Scope prints in diagnostics

[da-x]

• Rendered

Summary

The proposed change is to add context information in warnings and errors, so that not only source positions are printed, but also item names. For example, in fn main, or in fn mod::function, or in struct Foo. Also, relevant lines in annotation are added (but not repetitively).

error[E0124]: field `x` is already declared
 --> test.rs:9:5  in struct Foo
  |
1 | struct Foo {
...
6 |     x: u64,
  |     ------ `x` first declared here
...
9 |     x: u64,
  |     ^^^^^^ field already declared

Instead of:

error[E0124]: field `x` is already declared
 --> test.rs:9:5
  |
6 |     x: u64,
  |     ------ `x` first declared here
...
9 |     x: u64,
  |     ^^^^^^ field already declared

Related links

• Pre-RFC discussion
• Draft implementation (based on 1.38.0)

[estebank]

We can already do the following:

error[E0124]: field `x` is already declared
 --> test.rs:9:5  in struct Foo
  |
1 | struct Foo {
           ---
...
6 |     x: u64,
  |     ------ `x` first declared here
...
9 |     x: u64,
  |     ^^^^^^ field already declared

but it needs to be done on every error manually. Although this approach requires more effort, it also is more targeted, which means that we only show it when relevant. The devil's in the details, but the implementation of this seems "hard".

[da-x]

@estebank yeah, it's hard. There's also the issue of avoiding redundant scope prints when we have two or more different types of errors in the same scope [1], which probably means tracking scopes and perform unification of several unrelated errors based on their scope. Perhaps this can be done outside the call sites to emit diagnostics (i.e. only in librustc_errors), so that changes would not need to be tree-wide, making them easier. However, there's the issue of how to actually format the prints in that case. I wasn't able yet to come up with something sensible enough to put in the RFC as an alternative, but I'd like to mention it nonetheless.

On the bright side, simply avoiding duplicates in EmitterWriter based on Span, as I've done in the draft implementation is an easy way out implementation-wise, the result [2] may be good enough, perhaps.

[1]

error[E0124]: field `x` is already declared
 --> test.rs:9:5
  |
6 |     x: u64,
  |     ------ `x` first declared here
...
9 |     x: u64,
  |     ^^^^^^ field already declared

error[E0107]: wrong number of type arguments: expected 1, found 0
  --> test.rs:10:8
   |
10 |     z: Option<>,
   |        ^^^^^^^^ expected 1 type argument

[2]

error[E0124]: field `x` is already declared
 --> test.rs:9:5  in struct Foo
  |
1 | struct Foo {
...
6 |     x: u64,
  |     ------ `x` first declared here
...
9 |     x: u64,
  |     ^^^^^^ field already declared

error[E0107]: wrong number of type arguments: expected 1, found 0
  --> test.rs:10:8  in struct Foo
   |
10 |     z: Option<>,
   |        ^^^^^^^^ expected 1 type argument

[oli-obk]

cc @rust-lang/wg-diagnostics

[wesleywiser]

I realize this RFC is quite old at this point but in case anyone is interested in this, I agree with @estebank's comment. If there are specific errors where this context would be helpful, I think we would be willing to just review PRs that target those specific errors. For more general or wide ranging changes, I think it would be best to open an MCP to discuss those changes.

[estebank]

I'm going to close this RFC in favor of more targeted MCPs for specific error types.