dict views don't implement subtraction correctly

[MojoVampire]

BPO	26478
Nosy	@wimglenn, @MojoVampire, @zhangyangyu

^{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 2016-03-04.06:11:05.806>
created_at = <Date 2016-03-03.23:20:42.873>
labels = ['interpreter-core']
title = "dict views don't implement subtraction correctly"
updated_at = <Date 2016-03-08.04:55:26.092>
user = 'https://github.com/MojoVampire'

bugs.python.org fields:

activity = <Date 2016-03-08.04:55:26.092>
actor = 'python-dev'
assignee = 'none'
closed = True
closed_date = <Date 2016-03-04.06:11:05.806>
closer = 'python-dev'
components = ['Interpreter Core']
creation = <Date 2016-03-03.23:20:42.873>
creator = 'josh.r'
dependencies = []
files = []
hgrepos = []
issue_num = 26478
keywords = []
message_count = 4.0
messages = ['261180', '261182', '261193', '261322']
nosy_count = 4.0
nosy_names = ['python-dev', 'wim.glenn', 'josh.r', 'xiang.zhang']
pr_nums = []
priority = 'normal'
resolution = 'fixed'
stage = 'resolved'
status = 'closed'
superseder = None
type = None
url = 'https://bugs.python.org/issue26478'
versions = ['Python 3.5', 'Python 3.6']

[MojoVampire]

Don't know when the problem was introduced, but dictviews_sub is doing:

    tmp = _PyObject_CallMethodId(result, &PyId_difference_update, "O", other);

to implement subtraction (after creating result as a set of the keys in question). That's violating the CallMethod contract though (which states the format string should produce a tuple), and while it looks like CallMethod does fixups when the contract is violated, this creates some very odd behaviors.

With a list, everything is fine:

    >>> d = {0: 'zero', 1: 'one', 2: 'two', 3: 'three'}
    >>> d.keys() - [0, 2]
    {1, 3}
    >>> d.keys() - (0, 2)
    TypeError: 'int' object is not iterable

Basically, the fix up doesn't get applied when you subtract a tuple, so it's as if it's trying to call:

    result.difference_update(*(0, 2))  # Unpacking used to illustrate, effect is  result.difference_update(0, 2)

With the list, it's wrapping to make a one element tuple, so it behaves like:

    result.difference_update(*([0, 2],))  # Unpacking used to illustrate, effect is  result.difference_update([0, 2])

For more details, see http://stackoverflow.com/q/35784258/364696

Fix should be to change call line to:

    tmp = _PyObject_CallMethodObjArgsId(result, &PyId_difference_update, other, NULL);

(assuming _PyObject_CallMethodObjArgsId is a thing), or if it's not a thing, to fix the format string to force tuple wrapping:

    tmp = _PyObject_CallMethodId(result, &PyId_difference_update, "(O)", other);

[python-dev]

New changeset 0cae6b6e3d7d by Benjamin Peterson in branch '3.4':
properly use the ObjArgs variant of CallMethod in dictview binary operations (closes bpo-26478)
https://hg.python.org/cpython/rev/0cae6b6e3d7d

New changeset a5bc5c9490f5 by Benjamin Peterson in branch '3.5':
merge 3.4 (closes bpo-26478)
https://hg.python.org/cpython/rev/a5bc5c9490f5

New changeset 9532a8815a9c by Benjamin Peterson in branch 'default':
merge 3.5 (closes bpo-26478)
https://hg.python.org/cpython/rev/9532a8815a9c

[wimglenn]

Well that was patched quickly, impressive turnaround on this

[python-dev]

New changeset 309f3559a729 by Benjamin Peterson in branch '2.7':
properly use PyObject_CallMethod in dictview binary operations (closes bpo-26478)
https://hg.python.org/cpython/rev/309f3559a729