I'll need help to reproduce them.

But temporarily add some extra debugging to the makefile to see if we
get a clue.

The cpython failure in test_subprocess.py definitely began with this
branch (which doesn't make much sense). The PyPy failure was happening
long before.