Using &raw const instead of &raw mut produces invalid help message

[cyrgani]

Code

fn main() {
    let val = 2;
    let ptr = &raw const val;
    unsafe { *ptr = 3; }
}

Current output

error[E0594]: cannot assign to `*ptr`, which is behind a `*const` pointer
 --> src/main.rs:4:14
  |
4 |     unsafe { *ptr = 3; }
  |              ^^^^^^^^ `ptr` is a `*const` pointer, so the data it refers to cannot be written
  |
help: consider changing this to be a mutable pointer
  |
3 |     let ptr = &mut raw const val;
  |                +++

For more information about this error, try `rustc --explain E0594`.

Desired output

error[E0594]: cannot assign to `*ptr`, which is behind a `*const` pointer
 --> src/main.rs:4:14
  |
4 |     unsafe { *ptr = 3; }
  |              ^^^^^^^^ `ptr` is a `*const` pointer, so the data it refers to cannot be written
  |
help: consider changing this to be a mutable pointer
  |
3 |     let ptr = &raw mut val;
  |                    ~~~

For more information about this error, try `rustc --explain E0594`.

Rationale and extra context

The current help message is wrong:

fn main() {
    let mut val = 2;
    let ptr = &mut raw const val;
    unsafe { *ptr = 3; }
}

gives

error: expected one of `!`, `.`, `::`, `;`, `?`, `else`, `{`, or an operator, found keyword `const`
 --> a.rs:5:24
  |
5 |     let ptr = &mut raw const val;
  |                        ^^^^^ expected one of 8 possible tokens

whereas the correct syntax would be &raw mut val.

Other cases

No response

Rust Version

rustc 1.84.0-nightly (2024-10-20 662180b34d95f72d05b7)

Anything else?

EDIT: updated since addr_of! and addr_of_mut! diagnostics are not incorrect anymore due to #127675

[chenyukang]

I think the current suggestion is suggesting the code like this:

    let ptr = &mut val;

[chenyukang]

oh, the suggestion's code comes from here:

https://github.com/chenyukang/rust/blob/c8ed5c50edfd1b721c0541fa614864179251aaac/library/core/src/ptr/mod.rs#L2208

I think we should not suggest to change the library code here.

[tesuji]

I have playground link that represents the problem above: https://play.rust-lang.org/?version=nightly&mode=debug&edition=2021&gist=fee632ddd4f51a0390ac6918ef4eff96

However, when I add that code as a regression tests in tests/ui directory, the diagnostic suddenly changed to

error[E0594]: cannot assign to `*ptr`, which is behind a `*const` pointer
  --> $DIR/dont_suggest_raw_pointer_syntax-issue-127562.rs:9:9
   |
LL |         *ptr = 3;
   |         ^^^^^^^^ `ptr` is a `*const` pointer, so the data it refers to cannot be written
   |
help: consider specifying this binding's type
   |
LL |     let ptr: *mut i32 = std::ptr::addr_of!(val);
   |            ++++++++++

I don't have any ideas about this differences.

[chenyukang]

binding's type

This is because when run the test UI, the span here is remaped $SRC_DIR/core/src/ptr/mod.rs:LL:COL (#4):
https://github.com/chenyukang/rust/blob/b44a48416d06524da2ea247f1d1973d85ef514f8/compiler/rustc_borrowck/src/diagnostics/mutability_errors.rs#L1452

so the here we get an error and didn't continue the code branch:
https://github.com/chenyukang/rust/blob/b44a48416d06524da2ea247f1d1973d85ef514f8/compiler/rustc_borrowck/src/diagnostics/mutability_errors.rs#L1453

it's weird, maybe another issue need to be fixed.

[cyrgani]

Given that &raw syntax might get stabilized soon by #127679, suggesting the correct &raw mut syntax would become a more feasible option.

[RalfJung]

&mut raw const $place

This is completely invalid syntax so just suggesting nothing instead as a first step would already be a good idea.

Suggesting anything that involves references is a really bad idea -- code that uses raw pointers should keep using raw pointers throughout with no intermediate references.