gh-112821: Fix rlcompleter failures on objects with descriptors

[mdboom]

This issue came to light for me when it was discovered that autocompletion didn't work for a library I help maintain, written mainly in Cython. Cython implements @property-annotated methods on its cdef classes as getset descriptors. The rlcompleter logic doesn't exclude these (like it does regular Python properties) since isinstance(member, property) is False. It then goes on to call the getter, which in some cases (in our library at least) raises an exception. Any exception other than an AttributeError causes the rlcompleter to short circuit completely and not offer any options.

This is a more surgical (rlcompleter-only) solution than what was proposed in #112821. If adding ABCs is preferred, I could do that.

Restore the try/except removed in #27401
Create abstract classes for descriptors in _collections_abc (e.g. ImplementsGet, ImplementsSet, ImplementsDelete) and in rlcompleter, rather than checking for isinstance(obj, property), use isinstance(obj, ImplementsGet).

#112821 also mentioned:

This might cause false positives for method descriptors though.

I think this is handled here, but I also may not fully understand this specific issue.

• Issue: Reconsider gh-88915 changes rlcompleter since exceptions in descriptors cause auto completion to fail #112821

[python-cla-bot]

All commit authors signed the Contributor License Agreement.

[JelleZijlstra]

This could also run arbitrary code because the descriptor protocol also gets run on access through the class. To be fully safe we should probably use something like inspect.getattr_static. Still, your change is a strict improvement so I'm OK with landing it.

[mdboom]

I experimented with this but it breaks another test create_expected_for_none and changes behavior elsewhere. Upstream behavior and this PR:

>>> None.__class__<TAB>
None.__class__()

Changing this line to inspect.getattr_static(type(thisobject), word, None):

>>> None.__class__<TAB>
None.__class__

I could argue this either way, but I think I will merge this as-is as it doesn't change any existing tested behavior.

[miss-islington-app]

Thanks @mdboom for the PR 🌮🎉.. I'm working now to backport this PR to: 3.13, 3.14, 3.15.
🐍🍒⛏🤖

[bedevere-app]

GH-149656 is a backport of this pull request to the 3.15 branch.

[bedevere-app]

GH-149657 is a backport of this pull request to the 3.14 branch.

[bedevere-app]

GH-149658 is a backport of this pull request to the 3.13 branch.