Please, note the documentation has been updated. The recommended way to integrate Loguru with pytest is to override the caplog fixture this way:

@pytest.fixture
def caplog(caplog):
    handler_id = logger.add(caplog.handler, format="{message}")
    yield caplog
    logger.remove(handler_id)

Using PropagateHandler could result in dead-lock if inadvertently combined with InterceptHandler see #474.

Thank you all for code samples, they’re incredibly useful. I just want to share my own, in case any of you find it useful:

# conftest.py
import logging

import pytest
from loguru import logger

@pytest.fixture
def caplog(caplog):
    """
    override and wrap the caplog fixture with one of our own
    """
    logger.remove() # remove default handler, if it exists
    logger.enable("") # enable all logs from all modules
    logging.addLevelName(5, "TRACE") # tell python logging how to interpret TRACE logs
    class PropogateHandler(logging.Handler):
        def emit(self, record):
            logging.getLogger(record.name).handle(record)
    logger.add(PropogateHandler(), format="{message} {extra}", level="TRACE") # shunt logs into the standard python logging machinery
    caplog.set_level(0) # Tell logging to handle all log levels
    yield caplog

Ah, I wasn’t aware the loguru didn’t use the stdlib logging module. I guess I could have checked that first…

Well, this is actually not stated explicitly anywhere as far as I know. Good catch, I should add a word about this.

Anyway, between the 3 I’m thinking the easiest one would be the 3rd option. I’ll see what I can come up with and, if I find something, submit a PR to update docs with the results.

Would fit pretty well in the documentation page about migrating from logging to loguru I think.

Do you think using the sample in the Readme would work for your tests?

class PropagateHandler(logging.Handler):
    def emit(self, record):
        logging.getLogger(record.name).handle(record)

logger.add(PropagateHandler(), format="{message}")

Please, note the documentation has been updated. The recommended way to integrate Loguru with pytest is to override the caplog fixture this way:

@pytest.fixture
def caplog(caplog):
    handler_id = logger.add(caplog.handler, format="{message}")
    yield caplog
    logger.remove(handler_id)

Using PropagateHandler could result in dead-lock if inadvertently combined with InterceptHandler see #474.

@Delgan Thank you so much! Your snippet helped me avoid deadlock!

Looks like adding this to conftest.py works:

import logging

import pytest
from _pytest.logging import caplog as _caplog

from loguru import logger

@pytest.fixture
def caplog(_caplog):
    class PropogateHandler(logging.Handler):
        def emit(self, record):
            logging.getLogger(record.name).handle(record)
    logger.add(PropogateHandler(), format="{message}")
    yield _caplog

Technically you don’t even need to add from _pytest.logging import caplog as _caplog and can just:

@pytest.fixture
def caplog(caplog):

but I really don’t like that naming collision.

I’ll write up some docs for it come Monday or Tuesday and submit the PR. Discussion can continue there.

What am I even trying to achieve?

It sounds like you’re just interested in having pytest capture log output on failures. Is that correct? If so, none of this PropogateHandler mumbo jumbo needs to be done - pytest will already capture loguru output for tests. The “Captured stderr call” section might not be formatted the same way, but I don’t know if that matters to you.

from loguru import logger

def test_foo():
    print("hallo")
    logger.debug("foo")
    assert False

$ pytest test_foo.py
============================================ test session starts =============================================
platform linux -- Python 3.6.8, pytest-6.1.2, py-1.9.0, pluggy-0.13.1
rootdir: /home/dthor/temp/loguru-pytest
collected 1 item

test_foo.py F                                                                                          [100%]

================================================== FAILURES ==================================================
__________________________________________________ test_foo __________________________________________________

    def test_foo():
        print("hallo")
        logger.debug("foo")
>       assert False
E       assert False

test_foo.py:6: AssertionError
-------------------------------------------- Captured stdout call --------------------------------------------
hallo
-------------------------------------------- Captured stderr call --------------------------------------------
2020-11-10 14:32:31.733 | DEBUG    | test_foo:test_foo:5 - foo
========================================== short test summary info ===========================================
FAILED test_foo.py::test_foo - assert False
============================================= 1 failed in 0.14s ==============================================

The core of this issue is specific to pytest’s caplog fixture, which you only need if you want to assert what’s being logged. Meaning, you need the PropogateHandler if you want to do this:

from loguru import logger

def test_foo(caplog):
    print("hallo")
    logger.debug("foo")
    assert "foo" in caplog.text"     # This won't work without the PropogateHandler hack

@glass-ships Please open a new issue with fully reproducible example if possible.

Hi @SomeAkk.

The catch() decorator does not propagate exceptions by default. It will simply create a logging record and send it to the handlers as any other logged message. You need to specify reraise=True if you want to be able to explicitly catch it with try / except or with pytest.raises().

However, a little hack is possible to achieve what you want. That is, having a behavior similar to reraise=False in production but being able to switch to reraise=True during testing. Since the message is sent to each configured handler, you can add an error_handler() sink that will be in charge of re-raising the error. Be careful, it must also be added with the parameter catch=False parameter because Loguru prevents otherwise the propagation of the error.

def error_handler(message):
    exc = message.record["exception"]
    if exc is not None:
        _, val, _ = exc
        raise val

# Add this during testing
logger.add(error_handler, catch=False)

Hi @blueyed.

Do you think it makes sense for loguru to ship a pytest plugin that would do this?

(I just came here from the docs, have not read up on it, but think it is possible, and would be willing to do it)

Thanks for your proposition. However, I don’t wish for Loguru to expose such plugin, the code snippet in the documentation is sufficient enough as a workaround.

That’s great, thanks @dougthor42!

I think you should maybe remove() the added sink at the end of each test. Based on your snippet, I’m wondering if this is not addind a new sink each time you run a test.

handler_id = logger.add(PropogateHandler(), format="{message}")
yield _caplog
logger.remove(handler_id)