ABCs for MappingViews should declare __slots__ so subclasses aren't forced to have __dict__/__weakref__

[MojoVampire]

BPO	21421
Nosy	@rhettinger, @MojoVampire
Files	slots_for_mappingview_abcs.patch collections_abc_cleanup.patch

^{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/rhettinger'
closed_at = <Date 2014-05-04.02:09:57.033>
created_at = <Date 2014-05-03.00:02:41.932>
labels = ['type-feature', 'library']
title = "ABCs for MappingViews should declare __slots__ so subclasses aren't forced to have __dict__/__weakref__"
updated_at = <Date 2014-05-04.21:35:49.950>
user = 'https://github.com/MojoVampire'

bugs.python.org fields:

activity = <Date 2014-05-04.21:35:49.950>
actor = 'josh.r'
assignee = 'rhettinger'
closed = True
closed_date = <Date 2014-05-04.02:09:57.033>
closer = 'rhettinger'
components = ['Library (Lib)']
creation = <Date 2014-05-03.00:02:41.932>
creator = 'josh.r'
dependencies = []
files = ['35142', '35143']
hgrepos = []
issue_num = 21421
keywords = ['patch']
message_count = 7.0
messages = ['217809', '217810', '217811', '217815', '217852', '217853', '217893']
nosy_count = 3.0
nosy_names = ['rhettinger', 'python-dev', 'josh.r']
pr_nums = []
priority = 'normal'
resolution = 'fixed'
stage = None
status = 'closed'
superseder = None
type = 'enhancement'
url = 'https://bugs.python.org/issue21421'
versions = ['Python 3.5']

[MojoVampire]

Unlike the other collections ABCs, MappingView and its subclasses (Keys|Items|Values)View don't define __slots__, so users inheriting them to take advantage of the mix-in methods get a __dict__ and __weakref__, even if they try to avoid it by declaring their own __slots__. This is sub-optimal (I don't think any library class should leave __slots__ undefined, but it's particularly bad when they're intended to be used a base classes).

I've attached a patch that defines __slots__ for all of them; MappingView explicitly declares its _mapping slot, and the rest declare no slots at all.

Only negative I can think of is that if the user provides their own __init__, doesn't call the super().__init__, and uses a different name than _mapping for whatever data structure they actually use, then there will be a pointer reserved for _mapping that never gets set. That said, if they don't define _mapping, none of the mix-in methods work anyway, so they may as well not inherit directly, and instead just use *View.register to become a virtual subclass.

[MojoVampire]

Hmm... First patch isn't generating a review diff (I'm guessing because my VM's time sync went farkakte). Trying again.

[MojoVampire]

I also wrote a slightly more thorough patch that simplifies some of the method implementations (largely just replacing a bunch of for/if/return True/False else return False/True with any/all calls against a generator expression).

The one behavior difference is that I dramatically simplified Sequence's __iter__; the original just kept indexing until it got IndexError, the replacement just runs len(self) times, on the theory that a self-mutating Sequence should write its own __iter__, and we shouldn't jump through hoops to accommodate unsupported behavior like mutating a Sequence during iteration.

All tests pass under both patches.

[rhettinger]

At first glance, this seems like a reasonable request.
I leave it open for comments for a while before proceeding.

[python-dev]

New changeset 2c6a231e2c1e by Raymond Hettinger in branch 'default':
Issue bpo-21421: Add __slots__ to the MappingViews ABCs.
http://hg.python.org/cpython/rev/2c6a231e2c1e

[rhettinger]

I've applied the __slots__ patch. Thank you for submitting it.

Am leaving the rest of the ABCs code as-is. The current form may be a bit wordy but it is clean, fast, and easy to trace through a debugger. The expanded forms were chosen for a reason.

[MojoVampire]

Cool. Thanks for the feedback. I split it into two patches for a reason; I wasn't wedded to the simplification.