Fix data descriptor detection in inspect.getattr_static

@rhettinger

BPO	31184
Nosy	@rhettinger, @serhiy-storchaka
Superseder	bpo-26103: Contradiction in definition of "data descriptor" between (dotted lookup behavior/datamodel documentation) and (inspect lib/descriptor how-to)
Files	0001-Fix-data-descriptor-detection-in-inspect.getattr_sta.patch: Potential patch that fixes the issue

^{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 2017-08-11.16:22:11.131>
labels = ['type-bug', 'library', '3.11']
title = 'Fix data descriptor detection in inspect.getattr_static'
updated_at = <Date 2021-12-08.04:22:58.232>
user = 'https://bugs.python.org/davidhalter'

bugs.python.org fields:

activity = <Date 2021-12-08.04:22:58.232>
actor = 'rhettinger'
assignee = 'none'
closed = False
closed_date = None
closer = None
components = ['Library (Lib)']
creation = <Date 2017-08-11.16:22:11.131>
creator = 'davidhalter'
dependencies = []
files = ['47078']
hgrepos = []
issue_num = 31184
keywords = ['patch']
message_count = 4.0
messages = ['300170', '300173', '407976', '407985']
nosy_count = 3.0
nosy_names = ['rhettinger', 'serhiy.storchaka', 'davidhalter']
pr_nums = []
priority = 'normal'
resolution = None
stage = 'resolved'
status = 'open'
superseder = '26103'
type = 'behavior'
url = 'https://bugs.python.org/issue31184'
versions = ['Python 3.11']

Linked PRs

inspect.getattr_static is currently not identifying data descriptors the right way.

Data descriptors are defined by having a __get__ attribute and at least one of the __set__ and __delete__ attributes.

Implementation detail: Both __delete__ and __get__ set the same slot called tp_descr_set in CPython.

I have attached a patch that fixes the issue IMO.

See also bpo-26103.

This is not a duplicate. It is related to https://bugs.python.org/issue26103, because __get__ is not required anymore for an object to be a data descriptor. The current code on master (of inspect.getattr_static) still thinks a descriptor has both __get__ and __set__ set.

Since issue bpo-26103 has been fixed, it's now clear that my patch is slightly wrong, but I'm happy to fix that if someone is actually going to review it.

Can you give an example of where getattr_static() is not doing what you expect?

@rhettinger

Thanks for bringing it up again. I unfortunately somehow missed @rhettinger's comment on this issue. Will try to give an example as soon as possible.

@davidhalter

@davidhalter ping

As soon as possible doesn't mean I have time immediately. It will happen and it's still on my radar. Sorry, but I have been quite busy.

@iritkatriel

Here's my example:

import inspect

class DescriptorGet:
    def __get__(self, instance, klass):
        return "Foo"

class DescriptorGetDelete:
    def __get__(self, instance, klass):
        return "Foo"
    def __delete__(self, instance, klass):
        pass

class DescriptorGetSet:
    def __get__(self, instance, klass):
        return "Foo"
    def __set__(self, instance, klass, value):
        pass

class Foo:
    get = DescriptorGet()
    get_delete = DescriptorGetDelete()
    get_set = DescriptorGetSet()

    def __init__(self):
        self.__dict__['get'] = 42
        self.__dict__['get_delete'] = 42
        self.__dict__['get_set'] = 42

for attr in ['get', 'get_delete', 'get_set']:
    print(f'getattr        {attr}', getattr(Foo(), attr))
    print(f'getattr_static {attr}', inspect.getattr_static(Foo(), attr))

Output:

getattr        get 42
getattr_static get 42
getattr        get_delete Foo
getattr_static get_delete 42
getattr        get_set Foo
getattr_static get_set <__main__.DescriptorGetSet object at 0x7f88b10ab7f0>

The get_delete case is clearly wrong, because if a normal getattr returns the executed descriptor, we should never get the value of the instance's __dict__ as a result in the getattr_static case.

@iritkatriel @rhettinger Does this make sense? In case you think it's a bug as well, I'm happy to create a pull request with my patch & a unit test of the __delete__ case.

@davidhalter

Thanks @davidhalter. I agree that this looks like a bug. Just to give a slightly shorter reproducer:

>>> class DescriptorGetDelete:
...     def __get__(self, instance, klass):
...         return 'foo'
...     def __delete__(self, instance, klass): pass
...
>>> class Foo:
...     get_delete = DescriptorGetDelete()
...     def __init__(self):
...         self.__dict__['get_delete'] = 42
...
>>> foo = Foo()
>>> foo.get_delete
'foo'
>>> import inspect
>>> inspect.getattr_static(foo, 'get_delete')
42

The result of the last call should probably be <__main__.DescriptorGetDelete object at 0x000001D9BF998DA0> rather than 42.

If you file a PR for this issue, feel free to ping me on it -- I'd be happy to review it.

@davidhalter

The fix here is simply to also check for __delete__ method, not only __set__ method.

To be clear, despite the resolution of #70291 (which I think was resolved incorrectly), the check for __get__ method here in getattr_static is correct and should remain. In other words, I think @davidhalter 's original patch here remains the correct fix, despite the resolution of #70291.

Merge branch 'main' into pythongh-75367

gh-75367: Fix data descriptor detection in inspect.getattr_static (#1…

pythongh-75367: Fix data descriptor detection in inspect.getattr_stat…

[3.11] gh-75367: Fix data descriptor detection in inspect.getattr_sta…

Merge branch 'main' into 709funconly

@carljm

@carljm The issue seems to have been resolved. We can close the issue.

Navigation Menu

Repository navigation

Fix data descriptor detection in inspect.getattr_static #75367

Linked PRs

7 remaining items

Metadata

Assignees

Labels

Projects

Milestone

Relationships

Development

Fix data descriptor detection in inspect.getattr_static #75367

Description

Linked PRs

Activity

davidhalter commented on Aug 11, 2017

serhiy-storchaka commented on Aug 11, 2017

davidhalter commented on Dec 7, 2021

rhettinger commented on Dec 7, 2021

7 remaining items

davidhalter commented on Feb 21, 2023

arhadthedev commented on Mar 4, 2023

davidhalter commented on Mar 5, 2023

davidhalter commented on Mar 8, 2023

AlexWaygood commented on Apr 2, 2023

carljm commented on Apr 6, 2023

furkanonder commented on May 17, 2023

Metadata

Metadata

Assignees

Labels

Projects

Milestone

Relationships

Development

Issue actions