Add an allow-by-default lint against #[no_mangle]

[Nemo157]

Given that #[no_mangle] is unsafe there should be a lint that is capable of detecting it (there was a suggestion that unsafe_code should also detect it, but it might make sense to be able to lint it individually too).

[RalfJung]

So maybe unsafe_code should be a lint group, and include this new lint?

I am suggesting that because I think deny(unsafe_code) should suffice to trigger the new lint.

[Nemo157]

I've got the basic lint working now. @RalfJung do you think it's worth adding it as a separate lint, renaming unsafe_code to something else (which I can't think of a name for), and adding them both as lints in an unsafe_code group; or better to just integrate it into the current unsafe_code lint?

[RalfJung]

I'll defer to @rust-lang/wg-diagnostics for that :D

[estebank]

I am partial towards having a hard error when encountering multiple no_mangle functions with the same name in a given workspace, as originally suggested.

Adding this to unsafe_code might be problematic in the wild. Let's make a PR that rolls it into unsafe_code and see the behavior in the wild with a crater run, but if it causes too much fallout, lets make it an independent lint.

[Nemo157]

There's other ways of using no_mangle which create UB than just overriding within a workspace, e.g. the example in geiger-rs/cargo-geiger#104. While adding a hard error for cases that can be detected would be nice, I don't think it's possible in general.

[jyn514]

Removing I-prioritize since this is already tracked at #28179