FWIW: the false positive was not really related to 1 or more CPUs, but because ls(1) now calls stat(STDOUT_FILENO, ...) before writing to stdout, and the ls invocation to create the reference count didn't output anything; hence the difference to the expected outcome. Discussed at: https://lists.gnu.org/r/coreutils/2021-04/msg00050.html Upstream patch: https://git.sv.gnu.org/cgit/coreutils.git/commit/?id=b7091093bb