Test for process.terminal fails

[k4rtik]

=========================== FAILURES ===========================
___________________________ test_process ____________________________

flush = None

    def test_process(flush):
        cpid = os.getpid()
        psu = psutil.Process(cpid)
        psl = P.get_process(cpid)
        assert psu.pid == psl.pid
        assert psu.ppid() == psl.ppid
        assert psu.name() == ffi.string(psl.name)
        assert psu.exe() == ffi.string(psl.exe)
        assert " ".join(psu.cmdline()) == ffi.string(psl.cmdline)
        assert psu.create_time() == psl.create_time
        uids = psu.uids()
        assert uids.real == psl.uid
        assert uids.effective == psl.euid
        assert uids.saved == psl.suid
        gids = psu.gids()
        assert gids.real == psl.gid
        assert gids.effective == psl.egid
        assert gids.saved == psl.sgid
        assert psu.username() == ffi.string(psl.username)
>       assert psu.terminal() == ffi.string(psl.terminal)
E       assert '/dev/pts/2' == '/dev/null'
E         - /dev/pts/2
E         + /dev/null

test_process.py:26: AssertionError
==================== 1 failed, 16 passed in 4.08 seconds ===================

[k4rtik]

Spent some time on this yesterday. The problem is using any of the default file descriptors (0, 1 or 2) for a process to obtain the terminal name (as done at nibrahim@e6f5eaf#diff-bf673c7ccb2de4138124fef0d870a1cdR210) is not reliable. I tried using fd 1 and 2 instead of 0 and instead of /dev/null I received paths inside of /tmp/ when py.test is invoked.

For the OS X port, I was able to use devname() syscall but that is specific to BSD systems. The only correct way to do this (that I could find) was to use the same approach as psutil (maintain a map for controlling terminal number with device names, obtain the number (key) from /proc/pid/stat and then query the map).

[k4rtik]

@nibrahim BTW, does the test not fail at your end?

[nibrahim]

I don't think so. I'll look at all the bugs you've filed early next week
and we can chat to decide direction.

Cordially,
Noufal
http://nibrahim.net.in

[k4rtik]

Confirming that this fails on my dept's regular Debian 8 machines as well.