Name similarity metric should consider ordering of name parts being swapped

[meithecatte]

Code

fn main() {
    dbg!(u16::from_bytes_le([21, 37]));
}

Current output

error[E0599]: no function or associated item named `from_bytes_le` found for type `u16` in the current scope
 --> src/main.rs:2:15
  |
2 |     dbg!(u16::from_bytes_le([21, 37]));
  |               ^^^^^^^^^^^^^ function or associated item not found in `u16`
  |
help: there is an associated function `from_be` with a similar name
  |
2 -     dbg!(u16::from_bytes_le([21, 37]));
2 +     dbg!(u16::from_be([21, 37]));
  |

Desired output

error[E0599]: no function or associated item named `from_bytes_le` found for type `u16` in the current scope
 --> src/main.rs:2:15
  |
2 |     dbg!(u16::from_bytes_le([21, 37]));
  |               ^^^^^^^^^^^^^ function or associated item not found in `u16`
  |
help: there is an associated function `from_le_bytes` with a similar name
  |
2 -     dbg!(u16::from_bytes_le([21, 37]));
2 +     dbg!(u16::from_le_bytes([21, 37]));
  |

Rationale and extra context

No response

Other cases

Rust Version

current nightly on play.rust-lang.org

Anything else?

No response

[jhpratt]

As the person who most recently (to my knowledge) touched the algorithm for similarity, I think this is feasible. My immediate thought:

Take two branches. One is the status quo. The other branch will split the identifier into "words" based on naming conventions (i.e. snake_case, UpperPascal, and SCREAMING_CASE). Run the same algorithm based on this. If and only if the latter (newly-introduced) method is exactly one and the former is greater than one, prefer the latter. Otherwise, maintain the status quo.

This seems slightly complicated as I write it out, but I'm certain it wouldn't be hard to implement. My primary concern is performance, as it would be adding a fair amount of overhead to every similarity diagnostic.

cc @estebank — thoughts on above?

[estebank]

@jhpratt I think what you describe should be ok.

[jhpratt]

A quick attempt at coding up the above is almost entirely negative, so some more nuance will be necessary. I'm going to fiddle around some to see what I can get.