test_pyrepl.test_repl_eio is flaky on some machines

[cmaloney]

Bug report

Bug description:

Flakiness reports: #135329 (comment) (Github CI UBSan build)
Test added in gh-135329

From that flakiness report:

test test_pyrepl failed -- Traceback (most recent call last):
  File "/home/runner/work/cpython/cpython/Lib/test/test_pyrepl/test_unix_console.py", line 390, in test_repl_eio
    _, err = proc.communicate(timeout=support.LONG_TIMEOUT)
             ~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/runner/work/cpython/cpython/Lib/subprocess.py", line 1273, in communicate
    stdout, stderr = self._communicate(input, endtime, timeout)
                     ~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/runner/work/cpython/cpython/Lib/subprocess.py", line 2293, in _communicate
    self._check_timeout(endtime, orig_timeout, stdout, stderr)
    ~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/runner/work/cpython/cpython/Lib/subprocess.py", line 1320, in _check_timeout
    raise TimeoutExpired(
    ...<2 lines>...
            stderr=b''.join(stderr_seq) if stderr_seq else None)
subprocess.TimeoutExpired: Command '['/home/runner/work/cpython/cpython/python', '-E', '-S', '/home/runner/work/cpython/cpython/Lib/test/test_pyrepl/eio_test_script.py']' timed out after 300.0 seconds

CPython versions tested on:

CPython main branch

Operating systems tested on:

No response

[colesbury]

Here's another failure (on 3.14) https://buildbot.python.org/#/builders/1802/builds/838/steps/6/logs/stdio

======================================================================
ERROR: test_repl_eio (test.test_pyrepl.test_unix_console.TestUnixConsoleEIOHandling.test_repl_eio)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/srv/buildbot/buildarea/3.14.bolen-ubuntu/build/Lib/test/test_pyrepl/test_unix_console.py", line 390, in test_repl_eio
    _, err = proc.communicate(timeout=support.LONG_TIMEOUT)
             ~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/srv/buildbot/buildarea/3.14.bolen-ubuntu/build/Lib/subprocess.py", line 1220, in communicate
    stdout, stderr = self._communicate(input, endtime, timeout)
                     ~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/srv/buildbot/buildarea/3.14.bolen-ubuntu/build/Lib/subprocess.py", line 2153, in _communicate
    self._check_timeout(endtime, orig_timeout, stdout, stderr)
    ~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/srv/buildbot/buildarea/3.14.bolen-ubuntu/build/Lib/subprocess.py", line 1267, in _check_timeout
    raise TimeoutExpired(
    ...<2 lines>...
            stderr=b''.join(stderr_seq) if stderr_seq else None)
subprocess.TimeoutExpired: Command '['/srv/buildbot/buildarea/3.14.bolen-ubuntu/build/python', '-E', '-S', '/srv/buildbot/buildarea/3.14.bolen-ubuntu/build/Lib/test/test_pyrepl/eio_test_script.py']' timed out after 300.0 seconds

----------------------------------------------------------------------

[colesbury]

Claude was able to get a consistent repro with:

diff --git a/Lib/test/test_pyrepl/eio_test_script.py b/Lib/test/test_pyrepl/eio_test_script.py
index e3ea6caef58..693aa6f3c81 100644
--- a/Lib/test/test_pyrepl/eio_test_script.py
+++ b/Lib/test/test_pyrepl/eio_test_script.py
@@ -5,6 +5,7 @@
 import signal
 import sys
 import termios
+import time


 def handler(sig, f):
@@ -39,6 +40,7 @@ def create_eio_condition():
                         sys.exit(0)
                 else:
                     # Back to child
+                    time.sleep(1)
                     try:
                         os.setpgid(grandchild_pid, grandchild_pid)
                     except ProcessLookupError:

[colesbury]

I don't understand the purpose of this test or test script. It seems like it runs Python without pyrepl:

        proc = script_helper.spawn_python(
            "-S", script_path,
            stderr=subprocess.PIPE,
            text=True
        )

[colesbury]

Can we delete this test? As far as I can tell, it's only testing that the kernel returns EIO when you read from a pty master whose slave is closed. It's not testing anything about CPython.

[vstinner]

I don't understand the purpose of this test or test script. It seems like it runs Python without pyrepl

The test comes from the fix for #135329. From what I understood, the issue was on tcsetattr(TCSADRAIN) and this function is called by _pyrepl.unix_console (code: tcsetattr(self.input_fd, termios.TCSADRAIN, state)).

Lib/test/test_pyrepl/eio_test_script.py creates a complex group of processes and a pair of PTY file descriptors (master_fd, slave_fd). The script expects EIO error on os.read(master_fd, 1).

The whole script is run in a subprocess by test_repl_eio() which sends SIGUSR1 signal to the script.

Well, I didn't understand everything, the code is quite complex :-(

[vstinner]

I agree to remove this complex test which doesn't use pyrepl code.