Rustdoc: Allow implicit `fn main() -> Result<_, _>`

[llogiq]

Rustdoc now allows us to use ? in our tests. Unfortunately, this requires writing the fn main() -> .. manually, which really shouldn't be needed IMHO.

So we now have to write:

/// ```rust
///# fn main() -> Result<(), ()> {
/// let foo = Ok(1usize);
/// assert!(foo? > 0);
///# }
/// ```

where it should suffice to write:

/// ```rust
/// let foo = Ok(1usize);
/// assert!(foo? > 0);
/// ```

Doing so will make the tests shorter and hopefully lead to more ? and less unwrap in tests.

[Centril]

Feels like a no-brainer to me; would be great. :)

[scottmcm]

I think there are two things we want before doing this:

• try fn, so that it can always be try fn main() -> TBD {, and existing code will just work.
• The Try re-do, so we can make a type that accepts any error type (since it'll just error anyway).

[sinkuu]

fn main() -> impl std::process::Termination?

[llogiq]

Yes, that's what I'm thinking, and I'm setting up a PR for that. However, it's still unstable.

[QuietMisdreavus]

I talked about this back in March on the "? in main" tracking issue: #43301 (comment)

That sample you gave isn't even valid as-is, unless () coercing to Ok(()) is also in the pipeline?

[llogiq]

That'll be next, hopefully. I wonder if it suffices to impl From<T> for <Result<T, !>> or if we need to keep a generic error type (which might mess with type inference). In any event, having both the aforementioned and impl From<Result<T, !>> for T should be a no-brainer, right?