Use functools.cached_property on 3.8+
#5907
Conversation
DanielNoord
added
the
Maintenance
Pull Request Test Coverage Report for Build 1974339656
💛 - Coveralls |
| if sys.version_info >= (3, 8) or TYPE_CHECKING: | ||
| # pylint: disable-next=ungrouped-imports | ||
| from functools import cached_property | ||
| else: | ||
| # pylint: disable-next=ungrouped-imports | ||
| from astroid.decorators import cachedproperty as cached_property |
There was a problem hiding this comment.
I thought about this some more, especially since I wanted to see what it would take to resolve / hide the pylint error.
First, the reason it's emitted is that we don't check BoolOp nodes (in is_node_in_guarded_import_block) when we determine if ungrouped-imports should be emitted.
https://github.com/PyCQA/pylint/blob/dee07af04df93c4b5edeb2492eaa5c46ebcdb7b7/pylint/checkers/imports.py#L548-L550
The question then becomes if we should do that. To answer that it's worth taking a look at the condition itself and how cached_property is defined in typeshed. Basically typeshed also uses version guards, in particular if sys.version_info >= (3, 8). That means the or TYPE_CHECKING I suggested in astroid is basically useless. If we run mypy with --python-version 3.8 or above, the sys.version_info check is enough, whereas with 3.7 we just get an attr-defined error anyway.
Module "functools" has no attribute "cached_property" [attr-defined]
In the end or TYPE_CHECKING is useless and causes the unnecessary errors. So it might be best / easiest if we remove them.
Type of Changes
Description
Last PR to fix all deprecations as a result of
astroidupgrade 🎉