Make atexit state per interpreter

[vstinner]

BPO	42639
Nosy	@vstinner, @corona10, @shihai1991
PRs	#23763 #23770 #23771 #23699 #23683 #23777 #23779

^{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 = <Date 2020-12-15.16:17:11.642>
created_at = <Date 2020-12-14.12:00:33.149>
labels = ['library', '3.10']
title = 'Make atexit state per interpreter'
updated_at = <Date 2020-12-15.16:17:11.637>
user = 'https://github.com/vstinner'

bugs.python.org fields:

activity = <Date 2020-12-15.16:17:11.637>
actor = 'vstinner'
assignee = 'none'
closed = True
closed_date = <Date 2020-12-15.16:17:11.642>
closer = 'vstinner'
components = ['Library (Lib)']
creation = <Date 2020-12-14.12:00:33.149>
creator = 'vstinner'
dependencies = []
files = []
hgrepos = []
issue_num = 42639
keywords = ['patch']
message_count = 7.0
messages = ['382982', '383005', '383008', '383055', '383064', '383071', '383073']
nosy_count = 3.0
nosy_names = ['vstinner', 'corona10', 'shihai1991']
pr_nums = ['23763', '23770', '23771', '23699', '23683', '23777', '23779']
priority = 'normal'
resolution = 'fixed'
stage = 'resolved'
status = 'closed'
superseder = None
type = None
url = 'https://bugs.python.org/issue42639'
versions = ['Python 3.10']

[vstinner]

In Python 2.7, atexit was implemented in Python and registered itself using sys.exitfunc public attribute:
https://docs.python.org/2.7/library/sys.html#sys.exitfunc
https://docs.python.org/2.7/library/atexit.html#module-atexit

In Python 3.0, the atexit module was rewritten in C. A new private _Py_PyAtExit() function was added to set a new private global "pyexitfunc" variable: variable used by call_py_exitfuncs() called by Py_Finalize().

In Python 3.7, the global "pyexitfunc" variable was moved int _PyRuntimeState (commit 2ebc5ce), and then into PyInterpreterState (commit 776407f).

In Python 3.7, the atexit module was upgrade to the multiphase initialization API (PEP-489): PyInit_atexit() uses PyModuleDef_Init().

Since Python 2.7, the atexit module has a limitation: if a second instance is created, the new instance overrides the old one, and old registered callbacks are newer called.

One option is to disallow creating a second instance: see bpo-40600 and PR 23699 for that.

Another option is to move the atexit state (callbacks) into PyInterpreterState. Two atexit module instances would modify the same list of callbacks. In this issue, I propose to investigate this option.

[vstinner]

New changeset 83d5204 by Victor Stinner in branch 'master':
bpo-42639: Cleanup atexitmodule.c (GH-23770)
83d5204

[vstinner]

New changeset 357704c by Victor Stinner in branch 'master':
bpo-42639: atexit now logs callbacks exceptions (GH-23771)
357704c

[vstinner]

New changeset b8fa135 by Victor Stinner in branch 'master':
bpo-42639: Move atexit state to PyInterpreterState (GH-23763)
b8fa135

[vstinner]

New changeset f7049b5 by Victor Stinner in branch 'master':
bpo-42639: Add script_helper.run_test_script() (GH-23777)
f7049b5

[vstinner]

New changeset 3ca2b8f by Victor Stinner in branch 'master':
bpo-42639: atexit._run_exitfuncs() uses sys.unraisablehook (GH-23779)
3ca2b8f

[vstinner]

Ok, the initial issue is now fixed. It's possible to have more than one atexit module instance, registered callbacks in one or the other module are all called as expected. I also wrote an unit test for that.

I also took the opportunity of this issue to modernize the C code base, fix a bug in test_atexit (no longer clear atexit callbacks, run tests in subprocesses), and enhance atexit._run_exitfuncs(): log all callback exceptions using sys.unraisablehook so it's possible to catch them using the hook.