Cannot debug with gdb in WSL when MPI_InitFinalize is called

[yw5aj]

deal.II version: 8.5.1
WSL version: Ubuntu 18.04

(Google Groups discussion: https://groups.google.com/forum/#!topic/dealii/j0q7m6k6nNQ)

Dear all,

I am stuck in this remote debug issue - any pointer would be really appreciated.

I have set up Windows Subsystem Linux with Ubuntu 18.04 as instructed in https://github.com/dealii/dealii/wiki/Windows and it worked perfectly for tutorials without MPI, such as stpe-1 and step-6. The version for deal.II is 8.5.1.

However, when I use it for step-18, when I step into line 1894 where Utilities::MPI::MPI_InitFinalize mpi_initialization(argc, argv, 1); is called, the program hangs there and does nothing (no progress shows up in the Linux Console Window). Waiting for 30 minutes doesn't help.

Interestingly:

1. The compilation and linking worked fine; if I open Ubuntu terminal and run step-18 from there, everything works immediately;
2. Before when I used candi (rather than apt install) to install deal.II 9.0.0 it worked, making me think that whether I made a mistake somewhere in installing/upgrading the openmp related libraries;
3. The same workflow worked on my home computer but not my work computer - same Ubuntu version and deal.II version.

---

Update one day after my original post - I used gdb in the Ubuntu terminal:

gdb step-18
b main
r

And then either stepping into or over the line

1898          Utilities::MPI::MPI_InitFinalize mpi_initialization(argc, argv, 1);

Will hang the program.

I can neither step over nor step into that line - either way, it hangs there.

---

A quick update 7/13/2018 - just used candi to build and install all (including PETSc and deal.II 9.0.0) and the same problem still persists.

[drwells]

Thanks for reporting this.

Since GDB and WSL both intercept, in some way, system calls, I would estimate that they interact in some strange way that causes this problem.

I don't have windows 10 so I can only make guesses as to what is going on. Would you mind compiling and running

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

int main(int argc, char *argv[])
{
  MPI_Init(&argc, &argv);

  int rank = 0;
  MPI_Comm_rank (MPI_COMM_WORLD, &rank);
  if (rank == 2)
    puts("hello from rank 2");

  MPI_Finalize();
}

under GDB? If this hangs then we know that the problem is not with deal.II.

[masterleinad]

I can reproduce the issue and can report that @drwells' test code also hangs using gdb.

[drwells]

It sounds like this error should be reported upstream. @masterleinad since you can reproduce this, would you mind filing a WSL bug? If we do that then I believe we should document this bug in the wiki and then close the issue since there is no way for us to fix this.

[drwells]

Since @masterleinad can reproduce the issue and it is not dependent on deal.II, we should close this issue.

@masterleinad would you mind filling out a bug report at https://github.com/Microsoft/WSL?

[masterleinad]

Sure.

[masterleinad]

The bug report is in microsoft/WSL#3428.

[masterleinad]

Works with Version 10.0.17134.1 which is the April 2018 Update.

[drwells]

That is fortuitous. At least we followed up with Microsoft on it.