Rustdoc considers *const T and *mut T different

[jruderman]

https://doc.rust-lang.org/core/ptr/struct.Unique.html

is_null, offset, and as_ref are listed twice. This is confusing and makes the HTML invalid. I'm not sure what's going on here.

[alexcrichton]

The bug here is that *const T and *mut T are classified as the same kind of pointer, so when the Deref implementation is seen it inlines the inherent methods from those two pointer types, and both have an is_null method. This would likely be dealt by not having both pointer types be represented as the same type in rustdoc.

[pmarcelll]

Unique changed, so this specific issue is no longer relevant.

[steveklabnik]

So, in attempting to recreate this bug, I first looked at https://doc.rust-lang.org/1.0.0/core/ptr/struct.Unique.html to see what it looked like when Unique existed. I then used this code:

use std::ops::Deref;

pub struct Unique<T: ?Sized> {
    pointer: *const T,
}

impl<T:?Sized> Deref for Unique<T> {
    type Target = *mut T;

    #[inline]
    fn deref<'a>(&'a self) -> &'a *mut T {
        unsafe { mem::transmute(&*self.pointer) }
    }
}

This gives

with no methods at all! That seems... strange?

[jrvidal]

I spent some time looking into this, and there are a couple of intertwined issues.

• First, we don't seem to properly add the relevant Deref targets to the set of implementations that should be displayed, in particular when those targets are primitives. I tried the same example mentioned above and it works with slices and str, but not with any other primitive. This happens in the collect-trait-impls pass.
• Also, even when hacking a bit to make it find the relevant targets, we end up not displaying any methods:
  
  This happens because (almost?) all methods in primitives have by-value self, and we don't show those methods in "Methods from Deref" blocks.

So it seems that the first issue to solve is to make rustdoc display Deref-inherited methods with by-value self when the target is Copy. This, however, has been discussed before (#30607, #33396, #39550, #45645) and deemed too difficult.