NLL: "explicit lifetime required" error fails to highlight usage that introduces constraint

[pnkfelix]

There are a collection of tests where NLL is failing to highlight the expression relevant to debugging some lifetime error.

Here is the list of tests that pnkfelix believes falls into that category:

• ui/lifetime-errors/ex1-return-one-existing-name-early-bound-in-struct.nll.stderr
• ui/lifetime-errors/ex1-return-one-existing-name-if-else-2.nll.stderr
• ui/lifetime-errors/ex1-return-one-existing-name-if-else-3.nll.stderr
• ui/lifetime-errors/ex1-return-one-existing-name-if-else-using-impl-2.nll.stderr
• ui/lifetime-errors/ex1-return-one-existing-name-if-else-using-impl-3.nll.stderr
• ui/lifetime-errors/ex1-return-one-existing-name-if-else-using-impl.nll.stderr
• ui/lifetime-errors/ex1-return-one-existing-name-if-else.nll.stderr

Here's an explanation of the kind of (mis)behavior being described here:

In the following test ( ex1-return-one-existing-name-if-else-3.rs ):

fn foo<'a>((x, y): (&'a i32, &i32)) -> &'a i32 {
    if x > y { x } else { y } //~ ERROR explicit lifetime
}

here is the error output from AST borrowck ( ex1-return-one-existing-name-if-else-3.stderr )

error[E0621]: explicit lifetime required in parameter type
  --> $DIR/ex1-return-one-existing-name-if-else-3.rs:12:27
   |
LL | fn foo<'a>((x, y): (&'a i32, &i32)) -> &'a i32 {
   |            ------ consider changing type to `(&'a i32, &'a i32)`
LL |     if x > y { x } else { y } //~ ERROR explicit lifetime
   |                           ^ lifetime `'a` required

while this is the error output from NLL: ( ex1-return-one-existing-name-if-else-3.nll.stderr ):

warning: not reporting region error due to nll
  --> $DIR/ex1-return-one-existing-name-if-else-3.rs:12:27
   |
LL |     if x > y { x } else { y } //~ ERROR explicit lifetime
   |                           ^

error[E0621]: explicit lifetime required in parameter type
  --> $DIR/ex1-return-one-existing-name-if-else-3.rs:11:13
   |
LL | fn foo<'a>((x, y): (&'a i32, &i32)) -> &'a i32 {
   |            -^----
   |            ||
   |            |lifetime `'a` required
   |            consider changing type to `(&'a i32, &'a i32)`

error: aborting due to previous error

Note in particular that the error is failing to report the expression (namely the occurrence of y in an expression context that is expecting a &'a i32) that is injecting the lifetime constraint that lead it to suggest the change to the parameter type.

(It seems likely that the span of interest is probably immediately available, given that the warning we emit about not reporting a region error does highlight the span in which we are interested.)

[nikomatsakis]

Marking as deferred as this is sort of a subissue of #51188

[nikomatsakis]

This still seems to be a problem. Since we are closing #51188, I'm pulling this out into its own issue and putting it on the RC milestone, though it may be that it's more of a "final release" level issue still.