metaclass can't derive from ABC

[roalddevries]

BPO	9533
Nosy	@terryjreedy, @benjaminp, @merwok, @bitdancer, @durban
Files	bug.py: source file leading to the exception in the comment

^{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 2010-10-02.17:53:46.500>
created_at = <Date 2010-08-06.10:33:37.675>
labels = ['type-bug', 'library']
title = "metaclass can't derive from ABC"
updated_at = <Date 2010-10-02.17:53:46.498>
user = 'https://bugs.python.org/roalddevries'

bugs.python.org fields:

activity = <Date 2010-10-02.17:53:46.498>
actor = 'benjamin.peterson'
assignee = 'none'
closed = True
closed_date = <Date 2010-10-02.17:53:46.500>
closer = 'benjamin.peterson'
components = ['Library (Lib)']
creation = <Date 2010-08-06.10:33:37.675>
creator = 'roalddevries'
dependencies = []
files = ['18414']
hgrepos = []
issue_num = 9533
keywords = []
message_count = 5.0
messages = ['113094', '114025', '117802', '117880', '117891']
nosy_count = 6.0
nosy_names = ['terry.reedy', 'benjamin.peterson', 'eric.araujo', 'r.david.murray', 'daniel.urban', 'roalddevries']
pr_nums = []
priority = 'normal'
resolution = 'fixed'
stage = None
status = 'closed'
superseder = None
type = 'behavior'
url = 'https://bugs.python.org/issue9533'
versions = ['Python 3.1', 'Python 2.7', 'Python 3.2']

[roalddevries]

Exception raised::

  Traceback (most recent call last):
    File "bug.py", line 5, in <module>
      class derived(type, Sized):
    File "/opt/local/Library/Frameworks/Python.framework/Versions/2.6/lib/python2.6/abc.py", line 85, in __new__
      for name in getattr(base, "__abstractmethods__", set()):
  TypeError: Error when calling the metaclass bases
      'getset_descriptor' object is not iterable

[terryjreedy]

+- same traceback in 3.1
Since ABCmeta is not used (by name) its import is not needed.
I have no opinion on whether this should work.

[durban]

If we create a new class, which is a metaclass, and also inherits an ABC, we create a new instance of ABCMeta.

When ABCMeta.__new__ creates the __abstractmethods__ attribute of the new class, it iterates over the __abstractmethods__ attribute of every base class (and checks every name in it, if it's abstract in the new class). One of the base classes of a metaclass is of course type. The type.__abstractmethods__ object is a getset descriptor (its __set__ switches some flags in tp_flags), which isn't iterable, so it raises the exception.

Ideas for possible solutions:

1. In ABCMeta.__new__ special case type (do not check its __abstractmethods__attribute).
2. In ABCMeta.__new__ check if the __abstractmethods__ attribute is in fact an iterable, before we try to iterate over it. (I don't think this would be a very good solution, because if a base class' __abstractmethods__ isn't iterable, it should be an error, except if the base class is type itself. So we should special case type anyway.)

I can't come up with a use case for this right now, but I don't see why it shouldn't work.

[bitdancer]

Does the "fix" for bpo-10006 affect this? (I imagine that question is why Antoine made Benjamin nosy on this issue).

[benjaminp]

You can now create metaclass abcs. However, having __abstractmethods__ does not prevent instance creation. This is a problem with a builtins, though.