Python 3.11 crashes in some cases when loguru.logger.exception called

[gabus]

I'm sorry as I can't provide example but after digging into code the only change which is needed

formatter_record["exception"] = "".join(lines)
change to
formatter_record["exception"] = "".join(str(lines))
in loguru._handler.Handler.emit

it seems python 3.11 doesn't like implicit casting.
Though I'm second guessing my solution - don't know why it works..

[Delgan]

Thank you for the report. If you ever find out which exception reproduces the problem, let me know. I know there were different changes in exception handling in Python 3.11 which are not yet unit tested by Loguru, I'll try to find out what can cause this.

[gabus]

Managed to create a snippet which breaks

from loguru import logger
import falcon


class Ping:
    def on_get(self, req, resp):
        raise falcon.HTTPUnauthorized("test", "test unauthorised")


class ErrorHandler:
    def handle(self, req, resp, ex, params):
        logger.debug(ex)                   # works
        logger.exception(ex)               # dies here


def create_application():
    app = falcon.App()
    app.add_route('/ping', Ping())
    app.add_error_handler(falcon.HTTPError, ErrorHandler().handle)
    return app


application = create_application()

To build and run use

python3.11 -m venv venv
source venv/bin/activate
pip install pipenv

pipenv install loguru
pipenv install gunicorn
pipenv install falcon

venv/bin/gunicorn -p 8000 --reload --pythonpath ./ test_loguru

curl 'http://127.0.0.1:8000/ping'

[Delgan]

Thanks for the snippet, @gabus.

Unfortunately, it seems I'm unable to reproduce the issue.
Which versions of gunicorn and falcon are you using?
What is the observed output of the code you shared?

[gabus]

Interesting!
Can it be anything to do with macOS python version? I'm on Monterey 12.6 mac with M1 chip.

This is my setup:

Python 3.11.1

gunicorn==20.1.0
falcon==3.1.1
loguru==0.6.0
pipenv==2022.12.19
virtualenv==20.17.1

Output is very vague. It just shows that worker died and new one has been spawned. Then in separate pop up python crash window appears asking action: Ignore, Restart or Close Pyhton.

[2022-12-22 13:02:20 +0000] [53154] [WARNING] Worker with pid 53323 was terminated due to signal 11
[2022-12-22 13:02:20 +0000] [53339] [INFO] Booting worker with pid: 53339

[xzmeng]

I can reproduce it on macOS 13.1 with an M1 chip.

The process terminted with a segmentation violation when reading tb.tb_frame.

loguru/loguru/_better_exceptions.py

Line 182 in 16dbf79

frame = tb.tb_frame.f_back

[xzmeng]

I don't think it's a loguru's bug because accessing an instance attribute should never cause a core dump. (So maybe it's caused by Cython? I have tested with unCythonized falcon, and it works as expected.)

[Delgan]

Thanks @gabus and @xzmeng for the additional details.

I tested on Linux but have no access to MacOS computer unfortunately.

Are you both using Cython?
This recent pull request in python/cpython fixes a segfault in Python 3.11 when accessing frame.f_back, it does mention Cython as a environment where the bug happened: python/cpython#100182
I think this is what you faced @xzmeng.

[Delgan]

I'm closing this issue as it likely not a Loguru bug (as shown by ticket linked above). 👍