[tptacek]

Perhaps compared to Python, Ruby, and vanilla ES6; there is a class of epsilon-from-typo bugs that modern statically typed languages catch and that force rigorous Python and Ruby shops to waste time writing vast batteries of tests for the most banal functionality in their codebase ("does this setter set?").

Moving from one of those languages to a modern statically typed language can come with a feeling that your programs tend to run successfully as soon as they compile --- certainly they do more often than they would in Ruby. Because the compiler is doing some of that testing work for you. It's a good feeling.

But it hasn't been my experience that the extra rigor in Rust's model produces the same boost from Java or Go; in fact, my first-run experience with Rust is generally less reliable than Go; there are a few more things that can go wrong with a Rust program and slip by the compiler.

Which is what you'd expect; Rust isn't garbage collected, so the programmer has to do the garbage collection, and every program is responsible in some sense for designing and implementing that part of its own runtime. There are things you can get wrong there that GC'd languages don't make you bother trying to get right. And, of course, if you take the time to get your per-program runtime really right, there's a performance gain to be had from doing so. It's all tradeoffs.

[burntsushi]

> Which is what you'd expect; Rust isn't garbage collected, so the programmer has to do the garbage collection, and every program is responsible in some sense for designing and implementing that part of its own runtime.

This seems like a really odd thing to say in at least two dimensions. Firstly, I would say that "the programmer has to do the garbage collection" isn't an accurate way to describe Rust. Secondly, and particularly in comparison with Go, Rust's RAII applies more universally than Go's garbage collection. How many times have you seen a missing defer or a defer inside of a loop? I've seen both quite a bit. But those bugs basically don't happen in Rust precisely because the programmer almost never needs to deal with destruction directly. It's done automatically by the compiler.

As to the broader point, I would very strongly oppose the notion that Rust is less reliable than Go. While I don't think I would argue that Rust gives the same boost over Go that, say, Java or Go give over Ruby (that's a tough argument to make anyway, even if I agreed with it), I would say there is a meaningful boost. But that's a tough argument to make too. Aside from the obvious bits (nil errors and forgetting to check errors), for me, it basically comes to my belief that Go punishes its practitioners for abstraction. I wrote a bit more about that here: https://users.rust-lang.org/t/what-made-you-choose-rust-over...

[tptacek]

The RAII vs. defer thing is a point well taken. Also, the kind of thing that doesn't show up in the totally-informal "how many of my post-hoc tests pass the first time I get them to compile" metric.

Another point, which to your credit you didn't make but which is nonetheless true, is that I am not a good Rust programmer!

I don't think Rust is less reliable than Go in the large. I do think there are more things you have to get right in a Rust program than in a Go program, so it's "less reliable" in that weird twilight state when you're first bringing up your program.

I do agree that Go punishes practitioners for abstraction. If you're, like, you, that's a very bad thing. If you're working with a team of people for whom the project is a means to an end and not a brilliant-cut gemstone, putting the brakes on abstraction can be a good thing, which I think a lot of Rust programmers will quickly learn after the nth time they've had to do an edit-compile cycle just to `let () = something` to figure out a type.

[burntsushi]

FWIW, my experience when it comes to Go and its abstraction limitations _is_ in the context of collaborating with others on a team at work. That's where we feel the limitations of abstraction pretty acutely. Because it means we, to a lesser degree, cannot build APIs that are harder to misuse.

The point about having too much power is taken, and I don't have experience with that in a team context. And yeah, I am not exploring the Rust downsides as much here, and there are definitely others.

[estebank]

> I would say that "the programmer has to do the garbage collection" isn't an accurate way to describe Rust.

Although you could characterize it as "Rust makes the programmer think about and describe the garbage collection."