[Clp] potential memory leak in CLP with CBC trunk

[Stefan Vigerske]

Hi,

valgrind should also tell you where the memory has been allocated that
has not been freed. Such an info would be very helpful, I think :-).
You would need to specify --leak-check=full --show-reachable=yes as
valgrind options.
Most info you get when the code is compiled in debug-mode.

Stefan

Kish Shen wrote:
> Hi,
> 
> I am now able to compile the latest trunk version of Cbc/Clp with gcc 
> 3.3 after my post about it yesterday. Thanks to Michael Hennebry and 
> John Forrest for their help!
> 
> I ran valgrind with this latest download, and I am still seeing the same 
> memory leak. I should have made clear in my previous message that I 
> carefully checked that delete was done for all the pointers to the COIN 
> solvers, in this case to OsiClpSolverInterface.
> 
> The only line of code I comment out between the two versions is the call 
> to initialSolve()
> 
> 	    lpd->lp->Solver->initialSolve();
> 
> where lpd->lp->Solver is a OsiClpSolverInterface *, which is allocated 
> with a new:
> 
>      lp->Solver = new OsiXxxSolverInterface();
> 
> (OsiXxxSolverInterface is typdef'ed to OsiClpSolverInterface for the 
> Cbc/Clp version here).
> 
> and Solver is deleted before the end of the program:
> 
> 	delete lp->Solver;
> 
> With the initialSolve() line, I get the following leak summary,
> 
> ==9960== LEAK SUMMARY:
> ==9960==    definitely lost: 0 bytes in 0 blocks.
> ==9960==      possibly lost: 0 bytes in 0 blocks.
> ==9960==    still reachable: 6,912 bytes in 3 blocks.
> ==9960==         suppressed: 0 bytes in 0 blocks.
> 
> but with the initialSolve() line commented out, I get:
> 
> ==9993== LEAK SUMMARY:
> ==9993==    definitely lost: 0 bytes in 0 blocks.
> ==9993==      possibly lost: 0 bytes in 0 blocks.
> ==9993==    still reachable: 1,280 bytes in 1 blocks.
> ==9993==         suppressed: 0 bytes in 0 blocks.
> 
> I don't quite understand why these blocks are shown as reachable -- 
> Solver, which is the only pointer in my code to the OsiClpSolver 
> interface, is deleted (but I assume the delete method does not delete 
> the 2 blocks that are allocated when initialSolve() is called). These 
> blocks do show up as definitely lost in the real code I am using (rather 
> than the simplified version I am compiling here).
> 
> Cheers,
> 
> Kish
> 
> 


-- 
Stefan Vigerske
Humboldt University Berlin, Numerical Mathematics
http://www.math.hu-berlin.de/~stefan