If I can autoport my C++ to Rust, and the port is confirmed identical, and the Rust is confirmed safe, can't I use that to reason about the safety of my C++? Is safe C++ just a matter of proving it has a safe Rust equivalent?
It depends on what “confirmed identical” means. Most compilers (or translators as you like) only guarantee that the output program represents a subset of the behaviors of the input program so it could be that subset is the “safe as in Rust” subset.
GP is referring to the article's title, as in "safe Rust", as in "Rust with no unsafe keyword", as in whether it has undefined behavior or not.
Granted the premise is rather brittle, since all Cpp2Rust does is to convert UB to runtime crashes, which can be undecidable, and cannot be used to prove anything about the original C++ code.
If the concern is memory safety, I'd invite comparison with migration to the scpptool-enforced memory-safe subset of C++ [1]. If your C++ code is "idiomatic modern" C++, then the changes required to conform to the safe subset (in an idiomatic way) are often modest, and as demonstrated in the link, often something an LLM can handle for you. Since the safe subset does not impose a universal restriction on mutable aliasing the way Rust does, it doesn't require wrapping everything in `RefCell<>`s or anything like that (unless the objects in question are being shared between threads).
If your C++ code is more "legacy" than "modern", the migration is generally still straightforward, but will often involve additional run-time overhead, like it does with this cpp2rust, but to a lesser degree I think. Objects allocated on the stack can (safely) remain allocated on the stack even when they are the target of pointers. And still, no `RefCell<>` equivalents are required for objects that aren't shared between threads.
Interesting, but unpractical. As I know there is no way to perform automatic translation from one programming language to another without producing code which looks terrible in the destination language.
In this particular example I see no real safety benefits. If the source program is buggy, the result translated program willcrash at runtime, but if it's (mostly) bug-free and UB-free, such translation gives no benefits.
> our reference-counted translation model, where every variable is pessimistically wrapped inside a Rc<RefCell<T>> type, checks that would usually execute at compile-time are shifted to run time, degrading performance.
