Typing conformance results

No changes detected ✅

mypy_primer results

No ecosystem changes detected ✅

Memory usage report

Summary

Significant changes

Merging this PR will improve performance by 31.95%

⚡ 1 improved benchmark
✅ 25 untouched benchmarks
⏩ 30 skipped benchmarks¹

Performance Changes

_{Comparing charlie/mem (fac85b5) with main (2185bec)}

I'll note for posterity that privileging enum members over other attributes results in behavior that doesn't match runtime in the weird case where you have a mixin enum type (like IntEnum, where members are also subclasses of int), and then you also define an enum member that overrides a real attribute of the mixin class (e.g. an IntEnum with a member named numerator or denominator). In this case at runtime MyWeirdIntEnum.numerator.numerator actually accesses the real int attribute numerator on the enum value MyWeirdIntEnum.numerator, whereas in this PR you'd get back the enum member again.

But a) this case is invalid, and once we improve our Liskov checking we should emit a Liskov violation on it, and b) other type checkers all do the same. So I think this behavior is fine.