Hacker News Reader

I've seen as many of the tech geek stereotype where there's some level of masochistic appeal of C (especially C89) to them where "if it's hard and some people can't do it, then clearly it must be good and only for the best". Especially among hobbyists who've never had to maintain a production codebase or deal with changing user needs.

It's a bit ironic because Rust is also hard, but in a totally different way than C89 is hard. For instance, plenty of novice devs got started with C89 (or even C++) as their first "serious" programming language, but we're not even sure how Rust could be adapted to fit the same role. (Probably it would involve teaching the basics of Rust as if it was a quasi-Haskell, or at least a deeply weird ML. So quite different from how C89 is taught!)

On the contrary, I feel rust could be a great language for students because the borrow checker will slap sense into them from early on and each error it throws teaches them a little more on memory, data safety.

Being allowed to mutate an integer (or object) through two different pointers on the same thread is perfectly safe and even useful (in the general case, for non-tree-shaped problems) on a CPU. Rust adds unpleasant syntax to achieve it through Cell, unpleasant syntax and undefined behavior footguns to do it through raw pointers, and guaranteed undefined behavior to do it through aliasing &mut, so I no longer treat Rust as a general-purpose programming language. And when teaching beginners, this would give the incorrect impression that it's somehow undefined behavior to access memory through multiple pointers on the same thread (rather than merely a tradeoff which increases code flexibility at the cost of local reasoning).

And on a more personal level, I dislike the Rust community (and language, eg. for loops depend on trait resolution)'s heavy use of functional, generic, and higher-order function abstractions (rather than explicit imperative code amenable to sequential reasoning). I feel heavy use of generics is confusing in a language used to teach programming to students (and I think usefully learning mainstream programming revolves around algorithmic reasoning from the level of memory and bytes through structured programming, rather than pure functions and iterator combinators).

> Being allowed to mutate an integer (or object) through two different pointers on the same thread is perfectly safe

Not safe in the presence of variant records (i.e. enums in Rust), since one pointer might be used to derive a pointer to a field of one variant, and the object might concurrently be switched to a different variant. This is why Cell<> only relies on explicit get and set methods in the general case, where the entire object gets replaced. It's a well-known problem in temporal safety.

I find that Rust prefers to prevent programmers from committing undefined behavior, rather than giving them the tools required to achieve tasks, when these requirements come in conflict (unsafe code is built to allow implementing general functionality with unchecked code, but is more difficult to write without UB than the equivalent operations in safe Java or unsafe C++, partly because mutating a & or aliasing a &mut is instant UB)

You have a very odd definition of a general purpose language. All the functional ones are out, because immutability doesn’t let you mutate through any pointer.

In my observations, immutability-focused functional languages are generally either non-Turing-complete and built as configuration/declarative languages, or find niche usage and do not replace JS/C++/Java/C# for mainstream programmers building a spreadsheet or word processor (and aren't built for low-level programmers to replace C++ for a browser engine or video game or non-allocating real-time audio engine).

I already teach borrowing and ownership semantics as ways to avoid memory issues in C. While Rust is a more complex language on the whole, the fundamentals of borrowing and ownership are more broadly applicable concepts.

Rust 101 would definitely be a weed out class for students trying to get a CS degree!

Context: I took C++ back in college in 1999 and 2000. There were students who struggled with it then who dropped CS as a major. Later on I heard that Java took over as a 101 class. Now I think it’s Python.

Heap allocation doesn't really come up all that much in "101" programs, though. You have Strings of course, but that's about it. So I think teaching Rust could work, people just haven't gotten around to trying it. Most intros to Rust including the official book explicitly assume familiarity with some other programming language, and often fail to explain other things (at either the "low" or "high" level of semantics) that a 101 student wouldn't know.

And why would that be the case?

Indeed, the changing user needs aspect is inevitably the cause of so many memory corruption issues when old code, that's only really understood by that one or two grey beard still at the company and about to retire (or have already retired), needs to be changed. The change is put in and everything looks fine and then it's given to the customer who eventually complains of rare crashes on their previously-working-just-fine hardware. (Oh and you don't find out about the crashes until almost a year later given the long release cycle so have long forgotten what you did a year ago.) Inevitably it's used as an avenue for the sales engineers to upsell the customer on new hardware claiming their hardware is out of date. The software engineers all know this is a white lie given that the same software runs on both pieces of hardware and the hardware didn't actually change much.

It just ruins you over time and makes you wish you never had to work in this business.

> The people pushing for Rust are people who are highly experienced software engineers who have gone through a career of hardship because of memory unsafe languages and understand the tradeoffs.

The beginning of this statement is at odds with the end of it. Anyone who understands tradeoffs won't push anything because they know that which tradeoffs are acceptable is intimately dependent on the problem being solved, and without deep knowledge of the problem one cannot even begin to speak about accepting tradeoffs. Pushing someone who doesn't yet understand their problem towards a solution is the antithesis of engineering.

Moderately experienced code monkeys, who have been around enough problems to see failure, but not enough problems to understand that other people have different problems to solve seem to like to push languages onto other people, though. I'll give you that.

Not to mention most CVEs I see are C/C++ related. That’s not the sign of something we should keep building software in.

How much of that is due to the sheer volume of business critical software written in C/C++?

Surely you jest. Do you think most web services are written in C? Most enterprise software has been written in Java for a very long time. When you see exploits in other languages it’s often exploits due to the fact they were written in C/C++. Buffer overruns are no an artifact of the prevalence of C/C++, but that of the stack that you operate the C/C++ surface area is the weakest.

It attracts both. Which makes a lot of things harder than they need to be.