Rustdoc should not fail to generate documentation when deep types are involved

[nagisa]

Ref. #62059

To fix the referenced issue we raised the recursion limit, however if there is a type that exceeds this limit again, we will start failing to generate documentation for the crate in question and all the dependent crates using these deep types again.

rustdoc uses the regular trait selection mechanism to check for auto-traits, so it is not too easy to make rustdoc to ignore those failures and not generate an entry for the implementation in rustdoc instead of erroring.

Ideal behaviour would be a warning during documentation generation and a missing entry in the "Implementors" section.

[GuillaumeGomez]

How is the compiler handling such a case? If we could use the same in rustdoc, that'd be even better than warn about it.

[nagisa]

The compiler requires user crates to increase the recursion limit manually, but unlike with rustdoc, you will only ever hit those cases in regular compilation when you actually use types in a certain way that would require such recursion. rustdoc is significantly more prone to such recursion.

[steveklabnik]

Triage: not aware of any changes here

[jyn514]

discussed at compiler team meeting. @eddyb raised the idea that rustdoc could just silently ignore the cases where we hit the existing limit, and treat them as unimplemented traits.

I wouldn't expect this to be possible currently, since rustc has a nasty habit of aborting the process when it hits the recursion limit: #79459