I'm afraid that the current comm_spawn system really can't run unless started via mpirun. I'm going to take a shot at fixing it in the near future and can let you know when that is accomplished, but it might take a little while before I get the time.
I've tried my application with MPICH2 and OpenMPI only. The readline stuff works in OpenMPI as long as I don't start my application with mpirun. It does not work in MPICH2 irrespective of whether I start my application with mpirun or not.
My application starts on a workstation and spawns slave processes on cluster nodes. The application provides an interactive command shell and provides some GUI functionality using X.
Do you know how bad is the MPI_Comm_spawn issue if I don't start my application via an mpirun?
On 7/11/06, Jeff Squyres (jsquyres) <firstname.lastname@example.org> wrote:
Sorry for not replying earlier.
It is likely that Open MPI won't work properly in this situation -- Open MPI provides basic I/O forwarding. I don't know if curses-based or readline-based applications will work through our forwarding.
Do you know if this application works with other MPI implementations?
Sent: Friday, July 07, 2006 12:46 PM
To: Open MPI Users
Subject: [OMPI users] readline and mpirun
When I run my MPI enabled application via mpirun, my application command-prompt looses its readline editing capability. I was wondering if there is workaround.
users mailing list