Improve stdlib module initialization error handling.

[brandtbucher]

BPO	38823
Nosy	@Yhg1s, @vstinner, @asvetlov, @serhiy-storchaka, @miss-islington, @brandtbucher, @shihai1991, @erlend-aasland
PRs	bpo-38823: Clean up refleaks in _asyncio initialization. #17195 [3.8] bpo-38823: Clean up refleaks in _asyncio initialization. (GH-17195) #17196 [3.7] bpo-38823: Clean up refleaks in _asyncio initialization. (GH-17195) #17197 bpo-38823: Clean up refleaks in _contextvars initialization. #17198 [3.7] bpo-38823: Clean up refleaks in _contextvars initialization. (GH-17198) #17199 [3.8] bpo-38823: Clean up refleaks in _contextvars initialization. (GH-17198) #17200 bpo-38823: Clean up refleaks in _tkinter initialization. #17206 bpo-38823: Clean up _statistics initialization. #17215 bpo-38823: Clean up _xxtestfuzz initialization. #17216 [3.8] bpo-38823: Clean up refleaks in _tkinter initialization. (GH-17206) #17226 [3.7] bpo-38823: Clean up refleaks in _tkinter initialization. (GH-17206) #17227 bpo-38823: Clean up refleak in _tracemalloc initialization. #17235 bpo-38823: Clean up refleak in fcntl initialization. #17236 bpo-38823: Clean up refleaks in faulthandler initialization. #17250 bpo-38823: Clean up refleak in marshal initialization. #17260 [3.8] bpo-38823: Fix refleaks in faulthandler init error path on Windows (GH-17250) #17263 [3.7] bpo-38823: Fix refleaks in faulthandler init error path on Windows (GH-17250) #17264 [3.8] bpo-38823: Fix refleak in marshal init error path (GH-17260) #17271 [3.7] bpo-38823: Fix refleak in marshal init error path (GH-17260) #17272 bpo-38823: Clean up refleaks in _ast initialization. #17276 [3.8] bpo-38823: Fix refleak in marshal init error path (GH-17260) #17280 [3.7] bpo-38823: Fix refleak in marshal init error path (GH-17260) #17281 [3.8] bpo-38823: Fix refleak in _tracemalloc init error handling (GH-17235) #17282 [3.7] bpo-38823: Fix refleak in _tracemalloc init error handling (GH-17235) #17283 bpo-38823: Add a private _PyModule_StealObject API. #17298 bpo-1635741: Calling Py_INCREF() after PyModule_AddObject() success to run #18365 bpo-38823: Fix refleaks in _ctypes extension init #23247 bpo-38823: Always build _ctypes with wchar_t #23248 bpo-38823: Fix compiler warning in _ctypes on Windows #23258

^{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 = 'https://github.com/brandtbucher'
closed_at = None
created_at = <Date 2019-11-16.18:54:25.878>
labels = ['3.7', '3.8', 'type-bug', 'library', '3.9']
title = 'Improve stdlib module initialization error handling.'
updated_at = <Date 2020-11-13.13:44:17.121>
user = 'https://github.com/brandtbucher'

bugs.python.org fields:

activity = <Date 2020-11-13.13:44:17.121>
actor = 'vstinner'
assignee = 'brandtbucher'
closed = False
closed_date = None
closer = None
components = ['Library (Lib)']
creation = <Date 2019-11-16.18:54:25.878>
creator = 'brandtbucher'
dependencies = []
files = []
hgrepos = []
issue_num = 38823
keywords = ['patch']
message_count = 30.0
messages = ['356764', '356765', '356773', '356774', '356775', '356781', '356782', '356783', '356809', '356820', '356872', '356873', '356874', '356984', '357002', '357003', '357004', '357014', '357019', '357029', '357044', '357046', '357047', '357050', '357051', '357114', '361486', '380822', '380824', '380883']
nosy_count = 8.0
nosy_names = ['twouters', 'vstinner', 'asvetlov', 'serhiy.storchaka', 'miss-islington', 'brandtbucher', 'shihai1991', 'erlendaasland']
pr_nums = ['17195', '17196', '17197', '17198', '17199', '17200', '17206', '17215', '17216', '17226', '17227', '17235', '17236', '17250', '17260', '17263', '17264', '17271', '17272', '17276', '17280', '17281', '17282', '17283', '17298', '18365', '23247', '23248', '23258']
priority = 'normal'
resolution = None
stage = 'patch review'
status = 'open'
superseder = None
type = 'behavior'
url = 'https://bugs.python.org/issue38823'
versions = ['Python 3.7', 'Python 3.8', 'Python 3.9']

• PR: gh-83004: Clean up refleak in _io initialisation #98840

• PR: gh-83004: Clean up refleak in _pickle initialisation #98841

• PR: gh-83004: Clean up refleak in _zoneinfo initialisation #98842

• PR: gh-83004: Clean up refleaks in _decimal initialisation #99043

• PR: gh-83004: Clean up refleaks in ssl, socket initialisation #99044

• PR: [3.11] gh-83004: Clean up refleak in _io initialisation (GH-98840) #99059

• PR: [3.10] gh-83004: Clean up refleak in _io initialisation (GH-98840) #99060

• PR: [3.11] gh-83004: Clean up refleak in _zoneinfo initialisation (GH-98842) #99163

• PR: [3.10] gh-83004: Clean up refleak in _zoneinfo initialisation (GH-98842) #99164

• PR: [3.11] gh-83004: Clean up refleak in _pickle initialisation (GH-98841) #99165

• PR: [3.10] gh-83004: Clean up refleak in _pickle initialisation (GH-98841) #99166

• PR: gh-83004: Harden _socket init #103261

• PR: gh-83004: Harden msvcrt init #103383

• PR: gh-83004: Harden winsound init #103385

• PR: gh-83004: Harden winreg init #103386

• PR: gh-83004: Harden msvcrt further #103420

[brandtbucher]

Many of the C stdlib modules can benefit from improved error handling during initialization. I've now had two PRs where the authors had reference leaks on error conditions, but defended their decisions by pointing to examples of similar idioms all over the stdlib.

The problems fall into two related categories, mostly:

• Not DECREF'ing the new module object on failure.
• Not handling errors raised by the PyModule_Add* family of functions... specifically, the weird steal-on-success semantics of PyModule_AddObject. I've already improved the docs for this, so we should see the issue less, but our own code should still be fixed.

I intend to turn this into a longer term project. I'll be working my way through these modules bit-by-bit over time, using this issue to track all of them (since there are a few dozen cases). I'd rather not make one huge one that spams all of the code owners and is impossible to review.

If anybody want to make themselves available to review/merge these as I go along, that would be great! Many of the ones I'll start with are just adding trivial DECREFs to the more popular modules (_asyncio, _contextvars, _functools, _random, _warnings, etc...).

[brandtbucher]

How do others feel about the creation of a new private API? It would keep these diffs smaller and ease refactoring... and it would probably be good to have around anyways:

/* Like PyModule_AddObject, but steals o on success AND failure. */

int
_PyModule_StealObject(PyObject *m, const char *name, PyObject *o)
{
    if (PyModule_AddObject(m, name, o) < 0) {
        Py_XDECREF(o);
        return -1;
    }
    return 0;
}

[miss-islington]

New changeset c3f6bdc by Miss Islington (bot) (Brandt Bucher) in branch 'master':
bpo-38823: Clean up refleaks in _asyncio initialization. (GH-17195)
c3f6bdc

[miss-islington]

New changeset 48f4f75 by Miss Islington (bot) in branch '3.8':
bpo-38823: Clean up refleaks in _asyncio initialization. (GH-17195)
48f4f75

[miss-islington]

New changeset 825e91b by Miss Islington (bot) in branch '3.7':
bpo-38823: Clean up refleaks in _asyncio initialization. (GH-17195)
825e91b

[miss-islington]

New changeset 143a97f by Miss Islington (bot) (Brandt Bucher) in branch 'master':
bpo-38823: Clean up refleaks in _contextvars initialization. (GH-17198)
143a97f

[miss-islington]

New changeset 8a334af by Miss Islington (bot) in branch '3.7':
bpo-38823: Clean up refleaks in _contextvars initialization. (GH-17198)
8a334af

[miss-islington]

New changeset 1fe79a4 by Miss Islington (bot) in branch '3.8':
bpo-38823: Clean up refleaks in _contextvars initialization. (GH-17198)
1fe79a4

[asvetlov]

Is anything left to be done for the issue?

[brandtbucher]

Yes, there are still a few dozen modules I plan to update!

[miss-islington]

New changeset 289cf0f by Miss Islington (bot) (Brandt Bucher) in branch 'master':
bpo-38823: Clean up refleaks in _tkinter initialization. (GH-17206)
289cf0f

[miss-islington]

New changeset 9e4d031 by Miss Islington (bot) in branch '3.7':
bpo-38823: Clean up refleaks in _tkinter initialization. (GH-17206)
9e4d031

[miss-islington]

New changeset 42a4359 by Miss Islington (bot) in branch '3.8':
bpo-38823: Clean up refleaks in _tkinter initialization. (GH-17206)
42a4359

[Yhg1s]

New changeset 54b32c9 by T. Wouters (Brandt Bucher) in branch 'master':
bpo-38823: Clean up refleak in fcntl module initialization. (GH-17236)
54b32c9

[vstinner]

New changeset ac22354 by Victor Stinner (Brandt Bucher) in branch 'master':
bpo-38823: Fix refleaks in faulthandler init error path on Windows (GH-17250)
ac22354

[miss-islington]

New changeset 5bd2af9 by Miss Islington (bot) in branch '3.7':
bpo-38823: Fix refleaks in faulthandler init error path on Windows (GH-17250)
5bd2af9

[miss-islington]

New changeset a5ed2fe by Miss Islington (bot) in branch '3.8':
bpo-38823: Fix refleaks in faulthandler init error path on Windows (GH-17250)
a5ed2fe

[vstinner]

New changeset 33b671e by Victor Stinner (Brandt Bucher) in branch 'master':
bpo-38823: Fix refleak in marshal init error path (GH-17260)
33b671e

[vstinner]

The Azure Pipelines are sick tonight (unable to publish test results). PR bpo-17272 and PR bpo-17271 are blocked by this CI.

[brandtbucher]

Thanks Victor. These obviously aren’t urgent, so feel free to return to them whenever’s convenient.

I also pinged you on bpo-17235 (_tracemalloc) too.

[vstinner]

New changeset d51a363 by Victor Stinner (Brandt Bucher) in branch 'master':
bpo-38823: Fix refleak in _tracemalloc init error handling (GH-17235)
d51a363

[miss-islington]

New changeset 63f09e7 by Miss Islington (bot) in branch '3.7':
bpo-38823: Fix refleak in marshal init error path (GH-17260)
63f09e7

[miss-islington]

New changeset 2ea4c37 by Miss Islington (bot) in branch '3.8':
bpo-38823: Fix refleak in marshal init error path (GH-17260)
2ea4c37

[miss-islington]

New changeset daf7a08 by Miss Islington (bot) in branch '3.8':
bpo-38823: Fix refleak in _tracemalloc init error handling (GH-17235)
daf7a08

[erlend-aasland]

A lot of issues were fixed by Serhiy in #106858. Are there more actionable items left? (I did not check the _decimal PR yet.)

[vstinner]

I suggest to close the issue which has a long history, and maybe create a new issue with a better described scope (list C extensions) if more C extensions need to be fixed.

[erlend-aasland]

All right, there are no open PRs left (they were all obsoleted by Serhiy's recent work in #106858).

I suggest to close the issue which has a long history, and maybe create a new issue with a better described scope (list C extensions) if more C extensions need to be fixed.

I agree; let's close this :) Thanks to Brandt for starting this effort in the first place, and thanks to all who improved extension module error handling!