Underscore-named parameter makes hint point at invalid token

[Patryk27]

Hi,

I've got following code:

fn foo1<T: ?Sized>(_: T) {
    // here be dragons
}

fn foo2<T>(_: T) where T: ?Sized {
    // here be dragons
}

... that returns:

error[E0277]: the size for values of type `T` cannot be known at compilation time
 --> src/lib.rs:1:26
  |
1 | fn foo1<T: ?Sized>(_: T) {
  |         --               ^ doesn't have a size known at compile-time

error[E0277]: the size for values of type `T` cannot be known at compilation time
 --> src/lib.rs:5:18
  |
5 | fn foo2<T>(_: T) where T: ?Sized {
  |                  ^              - help: consider further restricting type parameter `T`: `, T: std::marker::Sized`
  |                  |
  |                  doesn't have a size known at compile-time

(playground)

In both cases the tip points at the very next token after the parameter list, instead of the underscore. Additionally, the foo2's help message (consider further restricting ...) has broken formatting.

I'd like to fix this myself, if only you'd give me a hint where to look :-)

Thanks;

[hellow554]

Interesting enough:

Before 1.15 ?Sized isn't valid.
From 1.15 to 1.19 this compiles just fine
Beginning with 1.20 the alignment is wrong.

[VirrageS]

@Patryk27 If you still want to look into this problem you may find this diff helpful:

diff --git a/src/librustc/traits/error_reporting.rs b/src/librustc/traits/error_reporting.rs
index 96d42dd73b1..0e021d658aa 100644
--- a/src/librustc/traits/error_reporting.rs
+++ b/src/librustc/traits/error_reporting.rs
@@ -1176,7 +1176,7 @@ impl<'a, 'tcx> InferCtxt<'a, 'tcx> {
                             // `fn foo<T>(t: T) where T: Debug {}` →
                             // `fn foo<T>(t: T) where T: Debug, T: Trait {}`
                             err.span_suggestion(
-                                generics.where_clause.span().unwrap().shrink_to_hi(),
+                                generics.where_clause.predicates, // TODO: fixme
                                 &format!(
                                     "consider further restricting type parameter `{}`",
                                     param_name,
diff --git a/src/librustc_typeck/check/mod.rs b/src/librustc_typeck/check/mod.rs
index 9107e993311..83a90e64eb1 100644
--- a/src/librustc_typeck/check/mod.rs
+++ b/src/librustc_typeck/check/mod.rs
@@ -1229,7 +1229,7 @@ fn check_fn<'a, 'tcx>(
         // for simple cases like `fn foo(x: Trait)`,
         // where we would error once on the parameter as a whole, and once on the binding `x`.
         if param.pat.simple_ident().is_none() && !fcx.tcx.features().unsized_locals {
-            fcx.require_type_is_sized(param_ty, decl.output.span(), traits::SizedArgumentType);
+            fcx.require_type_is_sized(param_ty, param.pat.span, traits::SizedArgumentType);
         }

         fcx.write_ty(param.hir_id, param_ty);

I've changed the decl.output.span() to param.pat.span and now it seems working correctly:

error[E0277]: the size for values of type `T` cannot be known at compilation time
  --> $DIR/unsized-fn-param.rs:1:20
   |
LL | fn foo1<T: ?Sized>(_: T) {
   |         --         ^ doesn't have a size known at compile-time
   |         |
   |         help: consider further restricting this bound: `T: std::marker::Sized +`
   |
   = help: the trait `std::marker::Sized` is not implemented for `T`
   = note: to learn more, visit <https://doc.rust-lang.org/book/ch19-04-advanced-types.html#dynamically-sized-types-and-the-sized-trait>
   = note: all function arguments must have a statically known size
   = help: unsized locals are gated as an unstable feature

as far for the HELP message in foo2 it probably needs iterating through generics.where_clause.predicates finding right predicate with correct name and print the span for it (if not found we should probably default generics.where_clause.span().unwrap().shrink_to_hi() or to generics.where_clause.span().unwrap().shrink_to_lo() or maybe find a better heuristic).

I'm not sure if these changes don't break other tests so be sure to run whole suite.

[Patryk27]

Great - thanks! :-)