Idea: Check variance by recursively expanding types and looking for invariant / contravariant types

[obi1kenobi]

Relevant reading: https://doc.rust-lang.org/nomicon/subtyping.html
Especially this section: https://doc.rust-lang.org/nomicon/subtyping.html#variance

TL;DR: For many types, a held T<'a> can be converted into T<'b> if 'a: 'b: we're agreeing to use the type "for shorter." If at a later time, the implementation of such a T gains (for example) an UnsafeCell field, it will lose the aforementioned property. We say that instead of covariant, it will have become invariant.

This is an extremely subtle way to break semver, and one for which downstream users have no recourse.

To craft an effective lint, we want to explain not just that the variance change happened but also which type(s) caused it. Ideally, we'd show the entire chain of types all the way down to the UnsafeCell or other field that causes invariance.

To be effective, this requires cross-crate analysis, which requires us to be able to effectively link and query rustdocs of multiple crates and the stdlib together.

[Nemo157]

Potentially this information could be provided by rustdoc for each generic type, we have rust-lang/rfcs#3515 about adding it to the UI.

[obi1kenobi]

Nice, TIL! That would certainly make this much easier to implement. Thanks!