PyPi once we add a new release, only after SCIP has a new minor release (shouldn't take more than a month). So, unfortunately, you would have to compile PySCIPOpt yourself.
    Local compilation of PySCIPOpt is done with

    git clone https://github.com/scipopt/PySCIPOpt.git
    cd PySCIPOpt
    git checkout fix-freeTransform-segfault
    export SCIPOPTDIR=your/SCIP/binary/directory # if not installed globally. if needed, you can download a precompiled packaged from https://www.scipopt.org/index.php#download
    pip install .

    Also, if feasibility is only what you're after, it can be more performant to use the following:

    from pyscipopt import Model, SCIP_PARAMEMPHASIS
    
    scip = Model()
    scip.setEmphasis(SCIP_PARAMEMPHASIS.FEASIBILITY)

    Please let me know if this fixes things, or if you're having problems with the compilation.

  6. keithbriggs commented on Jan 21, 2026

    @keithbriggs
    Author

    I was already doing SCIP_PARAMEMPHASIS.FEASIBILITY. I will try the local compile.

  7. keithbriggs commented on Jan 21, 2026

    @keithbriggs
    Author

    Trying to install scipoptsuite_10.0.0-1+trixie_amd64.deb on my Ubuntu 24 system, I get:
    scipoptsuite : Depends: libopenmpi40 (>= 5.0.7) but it is not installable

  8. Joao-Dionisio commented on Jan 21, 2026

    @Joao-Dionisio
    Member

    Strange... Maybe the easiest is to just do

      wget https://github.com/scipopt/scipoptsuite-deploy/releases/download/v0.10.0/libscip-linux.zip                                                                                                               
      unzip libscip-linux.zip                                                                                                                                                                                       
      export SCIPOPTDIR=$(pwd)/scip_install    
  9. keithbriggs commented on Jan 21, 2026

    @keithbriggs
    Author

    Thanks, that works and then the the pip install works.

    But the signal 11 problem is not fixed. It happens as soon as I try to do any manipulations on the model, or try to return from my function. Making the model global, which should mean that it is not deleted upon function return, still doesn't fix it.

    I made a small minimal example - but it did not show the problem. I probably have to make a bigger one.

  10. keithbriggs commented on Jan 21, 2026

    @keithbriggs
    Author

    Ok, I found that the bug is triggered by calling getSolTime(sol) in the infeasible case. When I take out this call, everything works.

  11. Joao-Dionisio commented on Jan 21, 2026

    @Joao-Dionisio
    Member

    Ah that makes sense, but it should still be fixed on our end. Thank you for the issue! I'll create a patch soon.

    Also, if you don't mind, I enjoyed browsing a bit through your website, quite interesting stuff :)

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions