New lint: Use Reverse with sort_by_key

[camsteffen]

Suggesting a new lint to prefer using Reverse and sort_by_key.

Bad:

vec.sort_by(|a, b| b.foo().cmp(&a.foo()));

Good:

vec.sort_by_key(|e| Reverse(e.foo()));

It should work the same for sort_unstable_by / sort_unstable_by_key.

Lint name idea: sort_by_key_reverse

Maybe there are other cases where Reverse can simplify a sort?

[JarredAllen]

I can implement this.

[camsteffen]

I'm curious if this will work with a lambda that does any transformation to a and b? Does there need to be a complexity limit?

Another (contrived) example: vec.sort_by(|a, b| Ord::cmp(&d(e(b.f().g())), &d(e(a.f().g()))))

[JarredAllen]

I think it should be possible to lint anything where b and a get the same transformation, so long as it looks completely identical (so things like b+b vs. a*2 would throw it off, even though they'd actually do the same thing), never mentions the other variable (so things like b+0*a would also throw it off), and never calls a function which could change one of it's arguments (by taking an &mut or some such).

[phansch]

Closed with #5623