Don't enable GC for classes that don't add new fields

[eltoder]

BPO	23726
Nosy	@pitrou, @benjaminp, @njsmith, @eltoder, @serhiy-storchaka
Files	class_gc.diff class_gc2.diff

^{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 2015-04-13.18:10:31.603>
created_at = <Date 2015-03-20.20:25:57.755>
labels = ['interpreter-core', 'performance']
title = "Don't enable GC for classes that don't add new fields"
updated_at = <Date 2015-04-13.19:30:52.528>
user = 'https://github.com/eltoder'

bugs.python.org fields:

activity = <Date 2015-04-13.19:30:52.528>
actor = 'eltoder'
assignee = 'none'
closed = True
closed_date = <Date 2015-04-13.18:10:31.603>
closer = 'pitrou'
components = ['Interpreter Core']
creation = <Date 2015-03-20.20:25:57.755>
creator = 'eltoder'
dependencies = []
files = ['38610', '38624']
hgrepos = []
issue_num = 23726
keywords = ['patch']
message_count = 11.0
messages = ['238718', '238721', '238727', '238730', '238732', '238733', '240690', '240691', '240699', '240710', '240711']
nosy_count = 6.0
nosy_names = ['pitrou', 'benjamin.peterson', 'njs', 'python-dev', 'eltoder', 'serhiy.storchaka']
pr_nums = []
priority = 'normal'
resolution = 'fixed'
stage = 'resolved'
status = 'closed'
superseder = None
type = 'performance'
url = 'https://bugs.python.org/issue23726'
versions = ['Python 3.5']

[eltoder]

As far as I can tell, if a new class does not add any new fields, and its base class doesn't use GC, there's no reason to enable GC for the new class.
This is useful for creating lightweight wrappers around classes implemented in C.

[serhiy-storchaka]

>>> class UserIntSlots(int):
...     __slots__ = ()
...     def __repr__(s): return '5'
... 
>>> (4).__class__ = UserIntSlots
>>> 2+2
5
>>> type(2+2)
<class '__main__.UserIntSlots'>

It looks weird.

[eltoder]

Agreed, but this is not new. This works without my change:

>>> class Tuple(tuple):
...   __slots__ = ()
...   def __repr__(self): return 'Imma tuple!'                                                    
... 
>>> ().__class__ = Tuple
>>> ()
Imma tuple!

[pitrou]

I wasn't aware of that :-o

[serhiy-storchaka]

May be we should forbid assigning the __class__ attribute of basic builtin
types (especially internable, such as int, str, bytes, tuple, bool, NoneType).

[eltoder]

Actually, this is rather new -- new in 3.5. The check was relaxed in bpo-22986: https://hg.python.org/cpython/rev/c0d25de5919e
Previously only heap types allowed re-assigning __class__.

[python-dev]

New changeset a60b7945ef87 by Antoine Pitrou in branch 'default':
Issue bpo-23726: Don't enable GC for user subclasses of non-GC types that don't add any new fields.
https://hg.python.org/cpython/rev/a60b7945ef87

[pitrou]

I've pushed the patch, thank you!

[eltoder]

Thank you!

Benjamin, Nathaniel, any opinion if we should restrict reassigning __class__ for types like tuple, int and str, where some/many instances are cached?

[njsmith]

Yes, it probably would be a good idea to disallow assigning __class__ for immutable types.

[eltoder]

Agreed. There's a small problem with that, as far as I know. Nothing on type declares that it is immutable.