Enum union and intersection with Any not simplifying correctly

[github-actions]

Run listed here: https://github.com/astral-sh/ty/actions/runs/17581914325

[AlexWaygood]

failures:

---- types::property_tests::stable::subtype_of_is_antisymmetric stdout ----

thread 'types::property_tests::stable::subtype_of_is_antisymmetric' panicked at /home/runner/.cargo/registry/src/index.crates.io-1949cf8c6b5b557f/quickcheck-1.0.3/src/tester.rs:165:28:
[quickcheck] TEST FAILED. Arguments: (Intersection { pos: [Union([Any, EnumLiteral("unsafe")]), EnumLiteral("unsafe")], neg: [] }, EnumLiteral("unsafe"))
note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace

[sharkdp]

It looks like (Any | E) & E = Any & E | E doesn't simplify to E, or that it is not considered equivalent to E.

(I'm using E instead of Literal[SafeUUID.unsafe] as a placeholder for some enum literal)

It looks like we do consider them equivalent if E is a singleton, e.g. if you remove the B variant from the enum in https://play.ty.dev/f0d5d174-f462-4dc8-9c97-3e2cb339e09d

But Any & E | E = E should hold for every (fully static) type, I think? Any & E | E has a bottom materialization of E, and a top materialization of E.

[carljm]

Could this have been caused by astral-sh/ruff#20285 ? I don't think that was intended to change any behavior that would affect this case, but it did touch the definition of is_single_valued for enums, and it seems like quite coincidental timing.

[sharkdp]

I can take a look tomorrow.

[sharkdp]

This has nothing to do with the newly merged PR.

It's a bug that has probably been around for two months, since I introduced enum literals originally. I think we just got very lucky here to generate just the ((Any | E) & E, E) combination with the same enum literal appearing three times.

=> astral-sh/ruff#20324