Lint for `&String::into(): String` on final use of input string

[pnkfelix]

PR #59825 added a impl From<&String> for String.

While perusing the rustc source, I am now seeing instances of String values being constructed and then passed via-reference into a function that wants a String. These would not have compiled without the change added by PR #59825.

It might be worthwhile to try to make a lint that detects the last use of a String that is then being used as an argument to & which is then itself passed to this conversion method (which effectively hides the clone taking place.

(You can see discussion of the hidden clone on #59827; I'm not seeking debate on the merits of PR #59825. I just am wondering if we could easily highlight the unnecessary clones via a lint.)

[awestlake87]

Yeah, I'm not sure I like that change. I always liked that potentially expensive operations like clone() were explicit and searchable. Also, I usually think of .into() and ::from as a move, not a copy, since it consumes the callee.

In C++, it's extremely easy to make copies accidentally and while impl From<&String> for String might be reasonably harmless, I feel like an impl <T: Clone> From<&T> for T kinda pushes Rust too far in that direction.

If that feature is here to stay, then I would probably always enable this lint as an error in my projects