Bad error message trying to use Box as a tuple struct

struct U <T> {
  wtf: Option<Box<U<T>>>
}
fn main() {
 U{ wtf: Some(Box(U{ wtf: None })) };
}

<anon>:5:19: 5:33 error: mismatched types:
 expected `core::ptr::Unique<_>`,
    found `U<_>`
(expected struct `core::ptr::Unique`,
    found struct `U`) [E0308]
<anon>:5  U{ wtf: Some(Box(U{ wtf: None })) };
                           ^~~~~~~~~~~~~~
<anon>:5:15: 5:34 error: mismatched types:
 expected `Box<U<_>>`,
    found `alloc::boxed::Box<_>`
(expected box,
    found struct `alloc::boxed::Box`) [E0308]
<anon>:5  U{ wtf: Some(Box(U{ wtf: None })) };
                       ^~~~~~~~~~~~~~~~~~~
error: aborting due to 2 previous errors
playpen: application terminated with error code 101

Could someone change the bad title? I am not good for english.

[Gankra]

Seems like more of a diagnostics issue. Box(..) is a private constructor, but we're erroring out on type-checking before we realize this.

The fix for this specific code is to replace Box(..) with Box::new(..)

[steveklabnik]

Triage: this now has a different, yet still confusing, error:

foo.rs:1:10: 1:11 error: parameter `T` is never used [E0392]
foo.rs:1 struct U<T> {
                  ^
foo.rs:1:10: 1:11 help: run `rustc --explain E0392` to see a detailed explanation
foo.rs:1:10: 1:11 help: consider removing `T` or using a marker such as `core::marker::PhantomData`

[Mark-Simulacrum]

That error is correct; but either way, the other error you get today is this, which while not ideal, is possibly the best we can do. I guess we could look for a new method with a single parameter? But that seems hacky.

error[E0423]: expected function, found struct `Box`
 --> test.rs:5:18
  |
5 |     U{ wtf: Some(Box(U{ wtf: None })) };
  |                  ^^^
  |                  |
  |                  did you mean `Box { /* fields */ }`?
  |                  constructor is not visible here due to private fields

error: aborting due to previous error

[estebank]

I guess we could look for a new method with a single parameter? But that seems hacky.

It is, but it is a common enough pattern and such a big enhancement in end user experience that I feel it'd be worth it to add it, specially when the ctor is not public. Then, the output should be:

error[E0423]: expected function, found struct `Box`
 --> test.rs:5:18
  |
5 |     U{ wtf: Some(Box(U{ wtf: None })) };
  |                  ^^^ help: did you mean to use the `new` method?: `Box::new`

error: aborting due to previous error

Similar output, but not quite the same cause in #22692.

[estebank]

In order to implement this there's a bit of roadblock at the moment. The verification if a struct has a given method is done in rustc_typeck, while this error is being emitted in src/librustc_resolve/lib.rs. I believe the error is being bubbled up, but it would have to be delayed on its entirety to the type check step in order to verify wether Box::new exists. :-/

You can look for traits that have a given method (get_traits_containing_item), but when I tried it with new it returned an empty vec. We could provide with a generic suggestion about trying out new, but I would much rather prefer if we could have some certainty that the method exists instead of potentially misleading the user.

[estebank]

Current output:

error[E0423]: expected function, found struct `Box`
 --> src/main.rs:5:15
  |
5 |  U{ wtf: Some(Box(U{ wtf: None })) };
  |               ^^^ constructor is not visible here due to private fields
help: possible better candidates are found in other modules, you can import them into scope
  |
1 | use syn::Expr::Box;
  |
1 | use syn::Pat::Box;
  |
1 | use syn::token::Box;
  |

error[E0392]: parameter `T` is never used
 --> src/main.rs:1:11
  |
1 | struct U <T> {
  |           ^ unused type parameter
  |
  = help: consider removing `T` or using a marker such as `std::marker::PhantomData`

error: aborting due to 2 previous errors