Post

@jsbarretto reminds me of when I converted a C codebase into Rust waybackwhen c2rust was new. It worked and now I had Rust code that replicated rhe behaviour of the C code perfectly. Which also meant that oh no I had Rust code that replicated the behaviour of the C code perfectly, unsafe memory accesses and all.

The other dimension, of course, is that this is obviously one big marketing ploy by Anthropic, who is currently experiencing major cash flow problems and wants desperately to put their tool on the frontpage of HackerNews.

If you're the sort of person that's talking about this as a watershed moment, I've got to ask: who's paying you? Because "the robot monkey can translate Shakespeare into the worst Spanish you've ever read" is not exactly the world-shattering achievement it's being made out to be. Perhaps in isolation 5 years ago it would be, but there's nothing here that demonstrates very much about anything from a capability standpoint. Automated tools have been producing similarly awful translations between programming languages for decades.

@jsbarretto tools like formatters, parser generators, etc are typically deterministic, though.

@jsbarretto 1-1 porting something to rust, which has a pretty aggressive posture on what the compiler is allowed to do when you violate safety constraints (and where those safety constraints aren't even well defined in areas), is a pretty questionable idea no matter how you perform it. Existing mechanical solutions have made attempts at grappling with this issue, and this simply decides to ignore it.