New lint: Unnecessarily complex types

[jyn514]

It would be nice to lint on return types that are more complex than necessary. Take for example the following function (playground):

fn unnecessary_result() -> Result<usize, ()> {
    Ok(1)
}

The Result here is not necessary, because the function always returns an Ok variant.
I would also like to see a similar lint for fn f() -> Option<T> where f always returns Some.

Note this is not the same as https://rust-lang.github.io/rust-clippy/current/index.html#type_complexity, which is about return types that are hard to read, not return types that are unnecessary.

Possible output for the lint (just making this up here):

warning: unnecessary `Result` in return type
 --> src/lib.rs:1:1
   |
1  |  fn unnecessary_result() -> Result<usize, ()> {
   |                             ^^^^^^^^^^^^^^^^^ help: try removing the `Result` type: `usize`
2  |         Ok(1)
   |         ^^^^^ note: only return value is an `Ok` variant

[jfrimmel]

Sometimes such a construct is necessary, e.g. to stay SemVer-compatible (e.g. you changed the internals of a function, so that it now cannot fail anymore, but your users still expect a Result/Option).
Another possibility is, that you want to be future-proof: currently the implementation cannot fail, but it is expected to be so in the future.
I still think, that this might be a useful lint, but the SemVer hazard should be documented in "known problems".

[jyn514]

Maybe it could be 'allow' by default and you'd have to explicitly set it to warn.

[flip1995]

Definitely an allow-by-default lint. It may be hard to determine all the return points in the function and find out what exactly is returned. Making this generic over multiple types may also be quite hard.

[ghost]

unnecessary_filter_map already does something similar. It warns if the closure never returns None.