Long story short: there is a race condition between the safe signal handlers (which just update a flag) and
pp_fork()invokes the system
pp_fork()doesn't reset all the flags, so the signals could be dispatched in the child process, even if they were meant for the parent.
What boggles me is that this bug is easily over 10 years old. I find it hard to believe that no one noticed it before. Of course, it will take a while for a patch to make it into perl. And all the perls installed out their still have the bug.
The work around is to block signals, fork(), then unblock signals.
I'll write (or someone) IPC::SafeFork to do the above. But also draw attention to the issue.