regression: rustc suggests `.as_ref()` at incorrect location and other spans have regressed

[camelid]

Code

Playground (regressed rustc version)

fn foo(opt: &Option<Box<i32>>) -> String {
    opt.map(|x| x.to_string()).unwrap_or_else(String::new)
}

Output (stable 1.56.0, good)

error[E0507]: cannot move out of `*opt` which is behind a shared reference
 --> src/lib.rs:2:5
  |
2 |     opt.map(|x| x.to_string()).unwrap_or_else(String::new)
  |     ^^^ move occurs because `*opt` has type `Option<Box<i32>>`, which does not implement the `Copy` trait
  |
help: consider borrowing the `Option`'s content
  |
2 |     opt.as_ref().map(|x| x.to_string()).unwrap_or_else(String::new)
  |        +++++++++

For more information about this error, try `rustc --explain E0507`.

Output (beta 1.57.0-beta.2, incorrect)

error[E0507]: cannot move out of `*opt` which is behind a shared reference
 --> src/lib.rs:2:5
  |
2 |     opt.map(|x| x.to_string()).unwrap_or_else(String::new)
  |     ^^^^^^^^^^^^^^^^^^^^^^^^^^ move occurs because `*opt` has type `Option<Box<i32>>`, which does not implement the `Copy` trait
  |
help: consider borrowing the `Option`'s content
  |
2 |     opt.map(|x| x.to_string()).as_ref().unwrap_or_else(String::new)
  |                               +++++++++

For more information about this error, try `rustc --explain E0507`.

[camelid]

My hunch is that this regression is caused by the change in error spans between stable and beta, which also feels like a regression to me since it makes the span less specific:

stable:

2 |     opt.map(|x| x.to_string()).unwrap_or_else(String::new)
  |     ^^^ move occurs because `*opt` has type `Option<Box<i32>>`, which does not implement the `Copy` trait

beta:

2 |     opt.map(|x| x.to_string()).unwrap_or_else(String::new)
  |     ^^^^^^^^^^^^^^^^^^^^^^^^^^ move occurs because `*opt` has type `Option<Box<i32>>`, which does not implement the `Copy` trait

[WaffleLapkin]

searched nightlies: from nightly-2021-09-29 to nightly-2021-10-21
regressed nightly: nightly-2021-10-01
searched commit range: 1149193...aa7aca3
regressed commit: 4aa7879

bisected with cargo-bisect-rustc v0.6.1

Host triple: x86_64-unknown-linux-gnu
Reproduce with:

cargo bisect-rustc --preserve --test-dir=tstdr --end=2021-10-21 --script=./test.sh

test.sh:

#!/bin/sh

cargo check 2>&1 | grep " ^^^ move occurs because"

[camelid]

cc @Aaron1011 @estebank

The new spans caused this suggestion regression, and I feel like they make the errors harder to read in some cases.

[Aaron1011]

@camelid: Setting aside the incorrect suggestion for the moment, I think the new span is better. By itself, pointing to opt does not explain why a move is occuring, as not every usage of a variable name is a move. Pointing to the the method call helps emphasize that the method call is responsible for the move.

However, I think adding a note could help emphasize that the method is causing the move (we already do this for other borrowcheck errors).

The incorrect suggestion is tricky - I'll see if I can do something about that.

[camelid]

Setting aside the incorrect suggestion for the moment, I think the new span is better. By itself, pointing to opt does not explain why a move is occuring, as not every usage of a variable name is a move. Pointing to the the method call helps emphasize that the method call is responsible for the move.

I understand, and I can see how highlighting the whole call is more correct in some ways, but then a lot of code is highlighted, which can be visually distracting. I wonder if it could make sense to highlight just the method name?

opt.map(|x| x.to_string())
^^^^^^^

Although that looks a bit weird. Another option is something like this:

opt.map(|x| x.to_string())
^^^ --- `opt` moved by `map` here

Additionally, a large part of why I feel the new span is visually distracting is that move occurs because `*opt` has type `Option<Box<i32>>`, which does not implement the `Copy` trait takes up a lot of space as a span_label with the new span. Perhaps the new span could be made easier to read if the label were moved to a note?