rustdoc regression: Lifetime in type alias definition is incorrectly elided

[camelid]

It'd be good to minimize this and reproduce this outside of rustc.

beta (incorrect)

stable (correct)

[inquisitivecrystal]

This bisects to #91430.

Stop looking at me like that! git bisect is still a thing! And no, before someone asks, I do not volunteer to do this sort of thing on a regular basis. It took all day. I did other stuff at the same time, but still...

[camelid]

Thank you so much! FYI, using cargo-bisect-rustc (with --prompt and -- doc --open) would've probably been a lot faster :)

[camelid]

cc @b-naber

[camelid]

I guess the issue here is that try_normalize_erasing_regions is erasing the lifetimes (=regions) in the type, hence its name.

cc @jackh726 as well

[GuillaumeGomez]

I'm working on it.

[inquisitivecrystal]

Thank you so much! FYI, using cargo-bisect-rustc (with --prompt and -- doc --open) would've probably been a lot faster :)

I feel very embarrassed. I figured that it wouldn't work, because generally cargo explodes whenever I try to use it on the compiler. I take it would actually work though?

[camelid]

Actually, I think you're right to use git bisect. cargo-bisect-rustc doesn't usually work on the compiler 🤦

Sorry about that!

[Mark-Simulacrum]

The underlying principle should work though, by specifying rustc = ... in config.toml pointing to a rustup-toolchain-install-master'd binary (and potentially cargo = and/or rustdoc =).