NLL diagnostics replaced nice closure errors w/ indecipherable free region errors

[pnkfelix]

The following tests all have a common weakness around how they report errors under NLL. Under AST borrowck they would say something somewhat understandable about closures. But under NLL they don't anymore.

List:

• borrowck/issue-45983.rs
• borrowck/issue-7573.rs
• borrowck/regions-escape-bound-fn-2.rs
• borrowck/regions-escape-bound-fn.rs
• borrowck/regions-escape-unboxed-closure.rs
• closure-expected-type/expect-region-supply-region.rs
• error-codes/E0621-does-not-trigger-for-closures.rs

(This list of tests is drawn from an informal paper document that I have been using to keep notes for myself as I work on this...)

Example (from ui/borrowck/issue-45983.rs):

AST borrowck says:

error: borrowed data cannot be stored outside of its closure
  --> $DIR/issue-45983.rs:17:27
   |
LL |     let x = None;
   |         - borrowed data cannot be stored into here...
LL |     give_any(|y| x = Some(y));
   |              ---          ^ cannot be stored outside of its closure
   |              |
| ...because it cannot outlive this closure

NLL says:

error: free region `` does not outlive free region `'_#2r`
  --> $DIR/issue-45983.rs:17:27
   |
LL |     give_any(|y| x = Some(y));
   |                           ^

[pnkfelix]

there is also the broader issue #49397; this one is more targeted and thus separate.

[KiChjang]

As @nikomatsakis has mentioned in the NLL triage docs, I was indeed looking at these errors, so I'm assigning this to myself.

[nikomatsakis]

I'm unassigning @KiChjang as I don't think they made any progress (ping me on Zulip @KiChjang if you still want to hack on this). I think that this is highly related to what I wrote on #51168 (that is, in this comment), and that the solution I described there would be good here too. So I'm going to mark this as a sub-issue and hence deferred.

[nikomatsakis]

Actualy, I'll make this the primary issue, since it has an amusing title. Subissues:

• NLL: missing note that "borrowed value needs to live until here" (sometimes?) #51168

[nikomatsakis]

The reason for this is indeed the closure, I think. We detect the error when analyzing the closure body, so it is hard for us to report a span that is outside the closure. I think this will be pretty tricky to fix and is maybe not that important. What we could probably do is take a different tactic: we could explain that x is owned by the closure and yet we are returning a reference to it (just as we would a top-level function, perhaps).

I'm imagining:

error[E0597]: `x` is owned by the closure, so you cannot return a reference to it
  --> $DIR/issue-11925.rs:18:35
   |
LL |         let f = to_fn_once(move|| &x); //~ ERROR does not live long enough
   |                                   ^
   |                                   |
   |                                   |
| cannot return reference to data owned by the closure

for bonus points, we can highlight the move keyword and/or the use of x that caused it to a move in the first place (we have that data available somewhere).