Deriving only one of Hash/Eq considered harmful

[Gankra]

If you derive Hash, but implement (Partial)Eq manually, you're probably doing it wrong, because those two need to agree for e.g. HashMap to work right.

[Manishearth]

This basically needs to look for the automatic_derive annotation on impls and compare.

[apasel422]

@Manishearth I assume the best way to do this is to have LateLintPass::check_struct_def query the context for the presence of PartialEq and Hash impls for the struct/enum and then see if one contains the automatic_derive annotation and the other does not. We can extract this data from cx.tcx more efficiently than recomputing it ourselves, right?

[Manishearth]

Yes. I don't know exactly which side tables it's in the tcx, but the info should be there.

Actually, a check_item for partialeq impls might be easier.

[oli-obk]

rust-lang/rfcs#1499 suggests that this issue is not yet resolved, since only derived Hash + implemented PartialEq is linted, but not derived PartialEq + implemented Hash