Import the 3.4 version of test_subprocess and make it (mostly) pass. Includes...

Import the 3.4 version of test_subprocess and make it (mostly) pass. Includes a test for #374; fixes #374.

.pep8

 [pep8]
 ignore=E702,E265,E402,E731,E266,E261,W503,E129
 max_line_length=160
-exclude=.tox,.git,build,2.6,2.7,2.7pypy,3.3,test_support.py,test_queue.py,patched_tests_setup.py,test_threading_2.py,lock_tests.py,_sslgte279.py
+exclude=.tox,.git,build,2.6,2.7,2.7pypy,3.3,test_support.py,test_queue.py,patched_tests_setup.py,test_threading_2.py,lock_tests.py,_sslgte279.py,3.4

changelog.rst

@@ -26,11 +26,17 @@ Unreleased
   Python 3, using a gevent SSL socket could cause the greenlet to
   block. See :issue:`597` by David Ford.
 - ``gevent.socket.socket.sendall`` supports arbitrary objects that
-  implement the buffer protocol (such as ctypes structurs), just like
+  implement the buffer protocol (such as ctypes structures), just like
   native sockets. Reported in :issue:`466` by tzickel.
 - Added support for the ``onerror`` attribute present in CFFI 1.2.0
   for better signal handling under PyPy. Thanks to Armin Rigo and Omer
   Katz. (See https://bitbucket.org/cffi/cffi/issue/152/handling-errors-from-signal-handlers-in)
+- The ``gevent.subprocess`` module is closer in behaviour to the
+  standard library under Python 3, at least on POSIX. The
+  ``pass_fds``, ``restore_signals``, and ``start_new_session``
+  arguments are still unimplemented.
+- An exception starting a child process with the ``gevent.subprocess``
+  module no longer leaks file descriptors. Reported in :pr:`374` by 陈小玉.
 
 1.1a1 (Jun 29, 2015)
 ====================

gevent/greenlet.py

@@ -451,15 +451,26 @@ def _kill(greenlet, exception, waiter):
 
 
 def joinall(greenlets, timeout=None, raise_error=False, count=None):
+    """
+    Wait for the ``greenlets`` to finish.
+
+    :param greenlets: A sequence of greenlets to wait for.
+    :keyword float timeout: If given, the maximum number of seconds to wait.
+    :return: A sequence of the greenlets that finished before the timeout (if any)
+        expired.
+    """
     if not raise_error:
-        wait(greenlets, timeout=timeout, count=count)
-    else:
-        for obj in iwait(greenlets, timeout=timeout, count=count):
-            if getattr(obj, 'exception', None) is not None:
-                if hasattr(obj, '_raise_exception'):
-                    obj._raise_exception()
-                else:
-                    raise obj.exception
+        return wait(greenlets, timeout=timeout, count=count)
+
+    done = []
+    for obj in iwait(greenlets, timeout=timeout, count=count):
+        if getattr(obj, 'exception', None) is not None:
+            if hasattr(obj, '_raise_exception'):
+                obj._raise_exception()
+            else:
+                raise obj.exception
+        done.append(obj)
+    return done
 
 
 def _killall3(greenlets, exception, waiter):

greentest/patched_tests_setup.py

@@ -197,6 +197,55 @@ if hasattr(sys, 'pypy_version_info'):
             # https://bitbucket.org/cffi/cffi/issue/152/handling-errors-from-signal-handlers-in
         ]
 
+if sys.version_info[:2] >= (3, 4):
+    disabled_tests += [
+        'test_subprocess.ProcessTestCase.test_threadsafe_wait',
+        # XXX: It seems that threading.Timer is not being greened properly, possibly
+        # due to a similar issue to what gevent.threading documents for normal threads.
+        # In any event, this test hangs forever
+
+        'test_subprocess.ProcessTestCase.test_io_buffered_by_default',
+        'test_subprocess.ProcessTestCase.test_io_unbuffered_works',
+        # These tests want to assert on the type of the class that implements
+        # `Popen.stdin`; we use a FileObject, but they expect different subclasses
+        # from the `io` module
+
+        'test_subprocess.POSIXProcessTestCase.test_terminate_dead',
+        'test_subprocess.POSIXProcessTestCase.test_send_signal_dead',
+        'test_subprocess.POSIXProcessTestCase.test_kill_dead',
+        # With our monkey patch in place,
+        # they fail because the process they're looking for has been allowed to exit.
+        # Our monkey patch waits for the process with a watcher and so detects
+        # the exit before the normal polling mechanism would
+
+        'test_subprocess.POSIXProcessTestCase.test_close_fds',
+        'test_subprocess.POSIXProcessTestCase.test_pass_fds',
+        'test_subprocess.POSIXProcessTestCase.test_pass_fds_inheritable',
+        # XXX: We don't implement the pass_fds option yet
+
+        'test_subprocess.POSIXProcessTestCase.test_restore_signals',
+        # XXX: We don't implement the restore_signals option yet
+
+        'test_subprocess.POSIXProcessTestCase.test_start_new_session',
+        # XXX: We don't implement the start_new_session option yet
+
+        'test_subprocess.POSIXProcessTestCase.test_exception_bad_args_0',
+        'test_subprocess.POSIXProcessTestCase.test_exception_bad_executable',
+        'test_subprocess.POSIXProcessTestCase.test_exception_cwd',
+        # These all want to inspect the string value of an exception raised
+        # by the exec() call in the child. The _posixsubprocess module arranges
+        # for better exception handling and printing than we do.
+
+        'test_subprocess.POSIXProcessTestCase.test_preexec_errpipe_does_not_double_close_pipes',
+        # Subclasses Popen, and overrides _execute_child. Expects things to be done
+        # in a particular order in an exception case, but we don't follow that
+        # exact order
+
+        'test_subprocess.POSIXProcessTestCase.test_small_errpipe_write_fd',
+        # Python 3 fixed a bug if the stdio file descriptors were closed;
+        # we still have that bug
+    ]
+
 # if 'signalfd' in os.environ.get('GEVENT_BACKEND', ''):
 #     # tests that don't interact well with signalfd
 #     disabled_tests.extend([