But you're willing to write many comments complaining that Rust is hard to refactor. Rust is the easiest language to refactor with I've ever worked in, and I've used a couple dozen or so. When you want to change something, you change it, and then fix compiler errors until it stops complaining. Then you run it, and it works the first time you run it. It's an incredible experience to worry so little about unknown side-effects.
Their refactoring comments look focused on C versus C++ to me, with a bit of guessing Rust is like C++ in a way that is clearly labeled as speculation.
So I don't see the problem with anything they said about refactoring.
Both the LLM and humans doing it are relying on the fabulous work the rust compiler team has done in generating error messages with clear instructions on how to fix problems.
Oh? So how do you refactor a closure into a named function in Rust?
I have found this to be of the most common failure modes that makes people want to punch the monitor.
(Context: in almost all programming languages, a closure is practically equivalent to an unnamed function--refactoring a closure to a named function tends to be pretty straightforward. This isn't true for Rust--closures pick up variable lifetime information that can be excruciatingly difficult to unwind to a named function.)
Adding features in particular is a breeze and automatically the compiler/language will track for you the places that use only old set of traits.
Tooling is still newer though and needs polish. Generic handling is interesting at times and there are related missing features for that in the language, vis a vis specializations in particular.
Basic concurrency handling is also quite different in Rust than other languages, but thus usually safer.
You won't understand it unless refactor some Rust programs.
Bunny summed it up rather well. He said in most languages, when pull on some thread, you end disappearing into a knot and you're changes are just creating a bigger knot. In Rust, when pull on a thread, the language tells you where it leads. Creating a bigger knot generally leads to compile errors.
Actually he didn't say that, but I can't find the quote. I hope it was something like that. Nonetheless he was 100% spot on. That complexity you bemoan about the language is certainly there - but it's different to what you have experienced before.
In most languages, complex features tend lead to complex code. That's what made me give up on Python in the end. When you start learning Python, it seems a delightfully simple yet powerful language. But then you discover metaclasses, monkey patching, and decorators, which all seem like powerful and useful tools, and you use them to do cool things. I twisted Python's syntax into grammar productions, so you could write normal looking python code that got turned into an LR(1) parser for example. Then you discover other peoples code that uses those features to produce some other cute syntax, and it has a bug, and when you look closely your brain explodes.
As you say C doesn't have that problem, because it's such a simple language. C++ does have that problem, because it's a very complex language. I'm guessing you are making deduction from those two examples that complex languages lead to hard to understand code. But Rust is the counter example. Rust's complexity is forces you to write simple code. Turns out it's the complexity of the code that matters, not the complexity of the language.
But it does! To qoute my top-level comment:
> What about about race conditions, null pointers indirectly propagated into functions that don't expect null, aliased pointers indirectly propagated into `restrict` functions, and the other non-local UB causes?
In other words: you set some pointer to NULL, this is OK in that part of your program, but then the value travels across layers, you've skipped a NULL check somewhere in one of those layers, NULL crosses that boundary and causes UB in a function that doesn't expect NULL. And then that UB itself also manifests in weird non-local effects!
Rust fixes this by making nullability (and many other things, such as thread-safety) an explicit type property that's visible and force-checked on every layer.
Although, I agree that things like macros and trait resolution ("overloading") can be sometimes hard to reason about. But this is offset by the fact that they are still deterministic and knowable (albeit complex)
Then we have the functions that might be re-entrant or not, in the presence of signals, threads,...
Really? How confident are you to change a data structure that uses an array with linear search lookup to a dictionary? Or a pointer that now is nullable (or is now never null)?
Unless you have rigorous test or the code is something trivial, this would be a project of its own.
I am pretty sure I can swap out the implementation of the dictionary in the rust compiler and by the time the compilation issues are worked out, the code would be correct by the end of it (even before running the tests)
Refactoring Rust projects is clearly the easiest because the compiler and type system ensure the program is correct at least in terms of memory access and shared resource access. It doesn't protect me from memory leaks and logical errors. But since Rust has a built-in testing framework, it's quite easy to prepare tests for logical errors before refactoring.
C/C++ refactoring is a nightmare - especially in older projects without modern smart pointers. Every change in ownership or object lifetime is a potential disaster. In multi-threaded applications it's even worse - race conditions and use-after-free bugs only manifest at runtime, often only in production. You have to rely on external tools like Valgrind or AddressSanitizer.
Python has the opposite problem - too much flexibility. You can refactor an entire class, run tests, everything passes, but then in production you discover that some code was dynamically accessing an attribute you renamed. Type hints help, but they're not enforced at runtime.
Rust forces you to solve all these problems at compile time. When you change a lifetime or ownership, the compiler tells you exactly where you need to fix it. This is especially noticeable in async code - in C++ you can easily create a dangling reference to a stack variable that an async function uses. In Rust, it simply won't compile.
The only thing where Rust lags a bit is compilation speed during large refactors. But that's a small price to pay for the certainty that your code is memory-safe.
Another area where Rust absolutely excels is when using AI agents like Claude Code. It seems to me that LLMs can work excellently with Rust programs, and thanks to the support of the type system and compiler, you can get to functional code quickly. For example, Claude Code can analyze Rust programs very well and generate documentation and tests.
I think Rust with an AI agent has the following advantages:
Explicit contract - the type system enforces clear function interfaces. The AI agent knows exactly what a function expects and what it returns.
Compiler as collaborator - when AI generates code with an error, it gets a specific error message with the exact location and often a suggested solution. This creates an efficient feedback loop.
Ownership is explicit - AI doesn't have to guess who owns data and how long it lives. In C++ you often need to know project conventions ("here we return a raw pointer, but the caller must not deallocate it").
Fewer implicit assumptions - in Python, AI can generate code that works for specific input but fails on another type. Rust catches these cases at compile time.
