Request for documentation: Is it safe to mutate contents of `Cell<T>` through `as_ptr`?

[FeldrinH]

Location (URL)

https://doc.rust-lang.org/std/cell/struct.Cell.html#method.as_ptr

Summary

Cell<T>::as_ptr allows you to obtain a pointer to T. This pointer can in theory be used to modify the contents of the cell, but is this sound? If I use ptr::write to write to the value, is that sound? Similarly, if I cast the pointer to &mut T and use that to modify the contents, is that sound (assuming I don't use set or get on the original cell while the mutable reference exists)?

I think answers to these questions should be present in the documentation of Cell<T>::as_ptr, otherwise it is unclear what the returned pointer allows.

[workingjubilee]

@rust-lang/opsem While this can be turned over to libs and whatnot as a pure documentation issue if you are okay with whatever they decide, I believe y'all should get right of first refusal and/or recommendation on any particular changes here.

[RalfJung]

Yes such a write is sound (as long as the shared reference that the pointer is derived from is still live). If you create a mutable reference you get into aliasing rules territory -- for the entire duration that the reference is live, no read or write may be performed via any other pointer/reference.

[workingjubilee]

Hm, we could even say "this pointer is correct to use if you treat it like Cell does: read or write, don't hold &mut unless you want to invite Way More Fun Than You May Expect".