Unreachable code in `before_log` reached

[zdebra]

Hello, I've reached code that is marked as unreachable in before_log function:

import logging

from tenacity import RetryError, Retrying, before_log

logging.basicConfig(level=logging.DEBUG, format="%(asctime)s - %(levelname)s - %(message)s")
logger = logging.getLogger(__name__)


def f():
    try:
        for attempt in Retrying(
            before=before_log(logger, logging.DEBUG),
        ):
            with attempt:
                raise Exception("hello")
    except RetryError as exc:
        exc.reraise()


f()

This code sample produces:

2025-01-28 15:11:42,725 - DEBUG - Starting call to '<unknown>', this is the 72232nd time calling it.
2025-01-28 15:11:42,725 - DEBUG - Starting call to '<unknown>', this is the 72233rd time calling it.
2025-01-28 15:11:42,725 - DEBUG - Starting call to '<unknown>', this is the 72234th time calling it.
2025-01-28 15:11:42,725 - DEBUG - Starting call to '<unknown>', this is the 72235th time calling it.
2025-01-28 15:11:42,725 - DEBUG - Starting call to '<unknown>', this is the 72236th time calling it.
2025-01-28 15:11:42,725 - DEBUG - Starting call to '<unknown>', this is the 72237th time calling it.

This is because retry_state.fn is probably not being set here.

Python 3.9.19 & tenacity 9.0.0.

[LotfiRafik]

This comment "# NOTE(sileht): can't really happen, but we must please mypy" is outdated. retry_state.fn can be None in the case of using retry inside context manager. its the expected behaviour not a bug.

[Mukller]

Root cause analysis

When Retrying is used as a context manager (the for attempt in Retrying(...): with attempt: pattern), __iter__ creates:

retry_state = RetryCallState(self, fn=None, args=(), kwargs={})

fn=None because there's no wrapped function — the user's code lives in the with attempt: block. Consequently, get_fn_name() falls through to:

return str(self.retry_object)  # → self._name or "<unknown>"

When no name= argument was passed to Retrying(...), self._name is None and __str__ returns "<unknown>". The comment # NOTE(sileht): can't really happen was valid only for the decorator path; it has been removed in the current codebase, but the underlying issue (unhelpful log output) remained.

Fix (PR linked above)

Inside __iter__, before the first yield, sys._getframe(1) refers to the frame that called next() on the generator — which is the for loop in the user's code. We capture f_code.co_qualname (Python ≥ 3.11) or f_code.co_name (older) and store it as retry_state._inferred_name.

get_fn_name() checks _inferred_name before falling back to "<unknown>", so before_log() now logs the actual enclosing function name:

# Before fix:
Starting call to '<unknown>', this is the 1st time calling it.

# After fix:
Starting call to 'my_module.my_function', this is the 1st time calling it.

Users who already pass name="my_name" are completely unaffected — that path is checked first.