Comment by roblabla | Hacker News Reader

Comment by roblabla | Hacker News Reader

Parent Full thread

> Having a complex programming language which limits you and controls how you use it does not sound appealing to those who need a feature done, fast.

This is so interesting to me. I'm a Rust programmer by trade (as in - I'm not a hobbyist, I actually write Rust for work). We've found that, while the feature work is a bit slower in Rust than in other languages the company used to use (mostly Python), they tend to require a lot less maintenance down the line (less bugs, easier refactoring), and so it ends up canceling out a bit.

I also never find that Rust limits or controls me in my day to day work. There are some things you cannot do easily, sure, but those things tend to not matter in application code. Linked lists are hard, graphs are hard, but when writing an app, I just use an existing dependency like petgraph or std::collections::LinkedList instead of reimplementing those datastructures. And if you need to write those things, it's not like it's impossible, you just need to drop down to unsafe rust, and ideally figure out some safe way to expose that API.

Really, the biggest disadvantage of Rust is that its learning curve makes onboarding newcomers much harder if they don't have experience in the language. For Python/Go/JS/Ruby you don't really have this problem, even if a potential recruit doesn't have experience in the language, they will probably be able to pick it up as they go without too much trouble. But in Rust, it can take a fairly significant amount of time to get up to speed and stop fighting the compiler, in the order of several months.

This feels like a common pattern. Languages like Rust and Go seem to pick up a lot of users from the world of dynamic languages, who then incorrectly associate the productivity gains of static typing with the specific language they picked.

A lot of programs written in Rust would almost certainly be better off being written in Kotlin. The one in the article is a good example. Why are they writing a messenger bot in Rust? That doesn't seem like the sort of use case Rust was targeted at. It's also a modern language with a lightweight syntax and pretty good static typing, great refactoring tools etc, but it's way more user friendly. There's no borrow checker because it's GCd, compile times are much faster (at least for the JVM version) etc.

You can also use Kotlin/Native or GraalVM native-image to produce standalone executables that don't need a full JVM, if that's a requirement for your use cases. The native images are astounding. They can start faster than programs written in C and their memory usage is also way less than a typical JVM app. Downside is of course compilation time but you can avoid that by just developing on the normal JVM and then AOT compiling at the end when it's time to release.

> Why are they writing a messenger bot in Rust?

AFAICT Rust is a general purpose programming language. People can use it how they want. How do you learn a recent programming language? "hey boss I'm going to write production code in a new language that I haven't learned. Lol YOLO"

The funny thing about this is, using rust for production is way safer than an old language like C or even C++. So the YOLO part isn't crazy here. If you don't use unsafe you cannot, read can not, get data races, among a host of other common bugs.

Now if we were talking about some other "new languages" not naming names, you can memory leak kilobytes writing hello world... Yikes.

Well, the dispatcher problem is applicable outside of messenger bots too. But your point is right. I think it was a mistake that teloxide is written in Rust. I just wanted to learn a new language so I was like, why not writing such a library in Rust. Don't make my mistakes.

> We've found that, while the feature work is a bit slower in Rust than in other languages the company used to use (mostly Python), they tend to require a lot less maintenance down the line (less bugs, easier refactoring), and so it ends up canceling out a bit.

Python to Rust is a pretty radical swing from one approach to language design to another. I'd expect you could solve most of the maintenance problems with Python by switching to something statically typed but with garbage collection (e.g., C#, Java, or go), without incurring the costs of moving to Rust.

https://dropbox.tech/application/our-journey-to-type-checkin...

Well, we have other reasons to chose rust rather than a GC'd language (mostly has to do a lot of FFI, which Rust makes much easier than go or Java). I do agree that, if GC'd language fits the problem space, they tend to be much better.

GC has little to do with FFI though. In C# P/Invoke, you basically just declare a static method as external and specify the library where it lives, and that's that. So far as I know, Java has a similar story these days, no need to write JNI wrappers by hand etc. Go is special not because of GC, but because of its green threads.

GC and FFI do have a lot to do w/ each-other though, because FFI usually introduces unmanaged objects.

When a lot of what you're doing is interacting with low-level platform APIs, you end up having a lot of those unmanaged objects. After a certain point, the upsides of using a GC kind of disappear because you still have a lot of places you have to worry about those objects.

Of course, this can be worked around by providing managed wrappers around those unmanaged object, but at a certain point, it becomes easier to just drop down to an unmanaged but safe language (like Rust) that model the unmanaged resources more accurately. In my experience, it's somewhat easier to provide Safe Rust wrappers around your average Windows API than it is to provide a C# Managed Wrapper around the same.

---

And yes, go is special because it has very smol stacks, so doing any kind of FFI on it is a bad idea.

Unless your code is strictly glue between those lower-level APIs, GC is still a benefit for the majority of it. And, on the other hand, the lack of it in FFI is, at worst, similar to C... except you still have all the other language features (like, say, null safety or pattern matching) at your disposal.

I'm genuinely curious as to what would make FFI in Rust easier than in C#, assuming an apples-to-apples comparison (i.e. use of "unsafe" and associated features in both cases"). Most complexity in P/Invoke shows up when you try to use it to automatically map data to managed data types; but in modern C#, you might as well just use raw pointers, stackalloc, spans etc.

Rust seems to excel in having an ecosystem that provides binding helper libraries like PyO3, Neon and napi-rs, Helix, etc. which both ensure "if it compiles, you're doing the C glue properly" and provide a layer for automatic type conversion.

...so requiring `use of "unsafe"` is often akin to forbidding Cargo/pip/NPM/etc. and then calling the language un-productive or faulting a procedural/imperative language for performing badly when you code as if you're writing Haskell.

https://www.hobofan.com/rust-interop/

https://areweextendingyet.github.io/

Using the ecosystem rather than reinventing it is a core tenet of Rust's value proposition.

I’m still learning Rust, but the idea of having to use unsafe features to implement something as simple as a linked list seems “wrong” to me. What am I missing?

I kind of hit a wall with Rust after realizing that something like doubly linked lists are difficult because when two nodes are referring to a node between them, you don't have a clear owner. So basically all situations where you have two or more references to an object need to be thought out carefully, and for me it was a bit of a let down (even though I fully understand the reasoning behind it and why it's useful). For now I stick with languages with GC.

Unpopular opinion but the whole memory safe idea is more niche than HN commenters would have us believe. Most of us are writing web apps and services that do not have strict memory requirements nor catastrophic failure modes.

Dynamic languages and GC'd languages cover most of what our employers are paying us for: web apps, backend services. It is ironic to build super safe software, then deploying them on kubernetes, written in a GC'd language prone to nil dereference errors.

Rust has its place, but it's a very small place. These days computers are so fast there's companies worth billions built on Ruby and Python, if you go native with a GC language you'll cover 98% of your needs.

Rust IMHO has gone too far towards safety at the expense of developer ergonomics and pragmatism. We're so expensive that adopting a subpar language that is easier to grok is often a savvier business decision.

> Most of us are writing web apps

Indeed. I'm a Rust dev somewhere it actually matters (video dev - we have to process 60 frames a second, for hours/days on end, with completely predictable performance, without crashing once) and I'd never pick it if I just wanted to build a webapp where nothing really matters and if the whole thing falls over we can just put it in a retry loop.

There's plenty of code out there that /cannot/ be written in Ruby/Python/Go/C#. Rust solves some really hard problems for people working on that type of code. But it's not what most people on HN are working on.

I think the problem is that a lot of web developers do not really understand that this type of code exists, and that languages which do not target web development can be exciting.

Your video app is a perfect example of where Rust shines. I'd add embedded development, data-intensive, fault-intolerant systems and game engines are perfect for this language.

But without even going as far as frontend web applications, stuff like backend services, system services, GUI applications, CLI applications, etc., don't actually need all that much memory safety and control over allocations.

This feels a bit like complaining that a tank is badly designed because nobody drives them on the highway...

My understanding is that Rust was designed to be a safe(er) language to write low-level system code than C/C++, not to compete with Python/Ruby/Java for web development applications.

True, yet what you see around HN and other places are people trying to use it for web development.

Front end or back end? I see nothing wrong with rust on the back end, not sure how it would even work on the front

WebAssembly.

Bingo, it's a true systems language. That said, the fact that people can and do write web applications(for practical reasons) with it kind of says something about it's breadth.

The whole point of system languages is that they reach the whole stack.

I'm not complaining about Rust. It's a very good language.

I'm complaining about everybody on programming forums, including HN, seeing Rust as the replacement for most, if not all languages. My opinion is that Rust's ideal problem space is much narrower than people think.

You might be right for the majority, but there are many other large groups who don't have computers that are more than fast enough. There are a lot of interesting and complex problems that can be solved in something like rust that the languages you name cannot be used for. Right now I'm using C++, but rust is of real interest to me (as is ADA, which in some variants lets me prove my code correct)

Completely agree! When you need to compute numbers or deal with raw bytes, you have C. When you need to be high-level, you have Python, Java, C#, etc. Rust is applicable only in areas where you need both a high-level and systems language. Examples include interpreters, operating systems, browsers, game engines, etc. The hype of Rust just exceeds its problem domain.

There seems to be a myth that Rust is unusual in being memory safe. All modern GCed languages are memory safe (assuming you don't do anything obviously unsafe like manipulate raw pointers, which some languages might let you do if you really want to).

>It is ironic to build super safe software, then deploying them on kubernetes, written in a GC'd language prone to nil dereference errors.

Nil dereference errors don't demonstrate an absence of memory safety. If the runtime checks for a nil reference before dereferencing and then raises an exception, that's perfectly safe.

I'm not sure if you understand what safe means in this context. One common example is modifying a vector element by reference. If you grab a reference to an item, push something else to the vector, then try to access through that reference, the vector may have re allocated during the push and you're either no longer looking at valid memory because it's been freed, or you're looking at a stale copy of the element.

I don't think the null handling in rust is about memory safety so much as it is their choice in error handling. If you come across a null pointer dereference and your code doesn't handle it, that's a hidden bug. Triggering an exception and handling it is a valid approach, but there are trade-offs. If no code handles the nil exception for example then it's no better than crashing as you would in C. Wrapping things up in try catch can lead to missing recoverable errors or unexpected states. The choice to disallow null is one to avoid common bugs, not one for memory safety, as far as I understand it.

I'm 100% sure that I understand what safe means in this context. All modern GCed languages are memory safe and will not access invalid memory regions unless you use specifically unsafe features.

> One common example is modifying a vector element by reference. If you grab a reference to an item, push something else to the vector, then try to access through that reference, the vector may have re allocated during the push and you're either no longer looking at valid memory because it's been freed, or you're looking at a stale copy of the element.

There aren't that many languages which both use GC and allow you to take references to elements of a vector. Go would be one example. I can assure you that you cannot access undefined memory regions by doing this in Go. (What would happen is that the original allocation backing the vector would be retained in addition to the new allocation.)

>If no code handles the nil exception for example then it's no better than crashing as you would in C

C programs are not guaranteed to crash when a pointer has an invalid value. Dereferencing a NULL pointer will reliably cause a crash if you're running the code on top of a modern general purpose operating system with memory protection, but invalid pointers (which don't necessarily have to be NULL) have the potential to access arbitrary memory regions without raising any kind of error.

Turns out I'm the one who forgot what memory safety is o.o I think the part that makes rust's memory safety special is that it also lacks garbage collection

> I can assure you that you cannot access undefined memory regions by doing this in Go.

Current implementation of slices and interfaces in Go is not memory safe in presence of data races:

https://blog.stalkr.net/2022/01/universal-go-exploit-using-d...

This is the kind of edge case you can find in lots of languages, including Rust (which has plenty of past and present soundness bugs).

Data races are violations of the memory model; of course any code which produces data races cannot be understood as memory safe?

As far as I remember, some languages like Java are memory safe even if the code produces data races. That means you can't have a reference to a value supposed to be of type A being actually of type B.

That's part of what the Rust borrow checker does for you: tracking who owns what and enforcing single ownership makes data races that much harder to cause.

Some people misunderstand unsafe for being anything more than plain memory corruption.

Nope, unsafe isn't a way to do dependent typing in Rust.

The reference to the slot in a vector has to be a kind of object which tracks the vector itself, and an offset into the vector. It can't just be a wrapper for a naked address.

"It is ironic to build super safe software, then deploying them on kubernetes, written in a GC'd language prone to nil dereference errors."

This is something that escapes a lot of people nowadays.

Everyone uses a browser and they are inherently unsafe.

I feel it is a bit ironic that Apple and Google rather introduce mitigations for C++ in their browser engines than introduce Rust into the codebase.

Meaning that Firefox's gains with Rust will eventually be lost given its market adoption is slowing reaching zero.

Google is investigating Rust as part of a multi-pronged strategy and, while it has the "this is not an official Google project" boilerplate, the `autocxx` crate (a header-parser/code-generator addon for the `cxx` interop crate) is by Google employees.

https://docs.google.com/document/d/e/2PACX-1vRZr-HJcYmf2Y76D...

https://www.chromium.org/Home/chromium-security/memory-safet...

https://github.com/google/autocxx

So here's a weird thought, why should you ever write your own doubly linked list? Why shouldn't you use one that a community of people has vetted and juiced for performance instead? In C what I just said is for some reason heresy.

But yes you are right, anytime you have two or more accesses to a data structure you do have to think a little harder about it. Usually you just want to use a reference to that object, sometimes it makes sense to clone it, sometimes you want to pass it by value and return it by value. In the case of doubly linked lists you probably want to drop into unsafe code.

I get it though, GC languages are more zen in that regard.

Writing your own data structures allows you to write it to suit the specific needs/constraints of your problem/domain. Typically that means you get something that is significantly simpler, faster and with a different feature set.

Libraries are typically optimized for generic use, so are often more complex, slow, and difficult to extend with new features (both coding wise or the features might not be suitable / have trade offs for other people).

Won't let me reply to fizzynut, but this is a reply to you.

Yes making data structures is important, and the reality of that is, there are people in the ecosystem creating fancy ones which can be leveraged for your use case.

If that doesn't exist then yea you have to write them. Sometimes that's difficult(you need to use unsafe), sometimes it's easyish and you can use all safe(I did this for a graph type). It can be done(lots of people are doing it), but I think what it shows is, data structures can be difficult to do correctly.

In a worst case scenario you could write it in asm inside rust if you really need a low level control, or write it in C/unsafe rust and ffi or just write it in.

Basically worst case scenario in rust is writing something closer to C.

Lots of people writing their own data structures and libraries should be a reflection of the different needs and diversity of the ecosystem, but if it's a reflection of difficulty then it tends to lead to a much narrower ecosystem.

That isn't to say a narrow ecosystem is necessarily bad, it can often be good if a particular language is used for a specific domain as all the libraries and data structures will likely be a better fit for your problems. Going outside of those constraints can often be very punishing in those languages though.

The fact is that writing a doubly linked list correctly is hard and it's easy even for veterans to miss things. Rust is going to put those invariants in your face.

My advice would be not to judge any language on how easy it is to implement a doubly linked list. 99% of code you write will likely not be that, and that particular data structure exposes a lot of choices and trade offs at a time when youd be better served learning other parts of the language.

the thing about Rust's ownership model is that a reference is not ownership, it's 'borrowing', and should be treated as such. But - and here's the kicker - the borrow happens as long as that reference is stored, not just when it's accessed i.e. dereferenced. This makes a number of data structures and patterns difficult and impossible to implement as you can't keep track of references to anything. You can't really make an opaque handle, for example, that can do operations through a stored reference. You have to store some ID or something and then when you want to do operations you need to pass the original container to the function anyway - so what's the point of even having reference types at that point?

IMHO Rust's lifetime analysis should consider borrows only at the point of dereference, not just whenever you encounter an & symbol. They already do that with raw pointers - you can have as many pointers as you want but you only need to mark `unsafe` if you actually dereference one. So at least someone on the team knows how to make that work functionally.

You’re missing that you don’t have to use unsafe, you use a wrapper type which is essentially a smart reference-counted pointer. Rust makes these things explicit.

you're not missing anything.

Because it's a library feature, exactly the thing you should use "unsafe" in, since it should be simple enough to be obviously correct

> And if you need to write those things, it's not like it's impossible, you just need to drop down to unsafe rust, and ideally figure out some safe way to expose that API.

I'm sure you know this already, but for the people reading this: first of all "unsafe" is a very rare, and it's only "less safe" more than fully "unsafe"; ie it's still safer than C++.

And the second thing is that, often you don't have to drop into unsafe Rust, it's possible to achieve most stuff without it, but might incur a small performance cost (idk if some of those can be optimised away by compilers or not).

Reply on news.ycombinator.com