The borrow checker is smart enough to track disjointed field borrows individually and detect that's fine, but if you have two methods that return borrows to a single field, there's no way of communicating to the compiler that it's not borrowing the entire struct. This is called "partial borrows", the syntax is not decided, and would likely only work on methods of the type itself and not traits (because trait analysis doesn't need to account for which impl you're looking at, and partial borrows would break that).

The solution today is to either change the logic to keep the disjointed access in one method, provide a method that returns a tuple of sub-borrows, have a method that takes the fields as arguments, or use internal mutability.

Ah, it wasn't clear from they they wrote that this is what they meant.