The test failed on a buildbot, see the message below. By the way, the test should use a value based on test.support.PIPE_MAX_SIZE rather than an hardcoded size of 2**20 bytes.

http://buildbot.python.org/all/builders/x86%20Windows%20Server%202008%20%5BSB%5D%203.x/builds/1661/steps/test/logs/stdio

test_communicate_epipe (test.test_subprocess.ProcessTestCase) ... ERROR
test_communicate_epipe_only_stdin (test.test_subprocess.ProcessTestCase) ... ok

======================================================================
ERROR: test_communicate_epipe (test.test_subprocess.ProcessTestCase)
----------------------------------------------------------------------

Traceback (most recent call last):
  File "E:\home\cpython\buildslave\x86\3.x.snakebite-win2k8r2sp1-x86\build\lib\test\test_subprocess.py", line 1095, in test_communicate_epipe
    p.communicate(b"x" * 2**20)
  File "E:\home\cpython\buildslave\x86\3.x.snakebite-win2k8r2sp1-x86\build\lib\subprocess.py", line 952, in communicate
    stdout, stderr = self._communicate(input, endtime, timeout)
  File "E:\home\cpython\buildslave\x86\3.x.snakebite-win2k8r2sp1-x86\build\lib\subprocess.py", line 1188, in _communicate
    self.stdin.write(input)
OSError: [Errno 22] Invalid argument

(...)
test_communicate_epipe (test.test_subprocess.ProcessTestCase) ... ok
test_communicate_epipe_only_stdin (test.test_subprocess.ProcessTestCase) ... ok

I don't know much about subprocess.py and I don't have access to Windows 8. But it looks like the kind of thing that might happen if the other end of the "pipe" is closed, which might happen if the subprocess exits early, either because it's just fast or because it's encountered some other bug.

If I add a sleep of 1 second before stdin.write(), the test always fail with OSError(22, "Invalid parameter"). So it becomes obvious that the write fails if the process already exited.

Attached patch should fix the issue. Can someone review it?

The test fails also on Python 3.3.

LGTM. Please commit in both repos and add to the list of cherrypicks for
Larry.

On Tue, Feb 18, 2014 at 1:40 AM, STINNER Victor <report@bugs.python.org>wrote:

STINNER Victor added the comment:

If I add a sleep of 1 second before stdin.write(), the test always fail
with OSError(22, "Invalid parameter"). So it becomes obvious that the write
fails if the process already exited.

Attached patch should fix the issue. Can someone review it?

The test fails also on Python 3.3.

----------
keywords: +patch
versions: +Python 3.3
Added file: http://bugs.python.org/file34131/communicate_epipe.patch

---

Python tracker <report@bugs.python.org>
<http://bugs.python.org/issue19612\>

---

New changeset 83013a7be836 by Victor Stinner in branch '3.3':
Issue bpo-19612: On Windows, subprocess.Popen.communicate() now ignores
http://hg.python.org/cpython/rev/83013a7be836

New changeset 423cb79640eb by Victor Stinner in branch 'default':
(Merge 3.3) Issue bpo-19612: On Windows, subprocess.Popen.communicate() now
http://hg.python.org/cpython/rev/423cb79640eb

LGTM.

Thanks for the review.

Please commit in both repos and add to the list of cherrypicks for
Larry.

Both repos? You mean Tulip? This issue is specific to subprocess not asyncio (asyncio.subprocess).

This sporadic bug is not new. IMO it exists since Python 3.3.

I'm not sure that it's need to hurry to fix it in Python 3.4.0.

Sorry, forget everything I said. (That includes the LGTM, sadly -- as I wrote earlier I don't know this code very well.)

Sorry, forget everything I said. (That includes the LGTM, sadly -- as I wrote earlier I don't know this code very well.)

Don't worry, I'm confident in my change :)

Oh, it looks like my patch is not enough: the test still fails sometimes :-/ It looks like the process was still running (poll() didn't return None) whereas stdin.write() got an OSError(errno.EINVAL).

http://buildbot.python.org/all/builders/x86%20Windows%20Server%202008%20%5BSB%5D%203.x/builds/2408/steps/test/logs/stdio

======================================================================
ERROR: test_communicate_epipe (test.test_subprocess.ProcessTestCase)
----------------------------------------------------------------------

Traceback (most recent call last):
  File "E:\home\cpython\buildslave\x86\3.x.snakebite-win2k8r2sp1-x86\build\lib\test\test_subprocess.py", line 1095, in test_communicate_epipe
    p.communicate(b"x" * 2**20)
  File "E:\home\cpython\buildslave\x86\3.x.snakebite-win2k8r2sp1-x86\build\lib\subprocess.py", line 951, in communicate
    stdout, stderr = self._communicate(input, endtime, timeout)
  File "E:\home\cpython\buildslave\x86\3.x.snakebite-win2k8r2sp1-x86\build\lib\subprocess.py", line 1187, in _communicate
    self.stdin.write(input)
OSError: [Errno 22] Invalid argument

New changeset 039ac3f01c4e by Victor Stinner in branch '2.7':
Issue bpo-19612: subprocess.communicate() now also ignores EINVAL when using at
http://hg.python.org/cpython/rev/039ac3f01c4e

I didnd't see the failure recently on Windows, so I close the issue.

I backported the fix on Python 2.7 for the issue bpo-21946.