document behavior of calling atexit.register() while atexit._run_exitfuncs is running

[smontanaro]

BPO	22867
Nosy	@smontanaro, @vadmium

^{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 2014-11-13.20:52:37.164>
labels = ['docs']
title = 'document behavior of calling atexit.register() while atexit._run_exitfuncs is running'
updated_at = <Date 2015-07-21.07:43:20.900>
user = 'https://github.com/smontanaro'

bugs.python.org fields:

activity = <Date 2015-07-21.07:43:20.900>
actor = 'ethan.furman'
assignee = 'docs@python'
closed = False
closed_date = None
closer = None
components = ['Documentation']
creation = <Date 2014-11-13.20:52:37.164>
creator = 'skip.montanaro'
dependencies = []
files = []
hgrepos = []
issue_num = 22867
keywords = []
message_count = 2.0
messages = ['231135', '231354']
nosy_count = 3.0
nosy_names = ['skip.montanaro', 'docs@python', 'martin.panter']
pr_nums = []
priority = 'low'
resolution = None
stage = None
status = 'open'
superseder = None
type = None
url = 'https://bugs.python.org/issue22867'
versions = []

Linked PRs

• gh-67056: document that registering/unregistering an atexit func from within an atexit func is undefined #104473
• [3.11] gh-67056: document that registering/unregistering an atexit func from within an atexit func is undefined (GH-104473) #104500

[smontanaro]

A discussion on comp.lang.python about prettying up the "if __name__ == 'main__'" idiom led to a suggestion that a decorator could simple register the main function using atexit.register. That looks like it will work, but leaves open the possibility that while main() is running via atexit._run_exitfuncs, other exit functions might be registered.

As currently defined (at least in the Python version with 2.7), I think everything will work fine. Still, the behavior of adding new exit functions during exit is not defined. Would be kind of nice if this behavior was blessed, and then mentioned in the documentation.

[ethanfurman]

From a post by Ian Kelly (https://mail.python.org/pipermail/python-list/2014-November/681073.html)
--------------------------------------------------------------
In fact it seems the behavior does differ between Python 2.7 and Python 3.4:

$ cat testatexit.py
import atexit

@atexit.register
def main():
  atexit.register(goodbye)

@atexit.register
def goodbye():
  print("Goodbye")
$ python2 testatexit.py
Goodbye
Goodbye
$ python3 testatexit.py
Goodbye

[iritkatriel]

The atexit functions are invoked in LIFO order, so I think it's not a good idea to add more after the first one started running. We don't currently forbid, it and I wouldn't argue in favour of doing so, but we shouldn't encourage it.

[smontanaro]

This has been idle for "awhile." Should we just put our heads together and decide how best to document this behavior?

[iritkatriel]

I'm not sure what the behaviour is. Or rather, what behaviour we want to commit to.

[smontanaro]

I'm not sure that committing to any particular behavior is the right thing to do without some more discussion. Maybe just warn that calling atexit.register after _PyAtExit_Call has been invoked is undefined. There's no mention of that corner case in the 3.11 or 3.12 documentation as it stands.