Add delay on error

[dschwoerer]

Sometimes the backtrace is not printed. I think it might be a race between flushing stdout and mpi_abort.

As this is racy, I am not sure how to test this, but think a delay of 0.1 seconds should not matter, and might help (at least sometimes).

[github-actions]

clang-tidy made some suggestions

[ZedThree]

I'm sure I've seen this behaviour too, but it's difficult to reproduce. From reading around, MPI implementations do seem to mess about with stdout's buffering, mostly to prevent interleaving output from ranks, and flushing might not be effective.

There are some options for OpenMPI's mpirun:

• --stream-buffering=0
• --output=RAW

But this workaround is probably fine -- do you know if it's possible to make the delay even smaller?

[dschwoerer]

I am not even sure the delay is sufficient. I mostly see this when other people complain that they see no backtrace.