Request: add, sub, and offset for NonNull

[Lokathor]

Currently, offsetting a non-null pointer is quite cumbersome. It would be better if these pointer methods were provided directly on the NonNull type instead of having to get the inner pointer, do the change, and then call NonNull::new_unchecked on that output.

[fee1-dead]

offsetting/subtracting a NonNull pointer needs to return an Option<NonNull>, since the offset could make the pointer null again. e.g. 0x1.offset(-1) should return None.

I think add is fine, though.

[hkratz]

add is not fine either:

let ptr = usize::MAX as *const u8;
unsafe {
    assert_eq!(ptr.add(1), core::ptr::null());
}

Edit: Nevermind that is UB even for normal pointers. The implementation should probably still assert in `NonNull.add()` that the result is not null.

[Lokathor]

The standard rules of add/sub/offset would still apply, specifically that the calculation does not go outside of an allocation or one byte past the end of an allocation, and also that the calculation does not wrap within the address space.

So it would already be UB to sub or offset to 0, separate from the non-null rule. Similarly it would already be UB to add or offset past the end of the address space, so wrapping around to 0 is also not a concern.

[antoyo]

Seems to be a duplicate of #72429.

[Dylan-DPC]

Closing this as duplicate of #72429