https://lists.gnu.org/archive/html/bug-parallel/2024-06/msg00000.html

It should be possible to run the very first job with -u, with the following jobs buffered.

It gives the benefit of --lb but without the CPU penalty for the first job.

It will not be possible to do -u for the next jobs, but for the very first, it ought to be possible.

Challenges:

Does this live up to POLA?

The first job will act differently from the rest:

    doit() {
      echo "$@" stderr >&2
      echo "$@" stdout
      sleep 0.$1
      echo "$@" stderr >&2
      echo "$@" stdout
    }
    export -f doit
    # Order is different due to buffering
    parallel -u doit ::: 1
    parallel doit ::: 1

The first job will have a tty, but the next ones will not.

A workaround might be:

    parallel -k -u doit {= 'seq() > 1 and $opt::ungroup = 0' =} ::: 9 1 2 3