suggested fix for undeclared hrtb lifetimes in where clauses in produces error "[E0316] nested quantification of lifetimes" when specified just before the trait

[polina4096]

Code

trait Trait<T>
  where T: for<'a> Trait<T> + 'b { }

Current output

error[E0261]: use of undeclared lifetime name `'b`
 --> src/lib.rs:2:31
  |
2 |   where T: for<'a> Trait<T> + 'b { }
  |                               ^^ undeclared lifetime
  |
  = note: for more information on higher-ranked polymorphism, visit https://doc.rust-lang.org/nomicon/hrtb.html
help: consider making the bound lifetime-generic with a new `'b` lifetime
  |
2 |   where for<'b> T: for<'a> Trait<T> + 'b { }
  |         +++++++
help: consider introducing lifetime `'b` here
  |
1 | trait Trait<'b, T>
  |             +++

Desired output

error[E0261]: use of undeclared lifetime name `'b`
 --> src/lib.rs:2:31
  |
2 |   where T: for<'a> Trait<T> + 'b { }
  |                               ^^ undeclared lifetime
  |
  = note: for more information on higher-ranked polymorphism, visit https://doc.rust-lang.org/nomicon/hrtb.html
help: consider making the bound lifetime-generic with a new `'b` lifetime
  |
2 |   where T: for<'a, 'b> Trait<T> + 'b { }
  |                  ++++
help: consider introducing lifetime `'b` here
  |
1 | trait Trait<'b, T>
  |             +++

Rationale and extra context

HRTB lifetimes can be put not only before the constrained type (for<...> T: ...), but also just before the constraining trait (T: for<...> Trait ...). When in this position, compiler suggests code which produces error "[E0316] nested quantification of lifetimes" as it ignores the existing HRTB just before the trait and only looks at constraint HRTB.

Other cases

No response

Rust Version

$ rustc --version --verbose
rustc 1.76.0 (07dca489a 2024-02-04)
binary: rustc
commit-hash: 07dca489ac2d933c78d3c5158e3f43beefeb02ce
commit-date: 2024-02-04
host: aarch64-apple-darwin
release: 1.76.0
LLVM version: 17.0.6

Anything else?

No response

[surechen]

I think this should suggest :

trait Trait<'b, T>
  where T: for<'a> Trait<'b, T> + 'b { }

Because 'b is bound for T

[surechen]

@rustbot claim

[polina4096]

I think this should suggest :

trait Trait<'b, T>
  where T: for<'a> Trait<'b, T> + 'b { }

Because 'b is bound for T

It already does suggest that, I think. There are two suggestions.

help: consider making the bound lifetime-generic with a new `'b` lifetime
  |
2 |   where T: for<'a, 'b> Trait<T> + 'b { }
  |                  ++++
help: consider introducing lifetime `'b` here
  |
1 | trait Trait<'b, T>
  |             +++

[fmease]

Desired output

2 |   where T: for<'a, 'b> Trait<T> + 'b { }
  |                  ++++

where T: for<'a, 'b> Trait<T> + 'b isn't valid either tho b/c for<> after : only universally quantifies (i.e., introduces universal parameters for) the trait ref immediately following the binder meaning 'b in + 'b is still undefined / unbound.

We should keep the preexisting trait Trait<'b, T> suggestion (which can lead to follow-up errors unfortunately) and maybe suggest:

- where T: for<'a> Trait<T> + 'b
+ where for<'a, 'b> T: Trait<T> + 'b

removing the inner binder (which might be a bit harder to do). The outer binder to the left of the : quantifies the entire bound.