> It's just... wow, that's pretty mission critical for what I gather is still a beta-phase tool.
I think this is very '90s thinking. Extreme "if it's not C++, you can't deploy it!" conservatism about programming languages hasn't been true for well over a decade. Paul Graham has written some nice essays about this very phenomenon.
Also, Rust is not "a beta-phase tool".
New languages are much easier to deploy since the PL Renaissance. There are a few holdout industries that still stick with C++ no matter what, such as games, but they'll come around eventually. :)
Back to beta-phase claim, you're saying the core features and compiler of Rust are no longer in development or testing? Totally stable now with only extensions, library features, and optimizations being added? If so, that's good news and I'll drop that claim in the future. Otherwise, it's in the beta phase. Another hint would be existence of any bugs with core functionality in your compiler or libraries.
Again, though, if the spec and core functionality are frozen with all bugs fixed then I won't call it beta. Is it there yet?
"All bugs fixed" rules out every language in existence except CompCert, so I don't think that's a very interesting metric.
I certainly won't expect it to be bug free. Nice comeback with CompCert, though. ;)
> I'd have first deployed existing, native code along with
> Rust in parallel using the data splitting technique so
> some is saved with original and some with Rust.
Yes, this is how Dropbox tests all of their features, and they did this with their new Rust-based storage engine as well. This tool has been in development for over a year by now and has seen extensive multi-stage testing (at first handling only a few terabytes of data, then "just" petabytes, and now exabytes).Thanks for the confirmation, though. I do look forward to reading the experience report. Hoping they include data on all key attributes vs prior language: productivity, debugging time comparison, CPU use, memory use, integration issues, and any changes in defect rates. Community evaluating Rust needs that data for something at this level.
The project was a whole team of engineers working for almost a year, with extensive testing. At least, that's what I've heard. So like, I think they did exactly what you said, it's just that it's ready by now.
"Edit: also worth adding that he Rust core team has been amazingly friendly and helpful. We've had several meetings with them where they came to our office and basically said "how's it going? what do you need? open up your laptop and show us your biggest problem." The project is under very good management."
When Rust was being developed, the two languages that got me most excited as next-gen, systems tech were Julia mainly and Nim somewhat. They had a great choice of features balancing all sorts of attributes as PL's that might get easy adoption as C++ alternatives. Thing is, many blog posts reported their communities and dev teams were downright vicious on everything from questions to compiler bugs (wtf!?). That made it a no-go as there's an upper-bound on how good those projects can get building on such a creaky, hostile foundation.
That leaves Rust as my default recommendation for new, non-critical projects in this space. Like Dropbox employee, what I've noticed is the two of you on Rust teams here are unusually helpful and respectful compared to many in PL communities defending their work. Even pcwalton when he's clearly less than happy with my comments. ;)
The combination of brilliant design and seemingly (from my little exposure) excellent community is why I promote Rust. So, keep up the great work all of you! :)
I think it's a risky move. I'd have first deployed existing, native code along with Rust in parallel using the data splitting technique so some is saved with original and some with Rust. Any compiler or library related issues on Rust side would be caught without data loss. I wonder if they did this in alpha stage and how long that lasted.
I mean, I love seeing Rust in production and some success stories would be great. It's just... wow, that's pretty mission critical for what I gather is still a beta-phase tool. I'd be testing on the middle or front-end layers first given they're usually stateless(ish).