__slots__ on PyVarObject subclass

[ronaldoussoren]

BPO	17295
Nosy	@ronaldoussoren, @amauryfa, @vstinner, @vadmium
Superseder	bpo-1173475: slots for subclasses of variable length types

^{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 = None
created_at = <Date 2013-02-25.16:24:22.254>
labels = ['interpreter-core', 'type-bug']
title = '__slots__ on PyVarObject subclass'
updated_at = <Date 2015-03-12.05:27:03.174>
user = 'https://github.com/ronaldoussoren'

bugs.python.org fields:

activity = <Date 2015-03-12.05:27:03.174>
actor = 'martin.panter'
assignee = 'none'
closed = False
closed_date = None
closer = None
components = ['Interpreter Core']
creation = <Date 2013-02-25.16:24:22.254>
creator = 'ronaldoussoren'
dependencies = []
files = []
hgrepos = []
issue_num = 17295
keywords = []
message_count = 5.0
messages = ['182959', '184637', '184640', '185397', '237914']
nosy_count = 4.0
nosy_names = ['ronaldoussoren', 'amaury.forgeotdarc', 'vstinner', 'martin.panter']
pr_nums = []
priority = 'low'
resolution = 'duplicate'
stage = 'resolved'
status = 'open'
superseder = '1173475'
type = 'behavior'
url = 'https://bugs.python.org/issue17295'
versions = ['Python 3.4']

[ronaldoussoren]

Currently a subclass of a PyVarObject (such as 'int') cannot use a non-empty slots, for example:

>>> class L (int):
...    __slots__ = ('a', 'b')
... 
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
TypeError: nonempty __slots__ not supported for subtype of 'int'
>>> 

However, subclasses that don't use __slots__ do have an implicit slot: the __dict__ value. Wouldn't it be possible to use the trick used for tp_dictoffset for slots as well, that is use negative values for the slot offsets and access them from the end of the object?

The major problem with an implementation is that PyMember_GetOne has a 'char*' argument instead of a 'PyObject*' one, but that's easily changed (PyMember_GetOne is currently only called in Object/descrobject.c and that call casts PyObject* to char*. This would be an API change, but AFAIK PyMember_GetOne is not documented at the moment (and changing the argument type would be binary compatible).

I hope to work on a patch in the near future.

[amauryfa]

Probably a duplicate of bpo-1173475. Do you want to work on the patch there?

[ronaldoussoren]

You're right, this is a duplicate.

[vstinner]

I closed the issue, but I didn't see that bpo-1173475 was closed. Sorry, I reopened this issue again.

[vadmium]

Encountered this when trying to add some fields to urllib.parse.SplitResult for bpo-22852. bpo-1173475 has a patch; I haven’t tried using it though.

[encukou]

For this to work, the interpreter must know the type's memory layout.
So, in 2023 in Python 3.12 3.13, we can solve it for types with PEP 697's Py_TPFLAGS_ITEMS_AT_END, but for tuple or int we'd need to make their layout known (possibly we could hardcode it for those types only, but even then, documenting what kind of access is safe re. other types or subclasses might turn out to be a PEP-sized chunk of work).

Let's use gh-103740 for enabling this with Py_TPFLAGS_ITEMS_AT_END.

[erlend-aasland]

Let's use gh-103740 for enabling this with Py_TPFLAGS_ITEMS_AT_END.

Let's close this as superseded by #103740, unless you have objections.

[ronaldoussoren]

Let's use gh-103740 for enabling this with Py_TPFLAGS_ITEMS_AT_END.

Let's close this as superseded by #103740, unless you have objections.

Fine by me.