"setup.py doesn't use multiprocessing. multiprocessing is super complex. Would it be possible to write a reproducer which doesn't use multiprocessing?"

But the problem is with the handling of fork() by Python modules, and specifically with multi-threaded fork()s. Without multiple processes there is no issue.

And setup.py does use use multiple processes via a thread pool, where each thread calls fork(), if it detects that "make" was invoked with the parallel (-j) option.