Stan whines about Rust¶
It’s not something I’ve spent a lot of letters talking about, but when it comes to programming languages, I have a clear favorite in C++. But I don’t think it’s perfect, and I may eventually publish a language of mine that will show some potential improvement areas. This may seem like a similarly crazy task as making a competitor to Git from scratch instead of supporting one of existing alternatives, but guess what: the alternatives aren’t exactly a net improvement.
There are two programming languages that are trying to position themselves as potential C++ replacements that I know of, Go and Rust. Go features mandatory garbage collection, which… you know, at that point you’re not that far a way from Microsoft’s .NET CLI setup, that’s nowhere near C++. With Rust, well. It’s definitely more honest. But it feels… thorny. It’s a Giant, Friendly Yet Clumsy Crab of C++-like programming languages. I’m not going to pretend I’m a Rust expert; this post is a rant written mostly after reminding myself Rust exists and reading through “The Book” on rustlang.org once afterwards. It’s likely that some points didn’t come through to me, maybe they could be clarified.
Let’s start by talking about Rust features which I like, and which I’m definitely considering to include in my language. Many of these are simply similar to C++ decisions, this includes a static typing system, equivalents of RAII, presence of object movement… However, it wouldn’t be fair to assume that “fixing obvious C++ flaws” is where Rust’s vision of improvement ends. The idea of compiler checking the lifetime of references seems somewhat obvious in retrospect, but that doesn’t change usefulness of it. Sometimes it’s interesting to see how some ideas at C++ Standardization Committee and in Rust are similar and wonder whether it’s simply convergence, or was there an inspiration going one way or another, as visible with C++’s std::spans and Rust’s slices as replacement for pointer arithmetic, or Herb Sutter’s suggested try operator and Rust’s question mark notation. The non-null notation is cool as well, though I wonder how well optimized it is - but that’s a rant about the language, not the compiler.
Some elements of Rust are what I would consider weird, in the sense that I’m not sure what were the authors thinking when they decided to make the language work like that, but I don’t really dislike them as much as I am befuddled by the choices. Usage of postfix exclamation mark for demarking macros is one example; I have no idea why didn’t they go with something else, but it’s not ultimately a problem.
A larger feature is lack of direct inheritance, with dataless interfaces becoming the virtual option vehicles, and in fact apparently vtable pointers being moved into “fat pointers” within references. I really don’t know what to think about this. It seems like it would result in a huge thought paradigm changes and this wasn’t particularly well explained in the book, aside from “Go also does this and massive OOP is bad”. There’s probably a point in suggesting use of interfaces instead of inheritance when separate interface facility is available, but this doesn’t leave much space for framework objects; maybe they can work differently.
There’s a few details that I’m not a huge fan of, but which I find ultimately understandable. All aspects of the Cargo build system feel more like something made because all the other “cool” languages have their own build systems which work similarly.
One thing that I want to focus on is lack of what I call “open typing”. To the best of my knowledge no widely known statically typed language made it work yet, it’s probably what one could call a “personal concept”. I mention it since multiple impl blocks for structs are the first step towards that, and I was feeling a little teased about the possibility of doing the same thing with fields - but, it’s not what I could really call a fault.
So I sounded pretty happy with Rust so far. What’s the problem with it, then?
Oh dear Lord, the notation.
This is actually a bit of a multifaceted issue. Let me explain.
Move by default¶
You know how people criticize C++ for deeply copying stuff all the time by default? Well, Java’s take on it was to involve a garbage collector and require explicit cloning of the values half the time. Rust is aligned to my interests enough to not involve a garbage collector, but deep copying of things still requires clone annotations. Instead, values are now moved by default, and proceed to disappear. This is slightly less of an issue than it sounds like since Rust encourages you to use references a lot, but it’s a huge gotcha, especially since it doesn’t apply to some types while applying to others.
But if you’re using references for your function parameters:
You need an ampersand to convert an object to a reference.
If you come to Rust without going through C++ it’s possible that you are squinting right now. So let’s move straight to the next point:
A vocabulary problem¶
The Rust terminology is in significant collision with C++ one. Given the amount of programming languages, that wasn’t completely avoidable, but some of the choices are extremely confusing. The best example is aforementioned reference, which is more like a C++ pointer than like a C++ reference in many ways. The Rust team could have tried to find a less used word for such an important element of the language, like they did with traits - there is a thing called a trait in C++, but it’s significantly harder to mix up with the Rust one than the reference. Imagine if it was called a concept or an interface.
let mut revenge¶
Rust has decided to make its equivalent of const object less verbose. That’s cool, given how cool const objects are in C++, and how Rust boosts their optimization potential by involving something similar to the C “restrict” business in its references. Unfortunately, the way they did it is simply taking the pain on all the non-const stuff by requiring a “mut” qualifier on it. Then, presumably due to the aforementioned “restrict” business, you have to reapply that keyword in parameter passes or it will go const again. Gently nudging me with a fork when I do a shadier thing instead of when I do a less shady thing still makes me angry about the fork, you know.
Generics, traits, and betrayal¶
Rust has a thing called a “trait” which is somewhere between a Java interface and C++-Wait-For-It-Edition concepts. Rust also has a thing called a generic which is supposed to work mostly like a C++ template. Unfortunately, while C++ Standards Committee decided to, at the last minute, pull concepts out of C++-Eventually-2011 due to how they required every template user to keep updating the bureaucratic declarations of conformance with concepts, Rust authors decided they were mostly fine with it, and in fact they forced generic authors to force their users to do it. In this case, C++ took… a while to come up with a suitable replacement, but the new concepts, in spite of the obvious handicaps, are a drastic improvement.
What is particularly hilarious is that Rust compiler comes with the ability of automatically making methods for some of the base traits, but you still have to tell it to do it in all cases.
I have no idea how one must feel in order to copy the triangle brackets and the namespace operator, of all things, from C++. Why? Just… why?
The Crab Flips Me On My Back¶
At this point, Mozilla has an existing Rust codebase and I doubt they would be particularly receptive to suggestions of changing half of their language by one guy with no experience. Certain elements could probably be adapted out without much damage to existing code, but they like it as is. I guess if I get less lazy I’ll roll out my own language over LLVM and we’ll see how that goes. Maybe I’ll even change my viewpoint on some of these.