Logged In: YES 
user_id=6656

This is deeply weird.

The most likely scenario that some *other* test (or combination of 
tests, sigh) is tickling a bug in glibc that test_ioctl is revealing.

Evidence for/against this could be acquired by adding '-r' to 
TESTOPTS and running make test a few times.  But I still don't 
understand why running regrtest.py from the shell passes.  Unless 
it's a Heisenbug, or just a flat out bug in the test.

Hmm.  Ten gets you one that it's test_fork1 that buggers it up.

It seems exceedingly unlikely that this points to a real problem in 
Python's ioctl code.  ioctl() is not the easiest thing in the world to 
write tests for...