The test sequence "test_ttk_guionly test_os test_pipes" hangs almost
every time when run on Gentoo x86 and Ubuntu x86_64 (at least).  Note
that this is without the gui resource, so the ttk tests aren't being run.

Commenting out test_closerange in TestInvalidFD in test_os clears the
hang.  Playing around with which module imports are commented out in
test_ttk_guionly can clear the hang, but test_pipes then produces the
following errors:

======================================================================
ERROR: testSimplePipe1 (test.test_pipes.SimplePipeTests)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/home/rdmurray/python/py3k/Lib/test/test_pipes.py", line 23, in
testSimplePipe1
    f = t.open(TESTFN, 'w')
  File "/home/rdmurray/python/py3k/Lib/pipes.py", line 148, in open
    return self.open_w(file)
  File "/home/rdmurray/python/py3k/Lib/pipes.py", line 168, in open_w
    return os.popen(cmd, 'w')
  File "/home/rdmurray/python/py3k/Lib/os.py", line 636, in popen
    bufsize=buffering)
  File "/home/rdmurray/python/py3k/Lib/subprocess.py", line 646, in __init__
    errread, errwrite)
  File "/home/rdmurray/python/py3k/Lib/subprocess.py", line 1138, in
_execute_child
    os.close(errpipe_read)
OSError: [Errno 9] Bad file descriptor

======================================================================
FAIL: testSimplePipe2 (test.test_pipes.SimplePipeTests)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/home/rdmurray/python/py3k/Lib/test/test_pipes.py", line 33, in
testSimplePipe2
    self.assertEqual(open(TESTFN2).read(), 'HELLO WORLD #2')
AssertionError: '' != 'HELLO WORLD #2'


The first of these appears almost every time, the second one seldom. 
Occasinally both tests pass.

It appears that TestInvalidFD.test_closerange is in fact closing a valid
file descriptor in this case.  I'm not sure how to fix the test...nor
why the close produced the results it does.

Here's another buildbot failure that's probably the same thing:

http://www.python.org/dev/buildbot/all/x86%20osx.5%203.x/builds/1220

Patch to skip test if it can't get a range of invalid file descriptors
to test.  The test sequence passes with this patch in place.

Oh, nice catch. That patch looks fine. However, I would probably add a
comment why the fstat checks are required.

Applied in r74171 to py3k and r74172 to 3.1.

It occurs to me that I should backport this to trunk, so I'm leaving the
issue open until I do that.

Backported to trunk in r74175, and 2.6 in r74176.