Hacker News Reader

Comment by mutatio | Hacker News Reader

The generic slice functions would certainly reduce boiler plate, but what I miss in comparison to Rust is the concept of iterators. In Rust if you were to iterate over words in a string you'd get borrowed words, so in practice lots of things are non-allocating (unless you choose to collect / take ownership) - in Go etc. instead of windowing over a string you typically fully allocate the []string result (e.g. strings.Split / strings.Fields). Sub-slicing in Go can be used to make your own iterators, but the ergonomics take a hit.

View on HN

Rust iterators are beautiful. If you've ever implemented C++ iterators, you'll be pleasantly surprised at how elegant it is in Rust by comparison (actually you could replace iterators with about any language feature and the above holds true.)

I really like Rust as a language, it's like a lightsaber, an elegant weapon from a more civilized era. Go feels a bit more like a blaster (sorry May 4th was yesterday.)

That said, when it comes to getting things done, the blaster seems more effective. I would chalk that up to garbage collection, goroutines vs async+await, and simplicity of the language in general. That said I use both languages regularly because they have different strengths and weaknesses and one is usually a clearly better option depending on the problem and requirements.

I really admire Rust iterators, especially compared to C++ iterators, but I lament that there is no easy way (that I know of, anyway) to build an iterator from a closure which yields `Option<T>`s. I don’t like that I have to define a type and implement a Trait (in general working with closures is cumbersome in Rust). That said, this isn’t a major problem either and on balance I’m very happy with Rust iterators.

It's probably possible to create a wrapper to do that. But if you need a closure, you can also just use a struct implementing the trait to manage the state. It's not perfect, but it's so much nicer than in C++.

Might you be looking for https://doc.rust-lang.org/std/iter/fn.from_fn.html ?

Oh wow, yeah, I think that’s it! Thanks very much!

Whoa I did not know about this. This would have saved me so much time during Advent of Code…

Thanks for sharing! I wish I'd known about this earlier.

> I lament that there is no easy way (that I know of, anyway) to build an iterator from a closure which yields `Option<T>`s.

Could you expand / explain what you actually mean here? Because I can think of multiple interpretations of this prompt which are straight up part of the stdlib, but that I can interpret this description in multiple ways shows the issue.

Let’s say I have a closure `fn() -> Option<T>`—how do I turn that into an Iterator? Maybe it’s possible with the stdlib and I’m just not familiar?

EDIT: I think this is it: https://news.ycombinator.com/item?id=27050861

Let's say std::iter::from_fn didn't exist. Then you'd need to make something that ends the iterator when the function returns none, you could do it like this:

    std::iter::repeat_with(f).take_while(|elt| elt.is_some()).filter_map(|elt| elt)

Where `.filter_map()` removes the Option layer and `repeat_with` creates an initially infinite iterator from your function `f`.

Long time ago, `repeat_with` might not have existed. Maybe you'd do something like this instead for that in Rust 1.0.0: `std::iter::repeat(()).map(|_| f())...`

You prefer to use std::iter::from_fn now that it exists, of course!

To be honest, making a simple iterator adaptor is quite easy too, so you could have a knack at making one you are missing - if you can't find it in std or in the crate itertools already.

> the blaster seems more effective

And you have to, er, attack the rebels with an army of highly variable talent and training, which contains maybe one or two actual Jedi Ninjas, so maybe it's better if everyone just uses blasters and nobody cuts off their own arm.

I'd love to work in some super elegant language for my own intellectual betterment, but I really hope my next real-world gig uses something like Golang if there are more than a half-dozen devs... ever.

Agreed. I love Rust for my personal projects, but I wouldn’t trust colleagues with it at any place I’ve worked. And it’s not that my colleagues aren’t intelligent or capable, but that Rust’s expressiveness invites bineshedding on the right amount of abstraction or which crate to use or the right way to do errors or how to configure rustfmt or when to just give up a fight with the borrow checker and use clone() or etc.

Rust is a really, really cool language—perhaps my favorite language, but I’ve worked in C++ shops and Python shops and I fully expect the same amount of bike shedding at a Rust shop.

Apparently I’m pretty much alone in thinking that Rust is a clumbsy, messy language. The syntax is all over the place. You get the feeling that there’s no design strategy and than anyone with half an idea is allowed to bolt on features.

Rust’s syntax is definitely complicated and that’s one of my least favorite things about it but it’s extremely consistent. What gives you the “all over the place” feeling?

I don’t really agree that it consistent, but maybe you’re right that it just complicated. There are so many different syntaxes for all sorts of features that you can’t really hold it all in your head.

It’s been a year since I last gave Rust a try, in the mean time I’ve been able to pick Go without much effort. Rust just seem like PHP, it’s fast, featurefull but not thought through.

You're assuming that it was difficult to learn because the features weren't thought through. Have you considered that maybe it was because there were genuinely new concepts not present in other languages?

Yeah, if anything, Rust feels to me like it's incredibly well thought through. All the pieces fit together well precisely as if they were all thought of as a whole. It's "hard" because it forces you to consider an entirely new constraint of your programs upfront. But (almost) every piece composes nicely with every other piece in a way that makes Rust development extremely pleasant.

Golang feels like the opposite, in that none of the parts are actually built to work with one another. Tuple returns exist but there are no language features that allow you to actually work with them. There's a strong type system, but you have to break out of it with interface{} and type punning to solve entire classes of common problems. Type punning exists, but go can't prove exhaustiveness of switch statements so doing this opens further gaping holes. Errors are handled explicitly everywhere, but again there are no tools to handle the repetitive lifting so every program becomes an endless series of error handling occasionally interrupted by bits of program logic. And let's not forget how badly interface pointers interact with nils: comparing a nil interface pointer against nil returns false, so a basic nil-check that works everywhere else fails to actually perform correctly when dealing with interfaces.

> anyone with half an idea is allowed to bolt on features.

I know you're talking about feel here, not about process, but to be clear, while we do have a pretty open and fairly democratic process, and while anyone can propose a feature, only the language team members, which is currently five folks, can decide to say "yes" to a proposal. The team may have been plus or minus a few over the years, but it's never been particularly large.

(We generally feel this process works well to balance all these factors; we want everyone's good ideas, but we also have to maintain coherence. An open suggestion process with a closed selection process has worked well in our eyes, but everyone can feel differently, of course.)

I’m guessing you’ve looked more at Rust code that uses incubating features from Rust nightly.

Most of those features don’t make it to stable—or if they do, their syntax has usually been refined by the time they do.

Inconveniently, a lot of the Rust code examples you’ll run into on HN uses incubating features, because a lot of the people posting about Rust on HN are language researchers talking about how they were finally able to implement new language feature/performance optimization/safety guarantee X using incubating language feature Y; or how they think nascent library Foo’s use of incubating language feature Z is/isn’t a good idea.

All that is inside baseball. It looks nothing like day-to-day Rust code, any more than e.g. .NET CLR static analysis code looks like day-to-day C#.

(Also, there’s the code examples of Rust stdlib code, which is like C++ STL code in that you have to use arcane language features “in anger” to bootstrap the language — e.g. safe concurrent data structures must necessarily at some point rely on unsafe code.)

(Or it’s code for a Rust unikernel / OS device driver / etc, because that’s a thing you can do. But it’s a still-nascent thing you can do, that’s not yet as ergonomic as regular code that can rely on virtual memory to exist. You have to build the safety yourself in such cases, or rely on a library that does; and there are no such libraries that have yet reached high code quality + wide cross-platform applicability.)

Rust stdlib code is a lot nicer to read than C++ stdlib code, which relies on _Mangled names to prevent conflicts, and (ab)uses template metaprogramming, SFINAE, overloading, & and && and forwarding, etc. heavily. Though Rust's trait system results in indirections comparable to (but somewhat more manageable than) C++ SFINAE/overloading/type traits. And some Rust stdlib types delegate to compiler intrinsics rather than being implemented in unsafe Rust (forgot which).

Additionally, third-party Rust data structure/concurrency crates often contain unsafe code comparable to Rust's stdlib in complexity. (And device drivers, as you mentioned, are probably highly unsafe too.)

Is that because Rust is an expression language? As much as I like Rust, I do find that returning variables from conditionals or match expressions can feel a little jarring (especially when you combine the two). In other places though, it really shines and just requires a little discipline.

> If you've ever implemented C++ iterators, you'll be pleasantly surprised at how elegant it is in Rust by comparison

You probably already know this, but if you're willing to take a dependency on Boost.Iterator, then boost::iterator_facade[1] makes it much easier and less error-prone to implement a standards-conforming iterator, and the pre-built iterator adapters[2] cover lots of common cases, so you don't need to directly implement an iterator at all.

Not arguing the relative merits of C++ vs. Rust, just leaving a pointer here for anyone who didn't know about it.

[1] https://www.boost.org/doc/libs/1_66_0/libs/iterator/doc/iter...

[2] https://www.boost.org/doc/libs/1_66_0/libs/iterator/doc/inde...

Rust is a beautiful language. Its only issue is that programming in it is painful for anyone who hasn't followed Modern C++ for a decade and then read everything on Rust, which is just Modern C++ with some niceness from ML thrown on top.

I couldn't even begin to imagine onboarding a Jr Engineer to a complex Rust project. I don't even know if it is possible.

What constitutes "Modern C++"?

Yes, iterators are typically a hard[1] problem in Go. I think this essay[2] describes perhaps the closest lazy eval iterator pattern in Go, which is much more complex to write than languages like Python.

[1] https://ewencp.org/blog/golang-iterators/index.html [2013]

[2] www.catb.org/~esr/reposurgeon/GoNotes.html [2020]

Clickable: http://www.catb.org/~esr/reposurgeon/GoNotes.html

I don’t understand what’s hard about iterators in Go after reading that article. I’ve written a lot of Python and Go (15 and 10 years respectively) as well as a bit of Rust and while iterators are certainly simpler in Python and Rust since Go lacks generics, most of the “challenges” presented in the article seem petty (complaints about for loops). The easiest way to write an iterator in Go is the closure/generator method and you can use it in a for loop like so:

    for x := next(); x != nil; x = next() { ... }

Not as nice as a for/range but not a real problem. With generics we can have libraries that operate on iterators like map, filter, reduce although I also think people make way too big a deal about a tiny bit of for loop boilerplate (there are legitimate reasons for generics but I don’t think small amounts of boilerplate are among them).

There's nothing hard about writing iterators in go.

What I miss in comparison to Rust is concise error handling. Rust has shown that it is entirely possible to have clear, type safe, concise error handling without try/catch rigoromol or if(err == nil) {...} pollution. My hope is that the introduction of generics in Go eventually leads to Result[] proliferating throughout golang.

Rust iterators are hard to implement without a big compile time hit.

Compile time hit meaning it's difficult to get the types right or something else?

Presumably that it’s inherently slow to compile. “Getting the types right” happens before compile time.

That they rely on a lot of optimisations to be really efficient (= have any chance of approaching a regular for loop).

Without that, every adapter adds a function call to the yielding of an element.

Reply on news.ycombinator.com