&mut [T;N] as *mut T does not compile, but &[T;N] as *const T does

[jagourq]

unsafe {
    let my_num: &[i32; 2] = &[10, 20];
    let my_num: *const i32 = my_num as *const i32; // This works.
    println!("{}", *my_num.offset(1));
}

unsafe {
    let my_num: &mut [i32; 2] = &mut [10, 20];
    // error: mismatched types:
    //  expected `*mut i32`,
    //         found `&mut [i32; 2]`
    let my_num: *mut i32 = my_num as *mut i32;
    println!("{}", *my_num.offset(1));
}

[alexcrichton]

cc @nrc, is this a coercion that's expected to work?

[alexcrichton]

triage: I-nominated

[nrc]

The first cast is from &[i32; 2] to *const i32, the second is from &mut [i32; 2] to *mut i32. I'm not actually sure if this is meant to work, I guess it should. The intention here is to give a way to treat arrays like C-arrays and get a pointer to the start of the array. It seems that if the array is mutable, then the pointer should be too.

I recommend this is P-low - it is backwards compatible, rare, and just a bug by omission, really. The fix should be pretty easy though.

[alexcrichton]

Ah I wasn't sure if this was intended, if it's actually intended then I don't mind removing the nomination as well!

[nikomatsakis]

I'm not sure how intended this is, but it doesn't seem too harmful. ;)

[pnkfelix]

P-low, (perhaps "just a bug", perhaps a minor feature.)

[steveklabnik]

Triage: no change

[osa1]

This still happens today (with e32f372), but the error message is slightly different:

error[E0606]: casting `&mut [i32; 2]` as `*mut i32` is invalid
  --> test.rs:13:32
   |
13 |         let my_num: *mut i32 = my_num as *mut i32;
   |                                ^^^^^^^^^^^^^^^^^^

error: aborting due to previous error

#80258 is another *mut-related issue.