WeakValueDictionary bug in setdefault()&pop()

[arigo]

BPO	19542
Nosy	@tim-one, @freddrake, @arigo, @pitrou, @serhiy-storchaka, @Tinche
PRs	[Do Not Merge] Convert Misc/NEWS so that it is managed by towncrier #552
Files	x.py: Raw test for Python 3. On Python 2, replace _thread with thread. weakref.slice: Proposed fix for these two methods (not as a diff because it's a bit unreadable) fix-weakvaluedict.diff: Test and fix for trunk fix-weakvaluedict-2.7.diff: Test and fix for 2.7

^{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 2016-12-26.20:58:59.392>
created_at = <Date 2013-11-10.08:51:04.410>
labels = ['3.7', 'type-bug', 'library']
title = 'WeakValueDictionary bug in setdefault()&pop()'
updated_at = <Date 2017-03-31.16:36:07.310>
user = 'https://github.com/arigo'

bugs.python.org fields:

activity = <Date 2017-03-31.16:36:07.310>
actor = 'dstufft'
assignee = 'none'
closed = True
closed_date = <Date 2016-12-26.20:58:59.392>
closer = 'pitrou'
components = ['Library (Lib)']
creation = <Date 2013-11-10.08:51:04.410>
creator = 'arigo'
dependencies = []
files = ['32557', '32558', '37992', '37993']
hgrepos = []
issue_num = 19542
keywords = ['needs review']
message_count = 15.0
messages = ['202510', '205603', '205609', '205614', '205645', '205674', '235301', '245470', '245471', '279455', '283599', '283601', '283602', '284040', '284046']
nosy_count = 8.0
nosy_names = ['tim.peters', 'fdrake', 'arigo', 'pitrou', 'Arfrever', 'python-dev', 'serhiy.storchaka', 'tinchester']
pr_nums = ['552']
priority = 'normal'
resolution = 'fixed'
stage = 'resolved'
status = 'closed'
superseder = None
type = 'behavior'
url = 'https://bugs.python.org/issue19542'
versions = ['Python 2.7', 'Python 3.5', 'Python 3.6', 'Python 3.7']

[arigo]

WeakValueDictionary.setdefault() contains a bug that shows up in multithreaded situations using reference cycles. Attached a test case: it is possible for 'setdefault(key, default)' to return None, although None is never put as a value in the dictionary. (Actually, being a WeakValueDictionary, None is not allowed as a value.)

The reason is that the code in setdefault() ends in "return wr()", but the weakref "wr" might have gone invalid between the time it was fetched from "self.data" and the "wr()" itself, thus returning None.

A similar problem occurs in pop(), leading it to possibly raise KeyError even if it is called with a default argument.

[pitrou]

I think the underlying question is: are weak dicts otherwise MT-safe?

[arigo]

As you can see in x.py, the underlying question is rather: are weakdicts usable in a single thread of a multithreaded program? I believe that this question cannot reasonably be answered No, independently on the answer you want to give to your own question.

[pitrou]

Ah, you're right. We just need to cook up a patch then.

[tim-one]

The weakref.slice fix looks solid to me, although it appears to be specific to 2.7 (the methods are fancier on the current default branch, fiddling with self._pending_removals too).

Does anyone know why the signature of pop is:

    def pop(self, key, *args)

? It doesn't make sense to me to accept any number of arguments following key. Perhaps it's just an obscure way to avoid doing:

_noarg = object()  # unique sentinel

def pop(self, key, default=_noarg):

?

[pitrou]

Changeset ea70032a24b1 is where the pop(*args) thing comes from.

[arigo]

Converted the test into an official test, which fails; and wrote the patch for CPython "3.trunk" and for CPython 2.7. Please review and commit.

[Tinche]

We're actually getting bitten by this in production through the Riak Python client, so this isn't a strictly theoretical problem.

[pitrou]

Yes, we need to fix this. Sorry for being a bit slow.

[arigo]

ping

[python-dev]

New changeset 5a542a2bca08 by Antoine Pitrou in branch '3.5':
Issue bpo-19542: Fix bugs in WeakValueDictionary.setdefault() and WeakValueDictionary.pop()
https://hg.python.org/cpython/rev/5a542a2bca08

New changeset f3706a9430da by Antoine Pitrou in branch '3.6':
Issue bpo-19542: Fix bugs in WeakValueDictionary.setdefault() and WeakValueDictionary.pop()
https://hg.python.org/cpython/rev/f3706a9430da

New changeset ac2715d04119 by Antoine Pitrou in branch 'default':
Issue bpo-19542: Fix bugs in WeakValueDictionary.setdefault() and WeakValueDictionary.pop()
https://hg.python.org/cpython/rev/ac2715d04119

[python-dev]

New changeset bcb1f0698401 by Antoine Pitrou in branch '2.7':
Issue bpo-19542: Fix bugs in WeakValueDictionary.setdefault() and WeakValueDictionary.pop()
https://hg.python.org/cpython/rev/bcb1f0698401

[pitrou]

This is finally fixed!