Confusing error involving lifetime elision

[ZainlessBrombie]

The following code:

struct A<'a> {
    backref: Option<&'a mut A<'a>>,
    value: String,
}

impl<'a> A<'a> {
    fn child_with_value(&'a mut self, value: String) -> A<'a> {
        A {
            backref: Some(self),
            value,
        }
    }
}

fn works(a: &mut A) {
    println!("{}", a.value);
}

fn errors(a: &mut A) {
    let mut b = a.child_with_value("foo".into());
    errors(&mut b);
}

Gives this error:

I expected to see this happen:
Given that the invocation should be legal, I expect this to compile without issues :)

Instead, this happened: Compiler error above. Note that the variable flows into itself, which seems strange

Meta

Tested both on stable and nightly:
rustc --version --verbose:

rustc 1.50.0-nightly (f0f68778f 2020-12-09)
binary: rustc
commit-hash: f0f68778f798d6d34649745b41770829b17ba5b8
commit-date: 2020-12-09
host: x86_64-unknown-linux-gnu
release: 1.50.0-nightly

Backtrace not applicable I think?

[camelid]

Backtrace not applicable I think?

Correct, you only need a backtrace if the compiler crashes.

---

I don't know enough about lifetimes to know whether this should work, but the diagnostics are definitely weird. Here's a more minimal example:

struct A<'a> {
    backref: Option<&'a mut A<'a>>,
}

fn child_with_value<'a>(this: &'a mut A<'a>) -> A<'a> {
    A {
        backref: Some(this),
    }
}

fn errors(a: &mut A) {
    let mut b = child_with_value(a);
}

(Playground)

Errors:

   Compiling playground v0.0.1 (/playground)
error[E0623]: lifetime mismatch
  --> src/lib.rs:12:34
   |
11 | fn errors(a: &mut A) {
   |              ------
   |              |
   |              these two types are declared with different lifetimes...
12 |     let mut b = child_with_value(a);
   |                                  ^ ...but data from `a` flows into `a` here

error: aborting due to previous error

For more information about this error, try `rustc --explain E0623`.
error: could not compile `playground`

To learn more, run the command again with --verbose.

[camelid]

It seems the reason my minimal example fails to compile is that the compiler is not able to infer the lifetimes properly. This fixes my minimal example:

struct A<'a> {
    backref: Option<&'a mut A<'a>>,
}

fn child_with_value<'a>(this: &'a mut A<'a>) -> A<'a> {
    A {
        backref: Some(this),
    }
}

fn errors<'a>(a: &'a mut A<'a>) {
    let mut b = child_with_value(a);
}

(Playground)

[camelid]

However, if I add explicit lifetimes to your original code, the compiler produces this error:

error[E0597]: `b` does not live long enough
  --> src/lib.rs:21:12
   |
19 | fn errors<'a>(a: &'a mut A<'a>) {
   |           -- lifetime `'a` defined here
20 |     let mut b = a.child_with_value("foo".into());
21 |     errors(&mut b);
   |     -------^^^^^^-
   |     |      |
   |     |      borrowed value does not live long enough
   |     argument requires that `b` is borrowed for `'a`
22 | }
   | - `b` dropped here while still borrowed

Now that I see the error, I think your code is actually incorrect. (Correct me if I'm wrong on that!) The diagnostics still need to be fixed though.

[ZainlessBrombie]

Yes my apologies, my code is incorrect - the problem is the incorrect message :)
Thank you!

[mbartlett21]

@camelid

Edit: moved to #79886

[camelid]

@mbartlett21 That example appears to trigger a different issue. Please open a new issue so we can discuss it separately.

[BGR360]

This seems like it would be a good candidate to apply the language of #90179 to.

Here's what I picture:

error[E0623]: lifetime mismatch
  --> src/lib.rs:12:34
   |
11 | fn errors(a: &mut A) {
   |              ------
   |              |
   |              these two types are declared with different lifetimes...
12 |     let mut b = child_with_value(a);
   |                                  ^ ...but data from `a` flows into `a` here
   = note: each elided lifetime in input position becomes a distinct lifetime
help: explicitly declare a lifetime and assign it to both
   |
11 | fn errors<'a>(a: &'a mut A<'a>) {
   |          ++++     ++      ++++

I think the key point that wasn't addressed in that PR is that there's an elided lifetime parameter on A, but that PR only acts upon elided lifetimes on references (or so it would seem).

cc @Nilstrieb

[Noratrieb]

sounds like a good idea, I'll look into that