Search code examples
emacsmpicommon-lispsbclslime

Slime\Emacs comint hangs when launching mpi process

I have a simple mpi program to demonstrate my problem:

#include <stdio.h>
#include <mpi.h>

int main(int argc, char *argv[])
{
    int rank, csize;

    MPI_Init(&argc, &argv);
    MPI_Comm_rank(MPI_COMM_WORLD, &rank);
    MPI_Comm_size(MPI_COMM_WORLD, &csize);

    printf("Hello from rank[%d/%d]\n", rank, csize);

    MPI_Finalize();
}

After compiling, I can successfully launch the executable using mpirun from the sbcl repl:

* (uiop:run-program '("mpirun" "-np" "10" "./hello_world") :output :string)

"Hello from rank[7/10]
Hello from rank[9/10]
Hello from rank[5/10]
Hello from rank[8/10]
Hello from rank[0/10]
Hello from rank[1/10]
Hello from rank[2/10]
Hello from rank[3/10]
Hello from rank[4/10]
Hello from rank[6/10]
"
NIL
0

However, when I run the same from within slime, the slime repl simply hangs. If I run the executable directly, rather than through the mpirun launcher, then the everything runs fine:

CL-USER> (uiop:run-program '("./hello_world")
               :output :string)
"Hello from rank[0/1]
"
NIL
0

I'm using sbcl-1.4.5 and slime 2.20 on a linux workstation. Does anyone have a solution to this problem or a starting point of where to look?

UPDATE:

The problem stems from emacs comint mode, which slime is based on. I observer the same hanging behavior if I launch sbcl via make-comint-in-buffer and then use uiop:run-program.

UPDATE2:

After reading up a bit on comint mode, I was able to capture some output from the hanging process. This emacs lisp code:

(make-comint "foo" "mpirun" nil "-np" "1" "/home/ptb/programming/c/hello_world")

produces the follow error on the hung process:

[warn] Epoll MOD(1) on fd 14 failed.  Old events were 6; read change was 0 (none); write change was 2 (del): Bad file descriptor
[warn] Epoll MOD(4) on fd 14 failed.  Old events were 6; read change was 2 (del); write change was 0 (none): Bad file descriptor

Any ideas as to what this means?

Solution

  • I guess that this is a problem with redirected stdin/stdout in openmpi or libevent (there have been such problems in the past, e. g. bugzilla.redhat.com/show_bug.cgi?id=1235044). Which version of those do you use?