Segfaults in C heap type subclasses

[eduardo-elizondo]

BPO	37879
Nosy	@vstinner, @encukou, @eduardo-elizondo
PRs	bpo-37879: Suppress subtype_dealloc decref when base type is a C heap type #15323 [3.8] bpo-37879: Suppress subtype_dealloc decref when base type is a C heap type (GH-15323) #15966 bpo-37879: Fix warnings in _testcapimodule #16004

^{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 2019-10-08.12:15:19.200>
created_at = <Date 2019-08-17.20:47:30.987>
labels = []
title = 'Segfaults in C heap type subclasses'
updated_at = <Date 2019-10-08.12:15:19.199>
user = 'https://github.com/eduardo-elizondo'

bugs.python.org fields:

activity = <Date 2019-10-08.12:15:19.199>
actor = 'petr.viktorin'
assignee = 'none'
closed = True
closed_date = <Date 2019-10-08.12:15:19.200>
closer = 'petr.viktorin'
components = []
creation = <Date 2019-08-17.20:47:30.987>
creator = 'eelizondo'
dependencies = []
files = []
hgrepos = []
issue_num = 37879
keywords = ['patch']
message_count = 6.0
messages = ['349905', '351773', '352067', '352082', '352344', '354194']
nosy_count = 3.0
nosy_names = ['vstinner', 'petr.viktorin', 'eelizondo']
pr_nums = ['15323', '15966', '16004']
priority = 'normal'
resolution = 'fixed'
stage = 'resolved'
status = 'closed'
superseder = None
type = None
url = 'https://bugs.python.org/issue37879'
versions = []

[eduardo-elizondo]

subtype_dealloc is not correctly handling the reference count of c heap type subclasses. It has some builtin assumptions which can result in the type getting its reference count decreased more that it needs to be, leading to its destruction when it should still be alive.

Also, this bug is a blocker for the full adoption of PEP-384.

The full details of the bug along with a fix and tests are described in the Github PR.

[encukou]

New changeset ff023ed by Petr Viktorin (Eddie Elizondo) in branch 'master':
bpo-37879: Suppress subtype_dealloc decref when base type is a C heap type (GH-15323)
ff023ed

[encukou]

New changeset 5e9caee by Petr Viktorin in branch 'master':
bpo-37879: Fix warnings in _testcapimodule (GH-16004)
5e9caee

[encukou]

New changeset 3562ae2 by Petr Viktorin in branch '3.8':
[3.8] bpo-37879: Suppress subtype_dealloc decref when base type is a C heap type (GH-15323, #60208) (GH-15966)
3562ae2

[vstinner]

I reopen the issue: the change caused a regression on the Refleak buildbot workers: bpo-38150.

[encukou]

Thanks to Stéphane and Eddie for fixing the refleak!