NLL: need to capture and present causes that led to region constraints

[pnkfelix]

Consider the test in-band-lifetimes/impl/dyn-trait.rs.

• An aside: Normally pnkfelix would copy the source text inline, but in this case the source text is testing a totally different bug and will probably obscure the main point I am making here.
  • There is probably a much smaller test case that would also succeed in making the point here; I may try to come up with such a minimal test, and post it here if I do.

For AST-borrowck we produce this output dyn-trait.stderr:

error[E0495]: cannot infer an appropriate lifetime due to conflicting requirements
  --> $DIR/dyn-trait.rs:32:16
   |
LL |     static_val(x); //~ ERROR cannot infer
   |                ^
   |
note: first, the lifetime cannot outlive the lifetime 'a as defined on the function body at 31:1...
  --> $DIR/dyn-trait.rs:31:1
   |
LL | fn with_dyn_debug_static<'a>(x: Box<dyn Debug + 'a>) {
   | ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
   = note: ...so that the expression is assignable:
           expected std::boxed::Box<std::fmt::Debug>
              found std::boxed::Box<std::fmt::Debug + 'a>
   = note: but, the lifetime must be valid for the static lifetime...
   = note: ...so that the types are compatible:
           expected StaticTrait
              found StaticTrait

error: aborting due to previous error

but for NLL we currently produce this output dyn-trait.nll.stderr:

error: free region `'a` does not outlive free region `'static`
  --> $DIR/dyn-trait.rs:32:5
   |
LL |     static_val(x); //~ ERROR cannot infer
   |     ^^^^^^^^^^^^^

error: aborting due to previous error

One can debate how much utility the end user gets from the potential overload of lines that the AST-borrowck note generates, but it is certainly more useful than the output that NLL is currently generated.

At its heart, the issue here is that NLL currently builds up a set of constraints and tells us if it couldn't resolve them, but it fails to present the causes that led to that constraint set.

[nikomatsakis]

We have to get a handle on this. @davidtwco and I have been working on it.

[nikomatsakis]

Related issues:

• NLL: "explicit lifetime required" error fails to highlight usage that introduces constraint  #51175
• NLL injects type infer failures in presence of subsequently-masked lifetime errors #51172

[pnkfelix]

Things have gotten a lot better here.

The only real diagnostic issue remaining with https://github.com/rust-lang/rust/blob/master/src/test/ui/in-band-lifetimes/impl/dyn-trait.nll.stderr is that it talks about "closures" when there is no closure in the source code from the view point of the end developer.

[davidtwco]

@pnkfelix With #52648 merged, errors now correctly refer to "closure" or "function" where appropriate.

[Mark-Simulacrum]

@rust-lang/wg-compiler-nll It looks like this issue might be done? Perhaps someone could take a look and update the current state.

[nikomatsakis]

I'm going to close this issue now. I re-opened #51175 which seem relevant.