I considered doing that, but I had already opened another issue (http://bugs.python.org/issue27145) in which I had proposed to replace that in-place negate in long_sub with a call to _PyLong_Negate.
But I guess I shouldn't worry about my patches colliding.

Anyway, the second assert would be 'assert(Py_SIZE(z) == 0 || Py_REFCNT(z) == 1);', because if someone does (in Python) 'x = (-2 ** PyLong_SHIFT) - (-2 ** PyLong_SHIFT)', x_sub would do 'return (PyLongObject *)PyLong_FromLong(0);'.

The updated diff file and the new test output are attached.

(No idea why test_netrc failed there. I ran it specifically five times right after that, and it passed all of them. Maybe some race condition? (To run the tests, I do 'python_d.exe -m test -j3'.)
Anyway, here is the relevant output (I am not sure the last line is relevant):

Warning -- files was modified by test_netrc
test test_netrc failed -- Traceback (most recent call last):
  File "C:\Users\orenmn\cpython\lib\test\test_netrc.py", line 52, in test_password_with_trailing_hash
    """, 'pass#')
  File "C:\Users\orenmn\cpython\lib\test\test_netrc.py", line 41, in _test_passwords
    nrc = self.make_nrc(nrc)
  File "C:\Users\orenmn\cpython\lib\test\test_netrc.py", line 13, in make_nrc
    with open(temp_filename, mode) as fp:
PermissionError: [Errno 13] Permission denied: '@test_3652_tmp'
minkernel\crts\ucrt\src\appcrt\lowio\write.cpp(49) : Assertion failed: (_osfile(fh) & FOPEN)

)