testing/test_gateway.py::TestPopenGateway::test_waitclose_on_remote_killed FAILED/PASSEDSegmentation fault

• Bitbucket: https://bitbucket.org/hpk42/execnet/issue/53
• Originally reported by: @danstender
• Originally created at: 2017-03-19T18:17:04.424

We've got this test failure here on the Debian package. It appears randomly (however, oftenly), so it's hard to tell what triggers this, nor how to successfully reproduce it.

=================================== FAILURES ===================================
_______________ TestPopenGateway.test_waitclose_on_remote_killed _______________

self = <test_gateway.TestPopenGateway object at 0x7fbe7777a518>
makegateway = <bound method Group.makegateway of <Group ['gw0']>>

    def test_waitclose_on_remote_killed(self, makegateway):
        gw = makegateway('popen')
        channel = gw.remote_exec("""
                import os
                import time
                channel.send(os.getpid())
                time.sleep(100)
            """)
        remotepid = channel.receive()
        py.process.kill(remotepid)
        py.test.raises(EOFError, "channel.waitclose(TESTTIMEOUT)")
>       py.test.raises(IOError, channel.send, None)
E       Failed: DID NOT RAISE <class 'OSError'>

testing/test_gateway.py:251: Failed

Anyway, any hint on this or a fix would be very welcome.
The corresponding bug report is https://bugs.debian.org/854494, please see there for moreinfo and buildlogs.
DS

Start with testing/test_gateway.py::TestPopenGateway::test_waitclose_on_remote_killed and reproduce the failure repeatedly, using the linked Debian report and build logs for context. Done means the remote-killed case behaves consistently, raising EOFError from waitclose and IOError from channel.send without a segmentation fault.