How to enable python selectors with pytest on linux?

[AyushExel]

I have a working solution for user input timeout using selectors in linux/uninx and msvcrt in windows. I have written a test that works with windows version but throws PermissionError: [Errno 1] Operation not permitted with pytest on linux.

stdin_timeout

def posix_stdin_timeout(prompt="", timeout=DEFAULT_TIMEOUT, timeout_log=""):
    echo(prompt)
    sel = selectors.DefaultSelector()
    sel.register(sys.stdin, selectors.EVENT_READ)
    keys_and_events = sel.select(timeout)

    if keys_and_events:
        key, _ = keys_and_events[0]
        return key.fileobj.readline().rstrip(LF)
    else:
        echo(LF)
        termios.tcflush(sys.stdin, termios.TCIFLUSH)
        raise InputTimeoutError(timeout_log)

test

def test_timeout(pytestconfig):
    capmanager = pytestconfig.pluginmanager.getplugin("capturemanager")
    capmanager.suspend_global_capture(in_=True)

    timeout_log = "input timeout!"
    try:
        stdin_timeout("waiting for input", 1, timeout_log)
    except InputTimeoutError as e:
        assert e.message == timeout_log
    capmanager.resume_global_capture()

These functions work fine when running as python applications but running the above test with pytest throws this error:

self = <selectors.EpollSelector object at 0x7f0e8d116390>, fileobj = <_io.TextIOWrapper name=0 mode='r' encoding='UTF-8'>, events = 1, data = None

    def register(self, fileobj, events, data=None):
        key = super().register(fileobj, events, data)
        poller_events = 0
        if events & EVENT_READ:
            poller_events |= self._EVENT_READ
        if events & EVENT_WRITE:
            poller_events |= self._EVENT_WRITE
        try:
            self._selector.register(key.fd, poller_events)
E           PermissionError: [Errno 1] Operation not permitted

/home/.pyenv/versions/3.7.11/lib/python3.7/selectors.py:359: PermissionError

Is there a way I can make this test succeed?

[RonnyPfannschmidt]

selectors is directly incompatible for stdio capture - doesntit have a test mode?

[initial-d]

suspend_global_capture(in_=True) is not enough to make this a good unit test target on Linux. Under pytest, stdin is often not a normal interactive terminal fd, and epoll/selectors can fail on the object pytest has installed for capture. So the failure is about the test environment, not necessarily about your application code.

I would test this in two layers:

1. Unit-test the timeout logic without real stdin

   Put the selector creation behind a small seam, or pass in the file object/selector. Then test the branches with a fake selector:

def posix_stdin_timeout(prompt="", timeout=DEFAULT_TIMEOUT, timeout_log="", selector_factory=selectors.DefaultSelector, stdin=sys.stdin):
    echo(prompt)
    sel = selector_factory()
    sel.register(stdin, selectors.EVENT_READ)
    keys_and_events = sel.select(timeout)
    ...

Then your unit test can make select() return [] and assert InputTimeoutError without touching pytest's captured stdin at all.

2. Integration-test the real POSIX stdin behavior in a subprocess/pty

   If you really want to prove the Linux terminal behavior, run a tiny script under a pseudo-terminal using pty/pexpect or a subprocess with controlled stdin. That gives your code a real fd instead of pytest's capture object.

For example, the high-level shape is:

def test_timeout_logic_with_fake_selector():
    class FakeSelector:
        def register(self, fileobj, events):
            pass
        def select(self, timeout):
            return []

    with pytest.raises(InputTimeoutError):
        posix_stdin_timeout(timeout=0.01, selector_factory=FakeSelector, stdin=object())

The direct test against sys.stdin inside pytest will always be brittle because pytest intentionally owns stdio during capture. I would keep the real selector test outside capture, preferably in an integration test that creates its own terminal-like input source.