Fix a crash in not-async-context-manager when the manager has no name

[binggao1230]

Type of Changes

	Type
✓	🐛 Bug fix

Description

async with slice(None): crashed pylint:

File "pylint/checkers/async_checker.py", line 92, in visit_asyncwith
    "not-async-context-manager", node=node, args=(inferred.name,)
AttributeError: 'Slice' object has no attribute 'name'

visit_asyncwith builds the not-async-context-manager message argument from inferred.name, but the inferred context manager isn't guaranteed to have a name. slice(None) infers to a Slice node, which has no name, so emitting the (otherwise correct) warning raised AttributeError and aborted the run.

It now falls back to the inferred type name when there is no name, so slice(None) reports Async context manager 'slice' doesn't implement __aenter__ and __aexit__. — consistent with how the other value managers are already reported (e.g. async with 42 → 'int'). Among the value nodes that can appear here only Slice lacks a name (Const/List/Tuple/Dict all expose their type name), so this covers the reported case.

Added the slice(None) case to the existing not_async_context_manager functional test (crashes on main, passes here).

Closes #11102

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 96.29%. Comparing base (2f47d86) to head (2027451).

Additional details and impacted files

@@           Coverage Diff           @@
##             main   #11103   +/-   ##
=======================================
  Coverage   96.29%   96.29%           
=======================================
  Files         178      178           
  Lines       19737    19743    +6     
=======================================
+ Hits        19006    19012    +6     
  Misses        731      731           

Files with missing lines	Coverage Δ
pylint/checkers/async_checker.py	100.00% <100.00%> (ø)
pylint/checkers/typecheck.py	96.58% <100.00%> (+0.01%)	⬆️

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[Pierre-Sassoulas]

Thank you for contributing to pylint ! Let's anticipate other similar issue and use domain knowledge to make the guard better.

[Pierre-Sassoulas]

Let's also add a with slice(None):. And move it to the functional test of the checker or checkers that are going to crash when we do :)

[binggao1230]

Good call. The synchronous with slice(None): crashes not-context-manager in the same way (astroid raises AttributeInferenceError on the missing .name), so I added a functional case for it in tests/functional/n/not_context_manager.py and fixed that checker too (verified red before the fix, green after).

[Pierre-Sassoulas]

We can often narrow down the type of the inferred node using the visitor pattern filtering and domain knowledge about python syntax in order to use isinstance check instead of getattr

[binggao1230]

Done. I checked what actually reaches this point: every other inferred result is name-bearing (a class/function, or an instance such as the int that astroid models as a named Const for 42); a Slice from slice(...) is the only nameless case. So the guard is now an explicit isinstance(inferred, nodes.Slice) that falls back to the builtin type name, instead of getattr(..., "name", None). Same guard applied to the sync not-context-manager site.

[Pierre-Sassoulas]

I'm not certain we can be sure that Slice is the only node without a name attribute here, I'd rather have an allow list of node known to work (baseinstance/classdef/functiondef/module etc.)

[binggao1230]

Good point, switched to that approach in c0a184e. Both sites now read .name only from an allow list of node types that define it (ClassDef, FunctionDef, Module, bases.BaseInstance) and fall back to the inferred type name otherwise, so any unforeseen nameless node degrades gracefully instead of crashing. The existing functional cases (instances, generators, int, etc.) are all BaseInstance/ClassDef so their messages are unchanged, and slice(...) takes the fallback and still reports slice.

[Pierre-Sassoulas]

Thank you for checking, let's just invert the logic from disallow list to allow list to avoid future crash if the list is not perfect and let's merge.

[binggao1230]

Mirrored your nodes.Lambda suggestion to the identical allow list in typecheck.py (the sync not-context-manager site) so the two stay in lockstep — the suggestion UI could only touch the async_checker.py hunk. Functional tests for both still pass. Ready to merge from my side.

[Pierre-Sassoulas]

Great, thank you !