Bug report: non-blocking collectives with user-defined operation gives segfault

test.c

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

typedef struct DensityEtc {
  double min;
  double max;
  double maxVel;
} DensityEtc;

void MpiOpUpdateFunc(void* invec, void* inoutvec, int *len, MPI_Datatype *datatype)
{
  const DensityEtc* iv = (const DensityEtc*)(invec);
  DensityEtc* iov = (DensityEtc*)(inoutvec);
  for (int i = 0; i < *len; ++i, ++iv, ++iov) {
    iov->min = iv->min < iov->min ? iv->min : iov->min;
    iov->max = iv->max > iov->max ? iv->max : iov->max;
    iov->maxVel = iv->maxVel > iov->maxVel ? iv->maxVel : iov->maxVel;
  }
}

int main(int argc, char* argv[])
{
  MPI_Init(&argc, &argv);
  
  MPI_Comm world = MPI_COMM_WORLD;
  int rank, size;
  MPI_Comm_rank(world, &rank);
  MPI_Comm_size(world, &size);

  MPI_Datatype density_t;
  MPI_Type_contiguous(3, MPI_DOUBLE, &density_t);
  MPI_Type_commit(&density_t);
  
  MPI_Op upDensity;
  MPI_Op_create(&MpiOpUpdateFunc, 1, &upDensity);
  
  FILE* log;
  if (rank == 0)
    log = fopen("log", "w");
   
  int a = rank, aMin, aMax;

  DensityEtc densLocal;
  densLocal.min = rank;
  densLocal.max = rank * 2;
  densLocal.maxVel = 0.5 * rank + 11;
  
  DensityEtc densGlobal;
  
  MPI_Request minReq, densReq;
  MPI_Comm minComm, densComm;
  
  MPI_Iallreduce(&densLocal, &densGlobal, 1, density_t, upDensity, world, &densReq);
  MPI_Wait(&densReq, MPI_STATUS_IGNORE);
  
  if (rank == 0) {
    fprintf(log, "Rank %d: min = %d\n", rank, aMin);
    fprintf(log, "Rank %d: min = %e, max = %e, maxVel = %e\n", rank, densGlobal.min, densGlobal.max, densGlobal.maxVel);
  }
  MPI_Finalize();
}