Feb. 3rd, 2013

tim: Mike Slackernerny thinking "Scientific progress never smelled better" (science)
On Twitter, [twitter.com profile] mcclure111 asked -- and I'm wildly paraphrasing here, since apparently it's impossible to search for a tweet that's more than a week old:

"Research languages seem to focus on only one question. Rust focuses on ownership types. I'm not sure if they're useful or not. They 'deform the code' and I don't know whether it's worth it."

It amuses me to see people looking at Rust and concluding that it focuses on ownership types, because when I started working on Rust, less than two years ago, it didn't have ownership types at all. When I started on the Rust team as an intern, I thought that Rust was centrally about investigating the usefulness of typestate. (In Rust, typestate was a system of assertions that were visible to the typechecker but relied on dynamic checks for soundness. We removed it long ago.)

But later, the team decided to get rid of typestate, and add ownership types. I think the decision to add ownership types happened during the three months in between the end of my internship and the start of my full-time employment at Mozilla, and I didn't pay a lot of attention while I was on leave. But, my understanding is that the decision got made not because anybody said "I really want to do research on ownership types", but because all of the other options had been exhausted. Way, way back in the day, Rust had neither automatic GC nor ownership types. It only had automatic reference counting (that the compiler manages for you), which necessitated a cycle collector that never quite got finished. But ref-counting everything just added too much overheard. Everyone figured that GCing everything would also add too much overheard. So that left what's known in the programing languages literature as "regions", though we've come to call them "lifetimes" instead.

Also, calling Rust a research language is funny to me because -- as its name reflects -- we've tried hard to avoid incorporating new technology into it. We haven't always succeeded at failing to be novel, but we have a rule of thumb of not including any ideas in the language that are new as of the past ten years of programming language research. The field of programming language is full of old technology that hasn't been put to use in solving problems that it's exactly suited for. The goals with Rust were to avoid reinventing wheels, and see what the past had to teach us. I can't blame anyone for thinking Rust is a research language, though, since it is being developed by Mozilla Research.

But anyway, back to ownership types. Sometimes, people new to Rust say that having three different sorts of pointers is complicated, and do we really need to have that? There was a recent thread on the rust-dev mailing list about whether Rust's syntax is "intimidating" for newcomers, and a large part of that discussion was about the syntax for ownership types. Part of the source of the original confusion had to do with explicit type declarations that weren't really necessary, that were only present to work around a bug in the compiler (that got fixed later). It's hard to say much more without an example, so here is one.

Cut for code, and lengthy discussion )


tim: Tim with short hair, smiling, wearing a black jacket over a white T-shirt (Default)
Tim Chevalier

September 2014

 1234 56
78 910 111213
1415 16 17181920

Most Popular Tags

Style Credit

Expand Cut Tags

No cut tags