[comex]

But how would you prevent use-after-free? In other words, how do you prevent a child object's method from reaching out to its parent object and, say, clearing the container that contains the child object? Specifically, how do you do that without reference counting or garbage collection.

[thomasmg]

Well the distinction between mutable and immutable borrows doesn't solve use-after-free. What prevents use-after-free is ownership. There is one owner, and the owner can not free the memory as long as there is a borrower (it doesn't matter whether the borrower is mutable or not). So that's the task of the borrow checker.

For my own programming language [1], I have only mutable borrowers (similar to Java), and a much simpler borrow checker: a method that holds a borrow can not call a method that potentially frees an object of the same type (directly or indirectly). I'm not sure yet if this is sufficient; it's still a borrow checker, just a very simple one. (My language uses reference counting by default, but does support single ownership + borrowing.)

[1] https://github.com/thomasmueller/bau-lang

[john-h-k]

> There is one owner, and the owner can not free the memory as long as there is a borrower (it doesn't matter whether the borrower is mutable or not). So that's the task of the borrow checker.

Yes, but what about:

``` let mut x = &mut some_vec; some_vec.push(10); ```

Sure the vec has an owner, but the second mutable borrow causes it to invalidate the first pointer. You need a _third_ category, “unstable mut”, which is exclusive as it can cause an object’s internal data to move. You can then collapse mut and immutable to one… but you end up with the exact some colouring problem between unstable mut and the others

[thomasmg]

Yes, I see your point. In Rust, the owner, and the mutable borrower can change the pointer itself (like C realloc). If multiple "mut" borrows are allowed, then this would be unsafe, so I understand "unstable mut" would solve this problem - but result in a new colouring problem.

My solution to this would be: in a new language, do not allow reallocation. Not for owners, and not for mut borrow. This is what Java does: an ArrayList is a wrapper around an array, and so adding entries will not move the ArrayList object, just the (wrapped) private array. In Java the programmer never sees dangling references. Java prevents the issue by design, at the cost of always paying for the wrapper. If you want to avoid this price, you need to use the array directly.

(I have to admit I was not aware that in Rust, push moves the memory... so thanks a lot for explaining! Always learning something new.)

I don’t have any problem with Rust’s focus on performance. But its design does make life harder for developers than it needs to be. It doesn't match my vision of a programming language that is at the same time easy to use, safe, and nearly as fast as C.

[Dagonfly]

> ArrayList is a wrapper around an array, and so adding entries will not move the ArrayList object, just the (wrapped) private array.

That's also how Vec works in Rust. Vec is just (buf_ptr, capacity, len) where capacity is the allocated size of the buffer.

The problem still exists though.

``` let mut v = vec![1, 2 ,3]; let x: &i32 = &v[0]; v.push(4); println!("First element {x}"); ```

The `push` might realloc the array. Then, x points into invalid memory. This is caused by the projection: You can create a reference to a field (aka member) from a reference to the base object.

A language without realloc sounds painful. Any growing container would lead to stale data.

[thomasmg]

In Rust, a vector is a "fat pointer" (with pointer to the array and length / capacity) which lives on the stack. In Java, the pointer to the array and length / capacity lives in the heap. So in Java, there is an indirection, and it is allowed to have multiple pointers to the ArrayList object.

> A language without realloc sounds painful. Any growing container would lead to stale data.

I think it's not so much about realloc, but about whether it's a fat pointer or not. (I could imagine that Java uses something like realloc for the array, if there is only one pointer to the array for sure).

Fat pointers have some advantages and some disadvantages. Rust chose fat pointers, I assume for performance reasons. That's fine. Java doesn't. But I don't think that's a _huge_ performance disadvantage for Java. What I'm arguing is not so much that one is better and the other is worse, just that there are advantages and disadvantages. Rust might be slightly faster, but a language without (this kind of) fat pointers could potentially be easier to use.

[Dagonfly]

That's true.

Though, for my example, the storage location of the array metadata/fat pointer is not relevant.

In Rust you can hold references directly into the buffer backing the array (`&v[0]`).

My Java knowledge is quite rusty at this point. AFAIK, a `ArrayList<MyType>` in Java stores pointers to MyType in the buffer. So, when indexing into a ArrayList, you never hold references into the actual buffer. Instead you get a shared-pointer to the class data. It's the indirection of the array entry that saves you during reallocation of the buffer.

Also because Java is a GC'ed VM, it wont dealloc the elements references by the array, as long as there are other references to an element.

The equivalent in Rust is `Vec<Rc<MyType>>` where holding an `&Rc<MyType>` referencing into the vec's buffer is problematic during reallocation. But, cloning the Rc and holding on to it is perfectly fine.

The initial point of this thread was that you can have a Rust-like language where you can hold multiple mutating (aliasing) references and prevent use-after-free. This won't work. Without a GC or RC, you can use one reference to "rug-pull" the memory that is aliases by the other reference.

[gf000]

I'm looking forward to your experience with this approach!

Bit off topic, but is there any particular reason you went with a go-like `ident type` syntax over the more common ones like the C or the ML one?

[thomasmg]

> any particular reason you went with a go-like `ident type` syntax

I just think that ":" is unnecessary. In my language, the type is mostly used in function declarations and types, eg.

    fun File read(data i8[], pos int, len int) int

    type List(T)
        array T[]
        size int

What would be the advantage of adding ":"?