$ ./python -S Lib/test/test_site.py 
Traceback (most recent call last):
  File "Lib/test/test_site.py", line 28, in <module>
    raise unittest.SkipTest("importation of site.py suppressed")
unittest.case.SkipTest: importation of site.py suppressed

This counts as fail test. So when you execute the whole unit test:

$ ./python -S Lib/test

You'll get 2 fail tests.

The first one is test_trace, already fixed in #19398, but not yet committed.

The second one is this one, test_site.

'raise unittest.SkipTest("importation of site.py suppressed")' will make the file counts as fail test.

Attached the patch to make it does not count as fail test.

See also issue 1674555, which aims to make test_site run without -S and everything else run with -S.  I think this issue is invalid, if I understand what you wrote correctly, since test_site *should* be reported as a skipped test if -S is specified.

Before patch:
$ ./python -S Lib/test/test_site.py 
Traceback (most recent call last):
  File "Lib/test/test_site.py", line 28, in <module>
    raise unittest.SkipTest("importation of site.py suppressed")
unittest.case.SkipTest: importation of site.py suppressed

After patch:
$ ./python -S Lib/test/test_site.py 
sssssssssssssssssssss
----------------------------------------------------------------------
Ran 21 tests in 0.003s

OK (skipped=21)

With this patch, ./python -S Lib/test will report test_site as *skipped* not *failed*.

My grand plan is to make ./python -S Lib/test successful.

That's not relevant, though.  In fact, the existing error message for running the module directly is exactly correct: the entire test module fails to run, which is a more accurate reflection of the reality than showing all of the individual tests as skipped.  

If you run test_site under the test harness you also currently get the correct result:

rdmurray@hey:~/python/p34>./python -S Lib/test test_site
[1/1] test_site
test_site skipped -- importation of site.py suppressed
1 test skipped:
    test_site

In other words, you don't need to do anything to test_site to get it to work correctly for your desired goal, it already does.

Yeah, but...

$ ./python -S Lib/test
== CPython 3.4.0b1 (default:373797990f57, Dec 2 2013, 16:46:35) [GCC 4.7.2]
==   Linux-3.5.0-37-generic-x86_64-with-debian-wheezy-sid little-endian
==   hash algorithm: siphash24 64bit
==   /home/ethan/Documents/code/python/cpython3.4/build/test_python_28533
Testing with flags: sys.flags(debug=0, inspect=0, interactive=0, optimize=0, dont_write_bytecode=0, no_user_site=0, no_site=1, ignore_environment=0, verbose=0, bytes_warning=0, quiet=0, hash_randomization=1, isolated=0)
[  1/387] test_grammar
[  2/387] test_opcodes
[  3/387] test_dict
...
[385/387/2] test_zipimport
[386/387/2] test_zipimport_support
[387/387/2] test_zlib
361 tests OK.
2 tests failed:
    test_site test_trace
2 tests altered the execution environment:
    test_idle test_importlib
22 tests skipped:
    test_codecmaps_cn test_codecmaps_hk test_codecmaps_jp
    test_codecmaps_kr test_codecmaps_tw test_curses test_devpoll
    test_gdb test_kqueue test_msilib test_ossaudiodev test_smtpnet
    test_socketserver test_startfile test_timeout test_tk
    test_ttk_guionly test_urllibnet test_winreg test_winsound
    test_xmlrpc_net test_zipfile64

Feel free to close this as wontfix. This issue is not a life and death situation anyway. :)

Sorry to forget to add the relevant detail:

$ ./python -S Lib/test
....
[284/387] test_shutil
[285/387] test_signal
[286/387] test_site
test test_site failed -- multiple errors occurred; run in verbose mode for details
[287/387/1] test_slice
[288/387/1] test_smtpd
[289/387/1] test_smtplib
....

The real issue here is that the test used to determine whether -S was passed or not is outdated: instead of checking sys.flags.no_site, it checks whether 'site' is in sys.modules.  This is no longer a valid test, since site's side effects are contained within site.main, which is only run if no_site is False.  In fact, distutils imports symbols from site unconditionally, as do a couple of test modules; hence why test_site currently fails running `python -S Lib/test`: test_distutils (and others) run before test_site and cause site to be present in sys.modules.

I think we should move away from a toplevel SkipTest, though; some of the tests may be applicable whether -S is passed or not, though of course the ImportSideEffectTests would not be.

Looking more closely, there's not much that looks like it really should be tested if explicitly called with -S, so perhaps the toplevel skip is still best after all.  Any other thoughts?

Attached patch fixes `python -S Lib/test` and cleans up an unused import. I'm a bit iffy on the added comment.

I would just change the comment to "it is not useful to run the tests if we were called with -S".  Otherwise it looks fine.

If someone adds tests that are meaningful with -S is specified, we can put them in a different class, and move the skip to the class level.  You could mention that in the comment as well.

New changeset 40884256f8dd by Zachary Ware in branch '3.3':
Issue #19828: Fixed test_site when the whole suite is run with -S.
http://hg.python.org/cpython/rev/40884256f8dd

New changeset 90834ad91d56 by Zachary Ware in branch 'default':
Issue #19828: Merge with 3.3
http://hg.python.org/cpython/rev/90834ad91d56

Fixed.  Thank you Vajrasky for raising the issue, and David for the much better wording.