[NLL] Lifetimes errors in a function can mask later errors in same function

[marmeladema]

For test src/test/ui/regions/region-multiple-lifetime-bounds-on-fns-where-clause.rs, under NLL, a lifetime may not live long enough error is missing.
This is a behavior change which should be investigated.

See comment posted by @jackh726 in #96212 (comment)

[marmeladema]

@rustbot label +A-NLL +NLL-diagnostics

[m-ou-se]

Reduced test case:

fn b<'a, 'b, 'c>(x: &mut &'a isize, y: &mut &'b isize, z: &mut &'c isize) {
    *x = *y;
    *z = *y;
}

Rustc output:

error[E0623]: lifetime mismatch
 --> src/main.rs:2:10
  |
1 | fn b<'a, 'b, 'c>(x: &mut &'a isize, y: &mut &'b isize, z: &mut &'c isize) {
  |                          ---------          --------- these two types are declared with different lifetimes...
2 |     *x = *y;
  |          ^^ ...but data from `y` flows into `x` here

error[E0623]: lifetime mismatch
 --> src/main.rs:3:10
  |
1 | fn b<'a, 'b, 'c>(x: &mut &'a isize, y: &mut &'b isize, z: &mut &'c isize) {
  |                                             ---------          ---------
  |                                             |
  |                                             these two types are declared with different lifetimes...
2 |     *x = *y;
3 |     *z = *y;
  |          ^^ ...but data from `y` flows into `z` here

error: aborting due to 2 previous errors

Rustc output with -Z borrowck=mir:

error: lifetime may not live long enough
 --> src/main.rs:2:5
  |
1 | fn b<'a, 'b, 'c>(x: &mut &'a isize, y: &mut &'b isize, z: &mut &'c isize) {
  |      --  -- lifetime `'b` defined here
  |      |
  |      lifetime `'a` defined here
2 |     *x = *y;
  |     ^^^^^^^ assignment requires that `'b` must outlive `'a`
  |
  = help: consider adding the following bound: `'b: 'a`

error: aborting due to previous error

It doesn't give an error about the *z = *y; line.

However, if you remove the *x = *y; line, it doesn't succeed. It then produces the right error:

error: lifetime may not live long enough
 --> src/main.rs:2:5
  |
1 | fn b<'a, 'b, 'c>(x: &mut &'a isize, y: &mut &'b isize, z: &mut &'c isize) {
  |          --  -- lifetime `'c` defined here
  |          |
  |          lifetime `'b` defined here
2 |     *z = *y;
  |     ^^^^^^^ assignment requires that `'b` must outlive `'c`
  |
  = help: consider adding the following bound: `'b: 'c`

So, this doesn't seem like a case where invalid code is accepted. Just a case where diagnostics are incomplete.

[m-ou-se]

When using two unrelated variables/lifetimes, it does produce both errors:

fn b<'a, 'b, 'c, 'd>(x: &mut &'a isize, y: &mut &'b isize, z: &mut &'c isize, z2: &mut &'d isize) {
    *x = *y;
    *z = *z2;
}

[m-ou-se]

When doing *x = *y; *x = *y; (the same thing twice), you only get the error once. It seems that in the new mode, that 'deduplicating' effect also applies to *x = *y; *z = *y;.

(*x = *y; *y = *z; (last assignment flipped) does produce two errors with and without nll.)

[jackh726]

Thanks @m-ou-se for looking at this!

I think based on this, it should be okay to move forward without "fixing" this. Seems like only a diagnostics thing instead of accidently allowing more code.