[MPICH] test/mpi/comm/cmfree.c

[Toon Knapen]

Rajeev Thakur wrote:
> Correction!  The test is ok as it is. The comm_split is on a different
> communicator. 


The communicator I am talking about is a communicator that is created 
directly after the MTest_Init as follows:

MPI_Comm_dup(MPI_COMM_WORLD,&comm);




> And a comm_free is called at the very end of the program if
> comm != null. 


I do not see this (at least not in mpich2-1.0.5p4). Additionally IMHO it 
would not make since to test it to be different from comm_null before 
comm_free'ing it because if this comm would be comm_null, the test would 
completely fail because this comm is used to determine the rank of every 
process etc which is used further down in the algorithm.


Actually, its a library of mine (that I link with the test) that 
detected the communicator-leak. I first thought it was an error of my 
library but finally I think that indeed there is a communicator-leak. 
But of course, I might be overseeing something.

Thanks again in advance,

toon