When using the following command from an interactive shell:
bash -c "sleep 1001 & sleep 1002 & set -m; sleep 1003" the sleep 1003 command ends up in its own process group, which also includes the other two sleep commands. PID PPID PGRP SID CMD 190870 12953 190870 190870 bash 193199 190870 193199 190870 sleep 1003 193200 193199 193199 190870 sleep 1001 193201 193199 193199 190870 sleep 1002 This does not look as satisfying the description of "set -m" for "sleep 1003": "All processes run in a separate process group." It seems that the fork optimisation would have been correct (reusing the shell's own process group obtained from the interactive shell) if no other processes were in that process group (which might be difficult to check if sub-processes in the same process group would not be direct shell's children). It might be possible to create examples of commands preceding "set -m" that could create such detached processes and leave the shell's own process group unsuitable for the fork optimisation later. From GNU bash, version 5.1.8(1)-release (x86_64-redhat-linux-gnu) on RHEL 9. -- Daniel Villeneuve
