Rust for Professionals(overexact.com) |
Rust for Professionals(overexact.com) |
If you’re coming from C++ especially, where the move/copy situation is ridiculously confusing (IMO), but also from a simpler “reference by default” language like Java, this has profound impact on what’s intuitive in the language.
For the C++ comparison, this is a pretty good article: https://radekvit.medium.com/move-semantics-in-c-and-rust-the...
This makes completely logical why some things must be Box<>'ed and borrowed. Why you cannot treat Trait or even impl Trait like any other type. Why sometimes it's ok to have impl Trait as your return type while in other cases it's impossible and you must Box<> it.
Third important realization is that things borrowed out of containers might be moved 'behind the scenes' by the container while you hold the borrow, so you are not allowed to mutate container while you are holding borrows to any of its contents. So it's ok, to instead hold indexes or copies or clones of keys if you need.
Another observation is that any struct that contains a borrow is a borrow itself. Despite using completely different syntax for how it's declared, created, it behaves exactly like a borrow and is just as restrictive.
Last thing are lifetimes, which don't have consistent (let alone intuitive) syntax of what should outlive what so are kinda hard to wrap your head around so you should probably start with deep understanding what should outlive what in your code and then look up how to express it in Rust syntax.
Rust is syntactically very similar to other languages, but semantically is fundamentally different beast. And while familar syntax is very good for adoption I'd also prefer tutorials that while showing the syntax explain why it means something completely different than what you think.
I found it frustrating enough that I put the language down and just went back to using C++.
In Rust it's also obvious because every = is a move. Confusion comes from tutorials pretending for too long that it's not.
> whereas in Rust, similar behavior seemed to depend on the object type.
It's best to think of that as an exception to the rule created specifically for numbers and other similar small, cheaply copied things.
If you need to move `a` but `a` has a trait Copy then you copy instead.
> Rust has its own move operator as well, despite destructive move being the default behavior for assignment.
I don't think that's true? Rust has a `move` keyword but it's a part of closure definition that makes it take all the variables from its environment by move, even if it doesn't need it. Unless you are talking about something else...
At a high level, I think the most unintuitive part of moving in C++ compared to Rust is that it can silently degrade into a copy without anything indicating it. In Rust, trying to use a value after it's been moved will give you a compiler error, at which point you can reconsider whether you do in fact want to explicitly copy or if you made a mistake or need to refactor. In C++, the only way I'm aware of to verify whether a value is actually moved or not is to use a debugger. The benefit for requiring explicit copies is similar to having bindings be immutable by default and requiring an explicit `mut` annotation; if you start out enforcing the constraint that things should be moved or immutable, fixing it later if you find out it won't work that way only requires adding `.clone()` or `mut` in one place. On the other hand, if you start out with implicit copies or mutability by default and then want to change it later, it can be a lot more work to refactor the places where the variable is used to not violate this, and it may not even be possible in some cases.
Any blog about learning Rust for beginners should just contain information that helps the reader decide _whether_ she should put in the time required for learning it, then refer to the great Rust Programming Language book that's really hard to surpass.
The reference is great as well, though personally I miss a part that formally defines the type system in its current form (there are some papers about lifetimes, but they are very hard to read).
(They seemingly don't use more apt comparisons with OCaml and Haskell, for instance, not expecting the reader to know them.)
So yeah there's quite a lot that people wouldn't know.
> Closures (lambdas). Rust supports closures (also called Lambdas, arrow functions or anonymous functions in other languages).
That's misguiding.
Closures are not lambdas. Lambdas are just syntax, but the whole point about closures is that they capture the enclosing environment (have access to variables where it's defined). Rust's documentation states just that. Closures may or may not be lambdas.
In above example of "Inner functions" (which is also a closure) that would be more clearly explained if the inner function used an outside variable. Not all languages can do that.
I really hope to start using Rust in 2023, probably for some kind of API gateway experimentation
So true. It’s never too late, you’re never too old, there’s never something else you need to learn first, just do it.
That you can write very few interesting programs without venturing into the heap with Box, Rc and such and into internal mutability with Cell and RefCell.
Then it quickly raises to the power of other languages and surpasses them with "pay for only what you use" mentality.
You can ignore that in other languages, but this comes at a cost later.
Joke btw. That thread is a hilarious trainwreck - surely the final nail in the coffin for the Rust advocates who so often deny anything about Rust is difficult to learn.
I don't mean that as an anti-Rust jibe, in fact I'm planning to get back to it this year (having given up in despair last). I like much about it, and think it's tremendously practical for many purposes. But it just is a difficult language, no question.
Easy is relative. I suspect a major reason I found it easy was because I didn't try to solve lifetime problems, I just cloned things. I also had primarily been using C++ in school so I was pretty familiar with pointers and, to some extent, ownership. Plus my initial foray into CS was driven by a desire to do exploit development professionally, so lower level details weren't scary at all.
1. "Operators" have different meanings. `1 | 2` and `1..=2` mean something different in patterns than in expressions. Here is a silly example: https://rust.godbolt.org/z/76Ynrs71G
2. Ambiguity around when bindings are introduced. Notice how changing a `const` to a `let` breaks the function: https://rust.godbolt.org/z/aKchMjTYW
3. Can't use constant expressions, only literals. Here's an example of something I expect should work, but does not: https://rust.godbolt.org/z/7GKE73djP
I wish the pattern syntax did not overlap with expression syntax.
This actually bit me in the a__ when I misspelled my enum variant and instead match created a variable named like that, that captured everything and I got only a warning and very wrong code.
I think there should be `let`s inside match block if matching creates some variables.
fn func<'a>() means that 'a must outlive execution time of func
but
T:'a means that references contained in T must live longer than 'a
and
'a:'b means 'a must live longer than 'b (that's consistent at least)
Maybe:
fn 'a:func() {
fn func() 'a:{
Maybe some directional character would be better than : (> is probably out of question because of <> generics)
----
I feel like structs that don't have 'static lifetimes because they contain some borrows should have it indicated in their name.
For example:
struct Handle&<'a> { n:&'a Node }
struct Handle&'a { n:&'a Node }
struct Handle& 'a:{ n:&'a Node }
Then you could use it:
let h = Handle& { n: &some_node };
Like
vec![].into_iter().map()
vec![].iter&().map()
Not sure if it would also be nice to differentiate structs with &mut
----
I would just like to separate lifetimes syntax from generics syntax because those two things have nothing to do with each other from the point of view of the user of the language.
----
I would also like to have
while cond {} else {}
where else is only entered if cond was false from the start. But that's a wish not specific to Rust.
Not as simple as if there was a proper borrow checker but its sort of possible:
https://awesomekling.github.io/Catching-use-after-move-bugs-...
I always get annoyed when I hear about people suggesting to push for a language in a company because they feel it would be a good fit. Sure, maybe it would be, but now the company will always require experts not in one language but two, and developers are no longer as fungible between teams.
Having said that, from a selfish perspective, I will use it to write some production Rust code in the future.
Here is an example that now frustrates me about JavaScript. I don't know if some variables "MyVal" declared somewhere is global, local or how transfer it around. Global variables no longer makes sense to me. In Rust, global variables are possible but they are obvious and their scope is completely within your control. Now, all the data is there and it's up to you how to structure it, move it around, store it, etc.. You have to think these through before doing any code. If you don't think well enough, the Rust compiler (borrow/ownership thing) will start to give you headaches.
Also check out CLiki (the Common Lisp wiki, https://www.cliki.net/), it's very helpful in finding useful libraries, like Alexandria, defstar, trivia, lparallel, and so on.
In the case of programming languages, yes, it's relative to the difficulty of other PLs. I've learned many over the years, and found Rust by far the hardest (it's the only one that defeated me). And it's not the most different from others I've learned - lisps are far further from the common languages than Rust is.
> I suspect a major reason I found it easy was because I didn't try to solve lifetime problems,
Well yes anything's easy if you skip the hard bits. Learn C without using pointers.
I personally didn't find ownership & borrows the hardest part - in my case it's the great complexity of many of the commonly used libraries. Rust's complexity bleeds out into the entire ecosystem (a cultural thing).
I've learned many many PLs at this point. Rust was one of the easiest for me.
> Well yes anything's easy if you skip the hard bits. Learn C without using pointers.
That's my entire point. Rust is not hard if you learn the easy parts first. Trying to learn everything at once is not easy, no matter the language. Once the problem becomes just learning the borrow checker it's not that big of a deal. The harder thing to do is learning the borrow checker and traits and closures and blah blah blah.
> In Rust it's also obvious because every = is a move.
If you assume = always moves you can just never use automatic reborrowing (do &mut * instead) and you'll loose nothing.
I don't think that's a very common pattern, fully optional and it can be just treated as another exception to the rule, just like Copy types. You don't have to manually write .copy() in some cases, and you don't have to manually write &mut* in some others. But that's what happens. Move is still done.
So that's total of two exceptions. Way easier to be tackled individually when the time comes than assuming = means something else, or something complex from the start.
> pub fn main() {
> let mut x = String::from("moo");
> let y = &mut x;
> let z = y;
> println!("{:?}", y)
> }
>
> pub fn main() {
> let mut x = String::from("moo");
> let y = &mut x;
> let z: &mut _ = y;
> println!("{:?}", y)
> }This guide is not going to magically teach you Rust is there's nothing to compare it to; there are other guides to help those without a background. But it could somehow help people who already have similar concepts in their mind to link them to corresponding concepts in Rust.
Lifetimes aren't hard or complicated, what's complicated is understand their interaction with other features (closures) when you don't even understand wtf closures in rust are, or traits, etc. I'd just focus on the other stuff for as long as you can.
Untrue - read https://users.rust-lang.org/t/what-is-a-good-mental-model-of.... Even Rust aficionados can't describe how they work without vast screeds. Or if you think the posters there are just wrong, try the RFC on which the current implementation is based: https://rust-lang.github.io/rfcs/2094-nll.html. You truly don't find that hard or complicated?
It may be the case that people can use simplified subsets of the language to ease their way. But when a Rust advocate says that in effect to understand lifetimes you have to read the standard library source (as is written in the forum thread linked above), because no current documentation is comprehensive enough to cover it, then you know the whole thing itself is pretty complex.
This is like saying: "You think a 64bit integer is simple? OK, let's dig into the C memory model, twos complement, overflow CPU flags, how CPU caches are implemented, architectural nuances that can lead to unsynced writes from registers to RAM, etc". In reality most people can just learn that it's a number that holds 2^64 values, not complicated. If you want to really understand integers though go read the intel manual and the C memory model specification.
Are lifetimes complicated? If you're a compiler developer who needs to consider the implications across all features and edge cases, yes. If you're learning the language and you just want to write "find me a &str in another &str but don't clone it", no, it's not complicated at all.
The vast majority of beginners can just learn "& means any number of readers and no writers, &mut means one writer and no readers".
It's completely optional fearure. You don't have to rely on it. Like auto-insertion of semicolons in JS maybe it's better if you don't. Although likelihood of this biting you is way lower than in case of relying on semicolons autoinsertion.
My claim is that for most Rust users, understanding the borrow checker is limited to a few heuristics that seem to work for them 80% of the time, and throwing up their hands and declaring an approach unworkable the remaining 20% of the time. To a first approximation, no one can mentally model the borrow checker beyond extremely simple examples of 4-5 lines of code.
But 'beginning' is never the hard part of learning any programming language, at least for a programmer. The hard part is going from having learned the basics to getting stuff done. Rust is harder than any other mainstream language to do that in. The writers in that thread can't even describe their mental model (NOT the underlying tech as you claim) of lifetimes without vast elliptical descriptions. And they can't point to any straightforward documentation.
I have learned multiple programming languages, of many different paradigms, and never had the trouble getting to the stage of writing useful software that I have had with Rust. I've witnessed the same again and again with all the people I know - all professional programmers. In fact I'm the only one I know who's stayed with it.
I find the denial of Rust's difficulty (and not even centred on the borrow checker - it's the use of just about every commmon library) just very very strange. Odd enough (and distant enough from the obvious, adn the experience of every person I've known) that I find it completely incomprehensible.
I’d argue C and C++ are both harder. It’s hard to even get these to build once you move past trivia examples (including libraries etc).
> I find the denial of Rust's difficulty (and not even centred on the borrow checker - it's the use of just about every commmon library) just very very strange
Some of us just didn’t find it that hard. Like, I get that its hard for people in theory, because I’ve seen enough people express this that I believe it must be a common experience. But my personal experience was a couple of weeks of slow progress and lots of puzzling, and then everything became much easier, coupled with a joy that most of the language is so much better designed than the previous generation of languages I was used to using (e.g. it actually has sum types, and everything is an expression - why on earth doesn’t every language work like this)
C++, on the other hand, is exceedingly complex and a difficult language to become proficient in, though modern C++ is an improvement in that regard. Some of this is due to decades of legacy cruft, some is due to it being an atypically expressive systems language, all of which has been stirred together.
> I’d argue C and C++ are both harder. It’s hard to even get these to build once you move past trivia examples (including libraries etc).
Agree I overstated that. What I mean is that, out of the difficult to learn languages, Rust is the only one whose community for opaque reasons denies the difficulty.
> Some of us just didn’t find it that hard.
Why the difference I don't know. So far everyone I know who has learned it has dropped out because they just couldn't get practical traction with it. My guess is that most people who say they don't find it hard are either longstanding C/C++ programmers, and/or are learning at work where they have immediate help & scaffolding.
What is odd is that rather than responding to difficulties with a curious "Oh, I didn't find it hard, I wonder where the difference lay", Rust advocates tend to come back with an aggressive 'proof' that Rust is factually not hard to learn (with implications we can all guess at). No other programming community frequently does this in my experience (and I have brushed against many). That community attitude itself must have causes that are worth thinking about.
You own a book.
`&` - You can lend others the book, they can't fuck with it. `&mut` - You can lend the book to one person, they can fuck with it `move` - You give someone else the book, it's theirs now
Or `many reader NAND one writer`
Is this a complete explanation? No. But it's quite simple and you can be plenty productive with just this amount of understanding.
Your experience is not my experience. I used Rust for the first project I wrote as an intern after dropping out, having never used it before, and I even used it to interact with mysql, which I had never used before.
At my last company I had multiple people pick up rust in a matter of days.
I'm not denying rust as being difficult, I'm saying it's easy for some and hard for some. I found it easy, I was writing productive code on day 0 with virtually no preparation other than that I wanted to try it out. Many people find it easy. Obviously some people, many people, find it hard. Life's weird like that.
- function arguments are moved into the called function
- you can call a function with a ref, and then you can keep using it in the calling function, because there is a blanket impl of Copy for refs
- you can call a function with a ref mut, and then you can keep using it in the calling function because ... ???* An object has a lifetime. This generally lasts until overwritten or destroyed at end of scope.
* You can loan out references to this object. You can either have a single &mut, xor an unlimited number of & references, but not both.
* References cannot outlast the lifetime of the object they point to.
* If you have a &mut reference to an object, you can give out another &mut reference to the same object. But you can't use the first reference for the duration of the second.
* Lifetimes can be named, and you can express criteria of the form "this lifetime must at least/most this long."
* Bonus: if you have &mut x, you can give out &mut x.a and &mut x.b at the same time. But you can't give out &mut x while such a subreference is live.
And... that's really all you need. Yeah, there's a lot more rules, and there's definitely fun edge cases around things like temporary objects' lifetimes. But there's no need to know all of that stuff. If you get things wrong, the compiler will come back and slap you in the face and give you an error message--that's the selling point of Rust, getting lifetimes wrong is an error, not silent nonsense--and your task at that point is to figure out if you're really breaking a cardinal rule (two or more mutable references to the same memory location, or references not lasting long enough), or if you didn't enforce sufficiently strict requirements for the bounds of the lifetimes. And the rust compiler is pretty good at telling you what you have to do to get necessary lifetime bounds (sometimes too good--it can suggest fixes to lifetime bounds even when such bounds are unachievable).
I'd compare this to things like name lookup and overload resolution in C++, which are actually horrendously confusing algorithms that almost nobody understands in their entirety. Yet plenty of people can be productive in C++ because the general principles are well-understood, and if you're at the point where you need to descend into the morass of exceptions and exceptions-to-exceptions, you're probably writing confusing code to begin with.
I personally never really had a problem with ownership model - probably because what I was doing fell into the simple buckets. Rust's difficulty goes well beyond ownership. I just thought the Rust user forum thread I quoted was a gently funny example of the Rust community's denial: "Rust's not hard, but to get a decent mental model of the borrow checker you need to read the standard library, or if not here's 10000 words on how I think of it". I'm not presenting it as "proof" that Rust is hard - the fact that I couldn't do anything practical with it after more troublesome attempts than with any other language is plenty enough for me to know that. Neither do I think difficulty is 'bad'. Denying it can be though.
It's not hard, you just add boilerplate. Cloning data, using interior mutability, or adding ref counting to deal with cases where multiple "owners" can keep an object around independently.
Then removing the boilerplate is how you do optimization, once you've gotten things to work. The opposite of other languages where low-level code is the most verbose and least intuitive.
Yep, that's the standard Rust aficionado flat insistence: "You're finding it hard. You're just wrong". Yet with more learning time than I've put into any other language I've learned, I have been unable to use Rust for real (ie. beyond beginner toys). A quick count of langs I've used professionally comes to about 10; I've learned many more to play with, much more successfully than with Rust, in much less time than I've spent with Rust. This reflects the experience of everyone I know who's tried it out (I'm the last man standing, having another crack at it this year).
The cultural peculiarity of this is that, in a field that generally lauds effort and intelligence, Rust advocates uniquely consider 'hard' or 'difficult' to be negative criticism. I don't hear this from (say) Haskell or Scala programmers. I certainly don't consider difficulty a negative - it depends on whether the payoffs are commensurate, and for Rust I think they are (I'd just quit otherwise).
As I'm 100% satisfied (no matter how often I'm unconvincingly 'corrected') that Rust is indeed harder than most programming languages to learn, I can only categorise this peculiarity as denial. Denial usually has ideological origins. I'm not entirely sure what they are in this case, though I have some ideas. These do feed into a sense that although I like the language, I don't think I like the Rust community much. I get strong religious/culty vibes from it - similar to the worst of the Linux community (for the record I also am a f/t Linux user, and I think that community has moderated considerably over time).
Yes, Rust's borrower checker is more complicated than equivalent features in other languages. If you read the writings of the Rust language developers, you'll notice that they basically admit that it's where Rust sinks its entire complexity budget, so there's no room to spend it anywhere else. But I don't think it's too complex--it's not complex enough for average developers to not be productive. I would contrast this with C++'s template metaprogramming, which I believe to be too complex for average developers to be productive (e.g., trying to write templates that switch based on the types of the parameters, especially before if constexpr).
Saying "rust is difficult" is silly to me because... it wasn't for me. Saying "rust is easy" is less silly to me because for me it was, but obviously for you it will be more silly because it isn't easy for you.
You'll find that many in the rust community are in fact very very sympathetic to your view that it's too hard to learn.
I have not found that. The very comment I'm responding to above is a correction, telling me "it's not hard", not saying on the writer's part that they didn't find it hard, and least of all with any curiosity towards any dev claiming that they find Rust hard, because that view is considered in the Rust commmunity plainly incorrect. This is what I have found, consistently in 2022. I plan to continue with Rust in 2023, but I'll probably largely go it alone.
Going it alone sounds like a recipe for it being very difficult. I'd recommend interacting with the community when you have questions, it'll be very helpful.
My only point this entire time is that saying "rust is difficult" is just as wrong as saying "rust is easy". Rust was easy for me, fact. Rust was hard for you, fact. Both things are true.
Hence there's no problem cloning a mutable reference implicitly, when calling a function. Notice that you can't for instance send it to another thread, however.
Not sure if that's good that this feature is automatic. Maybe it wouldn't be too bad to call functions like that for consistency:
go_break_stuff(&mut *my_stuff_ref);
There are also these patterns one has to learn like applying references to numbers in some corner cases, e.g. func(&123), and the fact polominus is being developed shows there is still room for improvement on the borrow checker logic.
Neither of those were the case for me. In fact, I'd previously tried and failed to learn C (to a level where I could achieve something useful). I did learn at work, which gave me time as I was able to work on it full time, but I had no help available as nobody else at the company knew Rust at the time.
> Rust advocates tend to come back with an aggressive 'proof' that Rust is factually not hard to learn (with implications we can all guess at). No other programming community frequently does this in my experience (and I have brushed against many). That community attitude itself must have causes that are worth thinking about.
I somewhat agree this behaviour isn't great. I suspect the cause is a whole bunch of people saying that Rust is too hard, and that the industry therefore shouldn't bother with it.
Factors I'd guess at:
- Previous experience with a language that uses functional patterns is a huge plus. Even if that's very lightweight functional patterns like you might find in JavaScript or Kotlin (past exposure to something like Haskell or Scala would be even better). - Previous experience with C++ helps. Especially if you write "modern" C++ with smart pointers, etc as this is rather close to how Rust works. - C experience can be a positive or a negative. If you are the sort of C programmer who is hyper aware of object lifetimes and programs carefully and defensively it'll likely help. If you're the kind of C programmer who likes to play fast and loose with safety so long as it works in practice then it'll likely hinder as you'll expect to be able to do things that you can't. - Java/C# and other strongly OOP languages can make learning Rust harder. OOP code tends to be full of graphs of objects that all point to each other, and you just can't do this in Rust. So if that's how you're used to programming your going have to completely relearn how to structure code.
And I guess some of it's just going to be how comfortable people are with abstraction. Rust is definitely heavier on the mathier side of programming (with a sophisticated type system, etc). Personally I find that makes it easier not harder than languages like Go and C which are much simpler in this respect, as when reading code in those languages I often find it hard to pick out those forest from the trees. But from what I've seen, programmers are pretty split on their preferences on this issue, and I imagine for some people this would make it harder.
Perhaps, though I haven't myself witnessed anyone saying it's 'too hard'. Just that people (like me and everyone I know who's tried to learn) find it 'hard'. Then we're informed it's not.
My thought about the underlying reasons was more along the lines of the Rust community's inclusion ethic. I find that entirely laudable in general. But like all ideologies it can lead to blind spots when insisted on against evidence. For example, it's even built into the (plainly false) official tagline: "A language empowering everyone to build reliable and efficient software." If someone finds Rust difficult, this creates a strong cognitive dissonance in fierce Rust advocates (for some reason so many are), which evokes vigorous denial.
> Previous experience with a language that uses functional patterns is a huge plus
My previous two 'learn for interest' languages were Clojure and Elixir. I could do more in both after a week than after some months of Rust.
In my case it's not unfamiliarity nor abstraction, but the sheer complexity that bleeds right through the ecosystem. For example: something I experience every time I try to do anything 'real': I search for the most common/recommended relevant library, go to its docs and find them utterly incomprehensible. It then takes days to glean enough to use the new library fluently. Even reading commandline args - detailed but conceptually simple - leads to a vastly complex horror of a library (I won't name names) that takes days to decipher. It's never taken me more than an hour or two in any other language. My Rust projects slow to a stupefying crawl.
For all that, I'm persisting (or rather re-starting, as my attempt earlier this year left me thoroughly demoralised) out of a combination of some real practical uses I have for it ("hard" doesn't to me in any way mean "bad"), and stubbornnes. But I'll do so without the 'help' of the community, which I find insufferable.
If you do interact with the Rust community, I suggest doing so in Shiny Happy Person guise.
[1] only the Rust community considers these 'complaints'