Fix enum attributes are not members #17207
Conversation
terencehonles
force-pushed
the
fix-enum-with-property-match-exhaustion
branch
from
f319685
to
ba435ac
Compare
This comment has been minimized.
This comment has been minimized.
| isinstance(sym.node, mypy.nodes.Var) | ||
| and name not in ENUM_REMOVED_PROPS | ||
| and not name.startswith("__") | ||
| and sym.node.has_explicit_value |
There was a problem hiding this comment.
This last condition changes the behavior of mypy which currently considers unassigned but annotated attributes as members of the enum.
I would say the new behavior is better but it must be noted in the PR title or description that this is a breaking change that must be considered on its own.
Looking at the mypy primer output, it looks like members of enums constructed using the enum call syntax do not have the has_explicit_value set which breaks type narrowing. I suggest changing the variable here to also set this flag perhaps with a comment explaining the reason -- something like "E = Enum("E", "A") is equivalent to class E(Enum): A = auto()"
There was a problem hiding this comment.
Thanks for noticing that, I've not really used them in that form, but hopefully the comment is not too wordy. I only glanced at the code that you pointed to, but I was cross referencing https://github.com/python/cpython/blob/7528b84e947f727734bd802356e380673553387d/Lib/enum.py#L828-L839 and how much of this is implemented by mypy? 🤔
This comment has been minimized.
This comment has been minimized.
|
It looks like this is actually related to the ongoing discussion python/typing-council#11 |
There was a problem hiding this comment.
Thank you @terencehonles.
The typing spec change has been accepted and this PR looks good to me.
Regarding the mypy primer diff, it is related to the linked typing spec change that requires assignment for enum attributes for them to be considered members of the enumeration and the fact that tanjun vendors an old copy of the stdlib inspect module and stub that relies on the old behavior. So the new errors are expected.
| def __new__(cls, value: int): pass | ||
|
|
||
| class Medal(int, Enum): | ||
| prize: str |
There was a problem hiding this comment.
Could you add a test case with a class attribute with an assigned value, e.g., prize: str = "nothing"?
There was a problem hiding this comment.
yup, should be able to tomorrow
There was a problem hiding this comment.
I thought about this a bit, and this is either a request for a nonmember(...) test, which is currently not supported, or an enum like the one above. I've added a test for nonmember that's been marked as expected to fail: https://github.com/python/mypy/pull/17207/files#diff-4f8d4b202949b4cf2221e3295fbbd1774e152e44b141880981895c349599e927R1-R36
Let me know if that's what you wanted, and/or if we should keep that
This adds on to the change in python#17182 and fixes enum attributes being used as members.
terencehonles
force-pushed
the
fix-enum-with-property-match-exhaustion
branch
from
5383c49
to
dfcc0c4
Compare
|
Diff from mypy_primer, showing the effect of this PR on open source code: Tanjun (https://github.com/FasterSpeeding/Tanjun)
+ tanjun/_internal/__init__.py:189: error: Statement is unreachable [unreachable]
+ tanjun/annotations.py:2646: error: Statement is unreachable [unreachable]
|
This adds on to the change in #17182 and fixes enum attributes being used as members.
@hamdanal / @hauntsaninja I noticed there was the function
get_enum_valuesin mypy types and instead of adding more to the case intry_expanding_sum_type_to_unionit seemed like it might make sense to move the changes from #17182 there. I don't think the other code touched in that PR can useget_enum_values.