Manually debugging isolates the point in the dolfinx code at
mesh/graphbuild.cpp l.144
graph::AdjacencyList<std::int64_t> recvd_buffer
= dolfinx::MPI::all_to_all(comm, send_buffer);
https://salsa.debian.org/science-team/fenics/fenics-dolfinx/-/blob/experimental/cpp/dolfinx/mesh/graphbuild.cpp#L144
I think that supports the argument that the segfault is in openmpi
rather than dolfinx. Unless it's related to the use of int64_t when
handling the mpi communication buffers. I'll flag that for upstream
dolfinx review.
Drew