Issue 2777: subprocess unit tests for kill, term and send_signal flaky

The unit tests for the kill, term and send_signal methods of the
subprocess.Popen object are still flaky and sometimes cause the test
suite to hang. I'm going to disable them for the upcoming alpha until
I've found a better solution.

What's the status of this?

I re-enabled the tests a few days ago and haven't seen any issues on
Windows or Linux. I attached a patch against r76222 which enables the
kill, term, and send_signal tests again, and also did some clean-up and
updating to the way tests are skipped.

I also changed the assertNotEqual at the end of the kill, term, and
send_signal tests to check against None instead of 0. I *think* it
should be checking against None there to make sure that the subprocess
isn't live anymore (where None would mean that it is).

Let me know what you think and if this needs anything else.

Minor patch change

The patch looks fine.
+1 to enable on trunk and follow the buildbots' behavior.

Patch based on Brian's patch, with additional cleanup, and use new unittest helpers: skipUnless, assertRaises.

And create a custom helper assertStderrEqual instead of "remove_stderr_debug_decorations" for debug builds.

Cleanup patch applied on r78508.
I will probably try to enable the tests on trunk.

I experienced this hang with Linux AMD64.

It occurs in test_send_signal.
It is because signal.SIGINT is not always handled (see #3137).

As a workaround, there's 2 choices: add a delay between Popen and send_signal, or retry SIGINT 2 or 3 times.

Windows should not be affected, because it uses signal SIGTERM in the test.

Re-enabled on r78662 and r78663.
Buildbots seems happy, except Windows XP + Cygwin buildbot:

======================================================================
FAIL: test_kill (test.test_subprocess.Win32ProcessTestCase)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "D:\cygwin\home\db3l\buildarea\trunk.bolen-windows\build\lib\test\test_subprocess.py", line 768, in test_kill
    self.assertNotEqual(p.wait(), 0)
AssertionError: 0 == 0

======================================================================
FAIL: test_send_signal (test.test_subprocess.Win32ProcessTestCase)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "D:\cygwin\home\db3l\buildarea\trunk.bolen-windows\build\lib\test\test_subprocess.py", line 761, in test_send_signal
    self.assertNotEqual(p.wait(), 0)
AssertionError: 0 == 0

======================================================================
FAIL: test_terminate (test.test_subprocess.Win32ProcessTestCase)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "D:\cygwin\home\db3l\buildarea\trunk.bolen-windows\build\lib\test\test_subprocess.py", line 775, in test_terminate
    self.assertNotEqual(p.wait(), 0)
AssertionError: 0 == 0

...and it hangs somewhere on ia64 Ubuntu.
http://www.python.org/dev/buildbot/all/builders/ia64%20Ubuntu%203.x/builds/557

All this flakiness is fixed:
 - r78736, r78759, r78761, r78767, r78788, r78789 on 2.x
 - r78797 on 3.x


Note: because of #3137, the send_signal(SIGINT) is retried 2 times on some platforms.

stage: patch review -> resolved

keywords: + needs review
stage: patch review

nosy: + brian.curtin
messages: + msg95157

keywords: + patch