It’s just another option, don’t gotta use it. Maybe you find yourself needing something like this, and the only other choice is making it public. At least with friend classes, you know which classes are friends so you can go look for any dependencies
There’s infinite ways to organize code. In C# or Rust where this isn’t an option, you might use nested classes or traits hidden behind a module/namespace.
Good use cases are data structures with associated helper classes. For example, a collection/tree and an iterator/tree-walker for working with elements of the collection. Or for something like a smart memory allocator (an arena or slab allocator), you might use a friend-class to wrap elements returned from the allocator, representing their connection back to it (for freeing up when done or to manage the allocation structure in ie a heap or sorted tree).
It allows for more fine grained access control and to implement afterthoughts.
Think having some private function that can break things if called improperly, but also allow you to avoid significant overhead when calling it the correct way. For example you could be avoiding input validation in a public wrapper for that function. If your friendly class already does it, or cannot produce invalid inputs, there is no need for that.
You could also implement logging after the fact, because your friendly logger object to read private members.
Arguably it’s a questionable design decision tho, as you could do all of this in other ways and it basically breaks any guarantees private would usually give you.
Add comment