Logger with a custom class breaks on copy

[govinda18]

BPO	46957
Nosy	@ericvsmith, @govinda18

^{Note: these values reflect the state of the issue at the time it was migrated and might not reflect the current state.}

Show more details

GitHub fields:

assignee = None
closed_at = None
created_at = <Date 2022-03-08.12:45:10.453>
labels = ['library', '3.9', 'type-crash']
title = 'Logger with a custom class breaks on copy'
updated_at = <Date 2022-03-09.07:11:41.061>
user = 'https://github.com/govinda18'

bugs.python.org fields:

activity = <Date 2022-03-09.07:11:41.061>
actor = 'eric.smith'
assignee = 'none'
closed = False
closed_date = None
closer = None
components = ['Library (Lib)']
creation = <Date 2022-03-08.12:45:10.453>
creator = 'govinda18'
dependencies = []
files = []
hgrepos = []
issue_num = 46957
keywords = []
message_count = 2.0
messages = ['414750', '414786']
nosy_count = 2.0
nosy_names = ['eric.smith', 'govinda18']
pr_nums = []
priority = 'normal'
resolution = None
stage = None
status = 'open'
superseder = None
type = 'crash'
url = 'https://bugs.python.org/issue46957'
versions = ['Python 3.9']

[govinda18]

The logging module allows you set a custom Logger class as the default logger. However, this breaks when we try to copy the logger which was created before setting the logger class

import copy
import logging

# Some module we don't control
oldLogger = logging.getLogger("abc")

# Some module where we want to change over to custom logging
class MyLogger(logging.Logger):
    pass

# Override the manager, root, etc., so everything uses our class
logging.setLoggerClass(MyLogger)
logging.root = root = MyLogger("", logging.WARNING)
logging.Logger.manager = logging.Manager(root)

newLogger = logging.getLogger("def")

# Later on this happens, which internally calls __reduce__
copy.deepcopy(oldLogger)

[ericvsmith]

In what way does it break? You haven't shown an error.

Why are you deepcopying the logger?

[serhiy-storchaka]

I see what happened.

Loggers are considered singletons and aren't supposed to be instantiated other than by calling logging.getLogger(name). They are pickled by name (like functions or classes). It fails if you get a different object by that name (same for functions and classes).

    def __reduce__(self):
        if getLogger(self.name) is not self:
            import pickle
            raise pickle.PicklingError('logger cannot be pickled')
        return getLogger, (self.name,)

See #74705.

The only logical behavior of copying the logger is returning the reference to the original logger. We can add __copy__ and __deepcopy__ which return self. "Copying" would not fail even for loggers not accessible by name. Will it solve the problem?

I am not sure that overriding instantiated loggers is supported strategy. @vsajip

[vsajip]

I am not sure that overriding instantiated loggers is supported strategy

No, it isn't. People sometimes use (and misuse) logging in odd ways for no particularly good reasons, so without knowing more about this specific use case, it's not yet clear whether it's worthy of any special support.