python3.11: failing test_log_messages_have_correct_metadata_echo

[arthurzam]

Describe the bug

On Gentoo we are currently testing and bumping packages for python 3.11 (currently in beta), during which we try to run the full test suite of package.
We noticed a failure in one single test test_log_messages_have_correct_metadata_echo in the file test/test_logging.py

To Reproduce

Running the following reproduces the test failure on python3.11:

`pytest -s -ra -l -vv test/engine/test_logging.py::TransactionContextLoggingTest::test_log_messages_have_correct_metadata_echo`

Same command for pypy3, python3.8, python3.9, python3.10 passes

Error

=============================================================================================== test session starts ===============================================================================================
platform linux -- Python 3.11.0b1, pytest-7.1.2, pluggy-1.0.0 -- /usr/bin/python3.11
cachedir: .pytest_cache
hypothesis profile 'default' -> database=DirectoryBasedExampleDatabase('/var/tmp/portage/dev-python/sqlalchemy-1.4.36/work/SQLAlchemy-1.4.36/.hypothesis/examples')
rootdir: /var/tmp/portage/dev-python/sqlalchemy-1.4.36/work/SQLAlchemy-1.4.36, configfile: setup.cfg
plugins: hypothesis-6.46.3, freezegun-0.4.2, asyncio-0.18.3
asyncio: mode=Mode.LEGACY
collected 1 item                                                                                                                                                                                                  

test/engine/test_logging.py::TransactionContextLoggingTest::test_log_messages_have_correct_metadata_echo 2022-05-14 14:58:49,640 INFO sqlalchemy.engine.Engine BEGIN (implicit; DBAPI should not BEGIN due to autocommit mode)
2022-05-14 14:58:49,640 INFO sqlalchemy.engine.Engine ROLLBACK using DBAPI connection.rollback(), DBAPI should ignore due to autocommit mode
FAILED

==================================================================================================== FAILURES =====================================================================================================
___________________________________________________________________ TransactionContextLoggingTest.test_log_messages_have_correct_metadata_echo ____________________________________________________________________
Traceback (most recent call last):
  File "/var/tmp/portage/dev-python/sqlalchemy-1.4.36/work/SQLAlchemy-1.4.36/test/engine/test_logging.py", line 788, in test_log_messages_have_correct_metadata_echo
    self._test_log_messages_have_correct_metadata(logging_engine)
    ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/var/tmp/portage/dev-python/sqlalchemy-1.4.36/work/SQLAlchemy-1.4.36/test/engine/test_logging.py", line 808, in _test_log_messages_have_correct_metadata
    assert any(
    ^^^^^^^^^^^
AssertionError: {'_log_info'}
assert False
 +  where False = any(<generator object TransactionContextLoggingTest._test_log_messages_have_correct_metadata.<locals>.<genexpr> at 0x7fdd269439f0>)
============================================================================================= short test summary info =============================================================================================
FAILED test/engine/test_logging.py::TransactionContextLoggingTest::test_log_messages_have_correct_metadata_echo - AssertionError: {'_log_info'}
================================================================================================ 1 failed in 0.10s ================================================================================================

Versions

• OS: Gentoo
• Python: 3.11.0b1
• SQLAlchemy: 1.4.36

Additional context

No response

[zzzeek]

hey there -

yes packagers testing on 3.11 are hitting some small issues and as yet we've not been able to test effectively on 3.11 due to the non-availability of greenlet and cython for python 3.11. Greenlet in particular. We hope to integrate 3.11 into our CI once we can get Greenlet to build (maybe it can now, wasn't working a couple of weeks ago).

from a packaging perspective I would add patchfiles to patch out this test for now.

[arthurzam]

@zzzeek Thanks for the reply.
This was the way we did for now (skipping the test). I know this is very early stages of python 3.11 and bugs are to be expected, and I wanted just to notify the awesome devs of sqlalchemy that there is a small issue.

[zzzeek]

we already hit this at Red Hat so I will likely have to run the tests in CI without the greenlet dependency (which is supported, just requires exception cases written out everywhere).

[zzzeek]

Ive pinged python-greenlet/greenlet#288 on this. I will try to seek adjusting tox to build without greenlet for py311

[sqla-tester]

Mike Bayer has proposed a fix for this issue in the main branch:

Install on py311 without greenlet; test support https://gerrit.sqlalchemy.org/c/sqlalchemy/sqlalchemy/+/3839

[zzzeek]

@arthurzam do you have a working greenlet build for py311? like a patchfile or something. If I could just put up a fork that builds it would make life a lot simpler on this end rather than skipping it everywhere

[zzzeek]

nm i got one from their PRs

[arthurzam]

@zzzeek Currently we skip greenlet on python3.11. You can see out build here.
We have issues with greenlet because of isn't working on arches like hppa or ia64, so we conditionally skip if it is missing

[sqla-tester]

Mike Bayer has proposed a fix for this issue in the rel_1_4 branch:

Install greenlet from a py311 compat patch https://gerrit.sqlalchemy.org/c/sqlalchemy/sqlalchemy/+/3841

[zzzeek]

to make it more fun, the issue is in py3.11.0b1 and not a4 which is what's on the CI machines, where we have a different failure, so having to upgrade the CI machines