Optimising with bounds only lead to instabilities

[icze]

Hi,

using the Cbc C interface (of Cbc 2.10.3) to solve a small model leads to an exception Corrupted start stamp for heap block. in void ClpModel::gutsOfDelete(int type) (called by ClpModel::~ClpModel()) when debugging my program. The method seem to try to delete an already deleted or not initalised ray:

delete[] ray_;

Here the model:

NAME          BLANK   
ROWS
 N  OBJROW
COLUMNS
    C0000001  OBJROW    0.0043124177
    C0000003  OBJROW    -0.00121674 
    C0000005  OBJROW    0.0013171039
    C0000007  OBJROW    0.0499413387
    C0000009  OBJROW    0.0138910224
    C0000013  OBJROW    0.0037925901
    C0000017  OBJROW    0.0199867314
    C0000019  OBJROW    -0.014416763
    C0000021  OBJROW    0.0508741733
RHS
BOUNDS
 FR BOUND     C0000001         1e+30
 FR BOUND     C0000003         1e+30
 FR BOUND     C0000005         1e+30
 FR BOUND     C0000007         1e+30
 FR BOUND     C0000009         1e+30
 FR BOUND     C0000013         1e+30
 FR BOUND     C0000017         1e+30
 FR BOUND     C0000019         1e+30
 FR BOUND     C0000021         1e+30
ENDATA

and the code:

  Cbc_Model* model = CBC_newModel();
  int error = CBC_readMps(model, "D:\\m.mps");
  if (error) {
    CBC_deleteModel(model);
    return;
  }
  error = CBC_solve(model);
  CBC_deleteModel(model);

Regards!
Ingo

[jjhforrest]

Ingo, Sorry I can't help as I am unable to reproduce the error on Linux either using stable or master. It might be worth doing a totally clean new build, just in case one of the dependencies in Windows is missing and so some functions were not recompiled when they should have been. John Forrest

…

On 30/09/2020 17:00, Ingo Czerwinski wrote: Hi, using the Cbc C interface to solve a small model leads to an exception |Corrupted start stamp for heap block.| in |void ClpModel::gutsOfDelete(int type)| (called by |ClpModel::~ClpModel()|) when debugging my program. The method seem to try to delete an already deleted or not initalised ray: |delete[] ray_; | Here the model: |NAME BLANK ROWS N OBJROW COLUMNS C0000001 OBJROW 0.0043124177 C0000003 OBJROW -0.00121674 C0000005 OBJROW 0.0013171039 C0000007 OBJROW 0.0499413387 C0000009 OBJROW 0.0138910224 C0000013 OBJROW 0.0037925901 C0000017 OBJROW 0.0199867314 C0000019 OBJROW -0.014416763 C0000021 OBJROW 0.0508741733 RHS BOUNDS FR BOUND C0000001 1e+30 FR BOUND C0000003 1e+30 FR BOUND C0000005 1e+30 FR BOUND C0000007 1e+30 FR BOUND C0000009 1e+30 FR BOUND C0000013 1e+30 FR BOUND C0000017 1e+30 FR BOUND C0000019 1e+30 FR BOUND C0000021 1e+30 ENDATA | and the code: |Cbc_Model* model = CBC_newModel(); int error = CBC_readMps(model, "D:\\m.mps"); if (error) { CBC_deleteModel(model); return; } error = CBC_solve(model); CBC_deleteModel(model); | Regards! Ingo — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#340>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ABWJYHEITX3H53CAVYNOZVLSINIY5ANCNFSM4R7I3IYQ>.

[icze]

Hi Mr. Forrest,

thanks for your help. I re-tested it on a newly build of Cbc 2.10.5 but problem is still there.
So, I started debugging and probably found the heap corruption:

Within the function ClpModel::emptyProblem() I found the following lines:

  if (returnCode == 2) {
    // create ray
    delete[] ray_;
    ray_ = new double[numberColumns_];
    CoinZeroN(ray_, numberColumns_);
    ray_[badColumn] = badValue;
  }

With the given example badColumn is -1 and therefore the heap is corrupted.

Regards!
Ingo

[odow]

Fixed in latest Clp: coin-or/Clp#142

[icze]

Thanks Oscar.

So, will the upcoming Cbc 2.10.6 use the new Clp 1.17.6 ?

[tkralphs]

Yes, the upcoming Cbc 2.10.6 use the Clp 1.17.6 (or whatever its latest release is by then), but I did want to also point out that coinbrew makes it easy to check out the latest compatible releases of dependencies to obviate the need for making releases of downstream projects solely for the purpose of incorporating changes in upstream projects. This is in concert with our effort to de-couple the projects in the Optimization Suite and in concert with the way package managers work generally. The incantation

coinbrew fetch Cbc --latest-release

should fetch Cbc 2.10.5, along with Clp 1.17.6. With that said, there should be a 2.10.6 release soonish.

[icze]

Hi Ted, thanks for your answer. I'll see if I get it run on Windows ...

[tkralphs]

Building on Windows is a bit more challenging, but if you follow the links to documentation in the README, there are pretty detailed instructions. You can also grab pre-built binaries on Bintray. I just kicked off a new build of 2.10, which should incorporate the fix. In a few hours, you can grab fresh binaries.