Slime/Emacs Comint启动MPI进程时挂起问题求助

Problem Description #

I've written a simple MPI program to reproduce an issue:

#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 it successfully, I can launch the executable via mpirun without any issues in the SBCL REPL:

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

However, when I try to start the MPI process in Slime/Emacs's comint, it hangs indefinitely. I'm looking for help to fix this.

Possible Solutions & Debugging Steps #

1. Fix Terminal/TTY Allocation Issues #

MPI implementations often rely on proper terminal access to spawn and manage processes. Slime's comint buffer might not provide the full TTY environment mpirun expects. Try these tweaks:

• Add the --oversubscribe flag (common for OpenMPI) and explicitly redirect input/output to avoid interactive prompts:

  (uiop:run-program '("mpirun" "--oversubscribe" "-np" "10" "./hello_world") 
                    :input nil :output :string :error-output :string)
  

• For other MPI implementations, use flags like -quiet to suppress interactive behavior.

2. Align Environment Variables #

Slime might set different environment variables compared to your standalone SBCL REPL, which can break MPI. Check variables like MPI_HOME, PATH, LD_LIBRARY_PATH, and OMPI_MCA_* (for OpenMPI) in both contexts with:

;; Run this in both SBCL REPL and Slime comint
(uiop:print-environment-variables)

If you spot discrepancies, set the correct variables in your Emacs config or right before launching the MPI process.

3. Run MPI in Batch Mode #

Some MPI launchers struggle with interactive terminals. Try detaching the process from comint by redirecting output to a file and running it in the background:

(uiop:run-program '("mpirun" "-np" "10" "./hello_world" "> output.log" "2>&1" "&") 
                  :shell t)

You can later check the output with:

(uiop:read-file-string "output.log")

4. Use Pseudo-Terminal Allocation #

Slime's comint might be blocking process I/O. Try allocating a pseudo-terminal to mimic a real terminal environment:

(uiop:run-program '("mpirun" "-np" "10" "./hello_world") 
                  :pty t :output :string)

Note that not all Lisp implementations support the :pty option with uiop:run-program, so test this first.

5. Enable Verbose MPI Logging #

Turn on verbose output from mpirun to pinpoint where the hang occurs. For OpenMPI, add -v or --verbose:

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

The logs will show if MPI is failing to spawn processes, waiting for input, or hitting network/resource issues specific to the Slime environment.

内容的提问来源于stack exchange，提问作者ptb