I'm kinda in the opposite camp. After doing a bunch of VB in my tweens and teens, I learned Java, C, and C++ in college, settling on mostly C for personal and professional projects. I became a core developer of Xfce and worked on that for 5 years.
Then I moved into backend development, where I was doing all Java, Scala, and Python. It was... dare I say... easy! Sure, these kinds of languages bring with them other problems, but I loved batteries-included standard libraries, build systems that could automatically fetch dependencies -- and oh my, such huge communities with open-source libraries for nearly anything I could imagine needing. Even if most of the build systems (maven, sbt, gradle, pip, etc.) have lots of rough edges, at least they exist.
Fast forward 12 years, and I find myself getting back in to Xfce. Ugh. C is such a pain in the ass. I keep reinventing wheels, because even if there's a third-party library, most of the time it's not packaged on many of the distros/OSes our users use. Memory leaks, NULL pointer dereferences, use-after-free, data races, terrible concurrency primitives, no tuples, no generics, primitive type system... I hate it.
I've been using Rust for other projects, and despite it being an objectively more difficult language to learn and use, I'm still much more productive in Rust than in C.
I think Rust is harder to learn, but once you grok it, I don't think it's harder to use, or at least to use correctly. It's hard to write correct C because the standard tooling doesn't give you much help beyond `-Wall`. Rust's normal error messages are delightfully helpful. For example, I just wrote some bad code and got:
--> src/main.rs:45:34
|
45 | actions.append(&mut func(opt.selected));
| ---- ^^^^^^^^^^^^ expected `&str`, found `String`
| |
| arguments to this function are incorrect
|
help: consider borrowing here
|
45 | actions.append(&mut func(&opt.selected));
|
I even had to cheat a little to get that far, because my editor used rust-analyzer to flag the error before I had the chance to build the code.
Also, I highly recommend getting into the habit of running `cargo clippy` regularly. It's a wonderful tool for catching non-idiomatic code. I learned a lot from its suggestions on how I could improve my work.
> I think Rust is harder to learn, but once you grok it, I don't think it's harder to use, or at least to use correctly. It's hard to write correct C because the standard tooling doesn't give you much help beyond `-Wall`.
When I say Rust is harder to use (even after learning it decently well), what I mean is that it's still easier to write a pile of C code and get it to compile than it is to write a pile of Rust code and get it to compile.
The important difference is that the easier-written C code will have a bunch of bugs in it than the Rust code will. I think that's what I mean when I say Rust is harder to use, but I'm more productive in it: I have to do so much less debugging when writing Rust, and writing and debugging C code is more difficult and takes up more time than writing the Rust code (and doing whatever less debugging is necessary there).
> Also, I highly recommend getting into the habit of running `cargo clippy` regularly. It's a wonderful tool for catching non-idiomatic code.
That's a great tip, and I usually forget to do so. On a couple of my personal projects, I have a CI step that fails the build if there are any clippy messages, but I don't use it for most of my personal projects. I do have a `cargo fmt --check` in my pre-commit hooks, but I should add clippy to that as well.
If you're using VS Code then you can add `"rust-analyzer.check.command": "clippy"` to your `settings.json`. I assume there's a similar setting for rust-analyzer in other editors.
Package managers are for running other people's code, I would not expect the same of static analysis tools, especially since they are of use while auditing other people's code before building/running it.
Cargo's threat model here is identical to that of rust analyser. If you trust your dependency tree sufficiently to run `cargo build`, then you trust it sufficiently to run rust analyser.
That's a fair distinction. Basically, it's easier to write C that compiles than Rust that compiles, but it's harder to write correct C than correct Rust.
Regarding Clippy, you can also crank it up with `cargo clippy -- -Wclippy::pedantic`. Some of the advice at that level gets a little suspect. Don't just blindly follow it. It offers some nice suggestions though, like:
warning: long literal lacking separators
--> src/main.rs:94:22
|
94 | if num > 1000000000000 {
| ^^^^^^^^^^^^^ help: consider: `1_000_000_000_000`
|
> it's still easier to write a pile of C code and get it to compile than it is to write a pile of Rust code and get it to compile.
As someone who is more familiar with Rust than C: only if you grok the C build system(s). For me, getting C to build at all (esp. if I want to split it up into multiple files or use any kind of external library) is much more difficult than doing the same in Rust.
Rust allows to provide more information about types (generic types, pointer usage) and checks it, while in C you have to rely on doc comments and checking the code manually. Or am I wrong and C allows to specify pointer nullability, pointer ownership and array bounds?
None of those things feature in any problem I deal with on a daily basis, whatever language I use.
So for example today I dealt with a synchronization issue. This turned out to not be a code bug but a human misunderstanding of a protocol specification saga, which was not possible to code into a type system of any sort. The day before was a constraint network specification error. In both cases the code was entirely irrelevant to the problem.
Literally all I deal with are human problems.
My point is Rust doesn't help with these at all, however clever you get. It is no different to C, but C will give you a superset of vulnerabilities on top of that.
Fundamentally Rust solves no problems I have. Because the problems that matter are human ones. We are too obsessed with the microscopic problems of programming languages and type systems and not concentrating on making quality software which is far more than just "Rust makes all my problems go away" because it doesn't. It kills a small class of problems which aren't relevant to a lot of domains.
(incidentally the problems above are implemented in a subset of c++)
> So for example today I dealt with a synchronization issue. This turned out to not be a code bug but a human misunderstanding of a protocol specification saga, which was not possible to code into a type system of any sort.
Maybe not in a reasonable language no, but there are advances in type systems that are making ever larger classes of behaviours encodable into types. For example, algebraic effects (can this function throw, call a remote service etc)
Vulnerabilities are bugs, so the C code will have more bugs than the Rust program.
You might say that the C and Rust code will have the same number of logic errors, but I'm not convinced that's the case either. Sure, if you just directly translate the C to Rust, maybe. But if you rewrite the C program in Rust while making good use of Rust's type system, it's likely you'll have fewer logic errors in the Rust code as well.
Rust has other nice features that will help avoid bugs you might write in a C program, like most Result-returning functions in the stdlib being marked #[must_use], or match expressions being exhaustive, to name a couple things.
> most Result-returning functions in the stdlib being marked #[must_use]
Actually it's a bit cleverer than that, and some people might benefit from knowing this. The Result type itself is marked #[must_use]. If you're writing a Goat library and you are confident that just discarding a Goat is almost always a mistake regardless of the context in which they got a Goat you too should mark your Goat type #[must_use = "this `Goat` should be handled properly according to the Holy Laws of the Amazing Goat God"] and now everybody is required to do that or explicitly opt out even for their own Goat code.
Obviously don't do this for types which you can imagine reasonable people might actually discard, only the ones where every discard is a weird special case.
Types I like in Rust which help you avoid writing errors the compiler itself couldn't possibly catch:
Duration - wait are these timeouts in seconds or milliseconds? It's different on Windows? What does zero mean, forever or instant ?
std::cmp::Ordering - this Doodad is Less than the other one
OwnedFd - it's "just" a file descriptor, in C this would be an integer, except, this is always a file descriptor, it can't be "oops, we didn't open a file" or the count of lines, or anything else, we can't Add these together because that's nonsense, they're not really integers at all.
How can it not be true? One of the primary features of the rust compiler is enforcing memory safety at compile-time. C doesn't have anything like that. There are an entire class of bugs that are impossible to implement in rust.
I have written a lot of C and Rust. The notion that identical projects written in both languages would have identical numbers of (or even severity of) bugs is laughable on its face.
I mean, literally just not being able to deref a NULL pointer by itself is enormous.
You can dereference null pointers in Rust, it's also easy to make null pointers. The thing is I'm not surprised you didn't realise this is possible because you would never actually do this in Rust since it's obviously a bad idea.
let p: *const i32 = std::ptr::null(); // Pointer to a signed 32-bit integer, but it's null
unsafe { some_local = *p }; // Dereferencing the null pointer, bang
If we instead use references they're nicer to work with, don't need unsafe, perform the same, and are never null. So obviously that's what real Rust programmers almost always do.
Of course it’s possible. I am well aware it’s possible. You can also just call into a C library that hands you a null pointer. Further, it’s wholly possible to invalidate every single one of Rust’s safety guarantees if you try hard enough.
In practice I have never encountered a null pointer in Rust. Most Rust programmers have never encountered a null pointer. For most non-FFI practical purposes you can act as if they don’t exist.
> It's hard to write correct C because the standard tooling doesn't give you much help beyond `-Wall`
I won't disagree that correct C is harder to write, but it's not 2005 anymore and standard tooling gives you access to things like asan, msan, ubsan, tsan, clang-tidy...
C-with-sanitizers is miles ahead of what C used to be, but just a couple weeks ago I ran into a dangling pointer bug that ASan doesn't catch. (Sidenote 5 here: https://jacko.io/smart_pointers.html) It seems like one of the big downsides of sanitizers is that they can't instrument other people's code, in this case the C standard library.
> Also, I highly recommend getting into the habit of running `cargo clippy` regularly.
You can also have that hooked up to the editor, just like `cargo check` errors. I find this to be quite useful, because i hace a hard time getting into habits, especially for thing that i'm not forced to do in some way. It's important that those Clippy lints are shown as soft warnings instead of hard errors though, as otherwise they'd be too distracting at times.
* Rust errors can be equally unhelpful. Also, the error you posted is hands down awful. It doesn't tell you what went wrong, and it's excessively naive to rely on compiler to offer a correct fix in all but the most trivial cases. When errors happen, it's a consequence of an impasse, a logical contradiction: two mutually exclusive arguments have been made: a file was assumed to exist, but was also assumed not to exist -- this is what's at the core of the error. The idiotic error that Rust compiler gave you doesn't say what were the assumptions, it just, essentially, tells you "here's the error, deal with it".
* In Rust, you will have to deal with a lot of unnecessary errors. The language is designed to make its users create a host of auxiliary entities: results, options, futures, tasks and so on. Instead of dealing with the "interesting" domain objects, the user of the language is mired in the "intricate interplay" between objects she doesn't care about. This is, in general, a woe of languages with extensive type systems, but in Rust it's a woe on a whole new level. Every program becomes a Sisyphean struggle to wrangle through all those unnecessary objects to finally get to write the actual code. Interestingly though, there's a tendency in a lot of programmers to like solving these useless problems instead of dealing with the objectives of their program (often because those objectives are boring or because programmers don't understand them, or because they have no influence over them).
I don't follow your first point—the compiler is pointing out exactly what the problem is (the argument has the incorrect type) and then telling you what you likely wanted to do (borrow the String). What would you see as a more helpful error message in this case?
The compiler says "expected X, but found Y". I don't know how to interpret this: is the type of the thing underlined with "^^^" X or Y? "Expected" and "found" are just like "up" and "down" in space: they are meaningless if you don't know what the compiler expects (and why should it?).
What it needs to say is something along the lines of "a function f is defined with type X, but is given an argument of type Y": maybe the function should be defined differently, maybe the argument needs to change -- it's up to the programmer to decide.
I dunno, I feel like if you've used a compiler regularly, "expected X, but found Y" is a pretty common idiom/shorthand that people understand. Your wordier version of that feels unnecessary to me.
C is a low level language and deals with things close to the metal. It's probably not fun to write a large business app in barebones C but you having control over low level things makes other things possible and very fast too. Depending on the type of problem you have use the appropriate and favorite language.
Since it's underlining code you wrote, it must be "found" that is highlighted, not "expected". Much like up and down, gravity exists to ground all of us in the same direction.
I'm over here with TTS: Underlining in a terminal rarely translates to audio. It isn't the only consideration that needs to be made, when making things clear.
One thing that could help here is that the compiler is able to offer output in JSON, allowing you to format the messages however you'd like: https://doc.rust-lang.org/rustc/json.html
I'm not aware of an existing tool to produce blind-friendly output, but this would at least be a part of that!
The code tend to be loaded with primitives that express ownership semantics or error handling. Every time something changes (for instance, you want not just read but also modify values referenced by the iterator) you have to change code in many places (you will have to invoke 'as_mut' explicitly even if you're accessing your iterator through mutable ref). This could be attributed (partially) to the lack of function overload. People believe that overload is often abused so it shouldn't be present in the "modern" language. But in languages like C++ overload also helps with const correctness and move semantics. In C++ I don't have to invoke 'as_mut' to modify value referenced by the non-const iterator because dereferencing operator has const and non-const overloads.
Async Rust is on another level of complexity compared to anything I used. The lifetimes are often necessary and everything is warpped into mutliple layers, everything is Arc<Mutex<Box<*>>>.
> Rust errors can be equally unhelpful. Also, the error you posted is hands down awful. It doesn't tell you what went wrong, and it's excessively naive to rely on compiler to offer a correct fix in all but the most trivial cases
What? It tells the user exactly what's wrong
> Every program becomes a Sisyphean struggle to wrangle through all those unnecessary objects to finally get to write the actual code
That is the cost of non-nullable types and correctness. You still have to do the Sisyphean struggle in other programming languages, but without hints from the compiler.
The biggest problem with C is that doesn't even have enough features to help you build the features and abstractions you need and want.
For example with C++ the language offers enough functionality that you can create abstractions at any level, from low level bit manipulation to high level features such as automatic memory management, high level data objects etc.
With C you can never escape the low level details. Cursed to crawl.
Back in 1994/95, I wrote an API, in C, that was a communication interface. We had to use C, because it was the only language that had binary/link compatibility between compilers (the ones that we used).
We designed what I call "false object pattern." It used a C struct to simulate a dynamic object, complete with function pointers (that could be replaced, in implementation), and that gave us a "sorta/kinda" vtable.
Worked a charm. They were still using it, 25 years later.
That said, I don't really miss working at that level. I have been writing almost exclusively in Swift, since 2014.
> We designed what I call "false object pattern." It used a C struct to simulate a dynamic object, complete with function pointers (that could be replaced, in implementation), and that gave us a "sorta/kinda" vtable.
You were not alone in this. It is the basis of glib's GObject which is at the bottom of the stack for all of GTK and GNOME.
Sure, that's a pretty common pattern in use in C to this day. It's a useful pattern, but it's still all manual. Forget to fill in a function pointer in a struct? Crash. At least with C++ it will fail to compile if you don't implement a method that you have to implement.
You still need to think about error handling, and it's not standardized because everyone else will also have to think about it ad hoc.
You'll also still need to think about when to copy and move ownership, only without a type system to help you tell which is which, and good luck ensuring resources are disposed correctly (and only once) when you can't even represent scoped objects. `goto` is still the best way to deal with destructors, and it still takes a lot of boilerplate.
No, but that tends to gradually happen all by itself from my experience, even in solo projects. Because it feels stupid writing something yourself when you have access to stdlib. Also, I feel rolling your own stuff in C++ is a major pita in comparison to C, too many rules to memorize for me.
That pretty much is the definition of C. It was designed to be a system language, that was "one step beyond" (my own Madness reference) the assembler.
It's a dangerous tool, and should be wielded by experts. The same goes for all kinds of similar stuff.
And "experts" is kind of a darwinian thing. There's not really that many folks that can successfully wield it in large quantities (a certain cranky Finn comes to mind). There's a ton of folks that are surfing the Dunning-Kruger wave, that think they are up to it, but they tend to face-plant.
For myself, I learned to do it right, but it was difficult, stressful work, and I don't miss it. Working in Swift is a joy. I don't have to worry about things like deallocations, and whatnot.
I've never met an expert who writes c well all of the time, so I have been led to believe it's simply not possible. Folks with 20+ years of experience will still regularly write a use after free but which is not caught until it hits production. I'm not sure I've actually seen C code free of concurrency bugs. If it's possible it takes years of battle hardening with no feature work. Humans are prone to make mistakes and we need to use as many tools as possible to help us avoid making those mistakes. They are recently looking into introducing thread annotations for locks in the Linux kernel and I expect them to find lots of real bugs lurking in otherwise well maintained subsystems. I've rewritten swaths of code into rust and always seen it find a bug in the original code's design that we were lucky to never run into.
haha every segfault puts me back down into the valley of despair xD. thats totally ok!
higher level languages often make it so you can focus on the language mostly. C and assembly, i think, you need a lot of context. what is running the code, and how.
Of course you can. It's quite the opposite actually. The downside is that in C you have to code a bunch of abstractions _yourself_. See how large projects like the Linux kernel make extensive use of macros to implement an object system.
so confused by this one. C is the most free to build systems and abstractions. it doesnt lock you down into any paradigm so you can build your own....
you can abstract away perfectly fine low level details and use your own high level constructs to build what you want in a memory safe way...
high level daya objects?
sure it doesnt have garbage collection, so it motivates you not to leave garbage laying around for some slow background process to collect.
actually you can build this in C you just do not want to...
you can give all your objects reference counters and build abstractions to use that, you can implement 'smart pointers' if u want, and have threads do garbage collection if u want... why not? what exactly is stopping you?
maybe its less convenient to go that route. but impossible? nope.
connect GDB to your c++ program and see how it works... its not like c++ suddenly doesnt become machine code. and c translates perfectly fine to machine code...
i never really understand why these get compared. i wouldn't expect that much overlap in the audiences
zig seems like someone wanted something between C and "the good parts" of C++, with the generations of cruft scrubbed out
rust seems like someone wanted a haskell-flavoured replacement for C++, and memory-safety
i would expect "zig for C++" to look more like D or Carbon than rust. and i'd expect "rust for C" to have memory safety and regions, and probably steal a few ocaml features
The single best feature (and I would say the _core_ feature separating it from C) that C++ has to offer is RAII and zig does not have that. So I don’t know which good parts of C++ they kept. Zig is more of its own thing, and they take from wherever they like, not just C++.
I really wish that were true but it isn’t. Modern C++ templates/constexpr are much more powerful and expressive than any Zig comptime equivalent.
The power and expressiveness of the C++ compile-time capabilities are the one thing I strongly miss when using other languages. The amount of safety and conciseness those features enable makes not having them feel like a giant step backward. Honestly, if another systems language had something of similar capability I’d consider switching.
I have written a lot of Zig comptime code and ended up finding the opposite. In C++ I find I have to bend over backward to get what I want done, often resulting in insane compile times. I've used metaprogramming libraries like Boost Hana before to have some more ergonomics, but even that I would consider inferior to comptime.
Out of curiosity, do you happen to have any examples of what you describe, where C++ is more powerful and expressive than Zig?
If it looks anything like what I read in "Modern C++ Design" 20+ years ago then I'll pass. That book made me realize the language wasn't for me anymore.
It looks nothing like C++ decades ago, it is effectively a completely different language. I found C++ unusable before C++11, and even C++11 feels archaic these days. Idiomatic C++20 and later is almost a decent language.
"Modern C++ Design" was the first book that highlighted template metaprogramming and showed you ways to use it as a Turing complete programming language in itself. Even a decade+ after it was printed my friend said it was recommended reading for employees within Google so I gave him mine - but I'm not sure about these days.
Talking about how much you can do with C++ templates made me think of that.
Zig has defer, which is arguably way simpler. Is there something RAII can do that defer can't?
Not everyone likes RAII by itself. Allocating and deallocating things one at a time is not always efficient. That is not the only way to use RAII but it's the most prevalent way.
defer can't emulate the destructors of objects that outlives their lexical scope. Return values and heap objects are examples of these since they outlive the function they were created in. defer only supports enqueuing actions at lexical boundaries.
If you destroy an object that outlives the lexical scope it was created in, then you have to clean up manually.
I have, and I do find Zig impressive, but it doesn't go far enough for me. I don't want a "better C", I want a better systems language that can also scale up for other uses.
I like strong, featureful type systems and functional programming; Zig doesn't really fit the bill for me there. Rust is missing a few things I want (like higher-kinded types; GATs don't go far enough for me), but it's incredible how they've managed to build so many zero- and low-cost abstractions and make Rust feel like quite a high-level language sometimes.
Rust is not a modern C++, their core models are pretty different. Both Rust and C++ can do things the other can’t do. C++ is a more focused on low-level hyper-optimized systems programming, Rust is a bit higher level and has stronger guardrails but with performance closer to a classic systems language.
I do think Zig is a worthy successor to C and isn’t trying to be C++. I programmed in C for a long time and Zig has a long list of sensible features I wish C had back then. If C had been like Zig I might never have left C.
What do you consider the difference in their core models.
Rust and C++ both use RAII, both have a strong emphasis on type safety, Rust just takes that to the extreme.
I would like to even hope both believe in 0 cost abstractions, which contrary to popular belief isn't no cost, but no cost over doing the same thing yourself.
In many cases it's not even 0 cost, it's negative cost since using declarative programming can allow the compiler to optimise in ways you don't know about.
C++ is (according to Bjarne Stroustrup) a general purpose programming language that can be used to build general business software and applications with, not just a systems PL. This is why perf is all over the place —- the core language is fast like C but the stdlib contains terribly slow code (regex, exceptions) and ways to waste lots of cycles (wrapping everything in smart pointers, using std::map instead of unordered map).
The answer there [0] doesn't support this. From the answer:
> There is no question that libstdc++'s implementation of <regex> is not well optimized. But there is more to it than that. It's not that the standard requirements inhibit optimizations so much as the standard requirements inhibit changes.
The answer's one comment expands on this, it sounds like they're not able to add a sophisticated optimising regex engine into the libstd++ shared library (i.e. non-inline) as this would be an ABI break.
Perhaps other implementations of the C++ standard library perform better.
In my understanding, other implementations cannot perform better, because the root cause is how it is defined in the standard. Basically, any better implementation would be non-conforming. It’s not the kind of ABI issue where simply choosing a different one would help; the flaw comes directly from the definition, which cannot be changed.
Of course, non-std implementations of regexes in C++ don’t have this issue. Were strictly talking about the standard library one.
Almost everyone wants to claim to be modern c and not modern C++ because everyone shit on C++
If anyone is the modern c++ its D. They have template metaprogramming while rust has generics and macros.
Rust doesn't have constructors or classes or overloading. I believe its type system is based on Hindley–Milner like ML or Haskell and traits are similar to Haskell type classes. Enums are like tagged Union/sum types in functional programming and rust uses Error/Option types like them. I believe rust macros were inspired by Scheme. And finally a lot of what makes it unique was inspired by Cyclone (an obscure excitemenal language that tried to be a safer C) and other obscure research languages.
I guess rust has RAII that's one major similar to c++. An there's probably some similarities in low level memory access abstraction patterns.
But I'd describe rust as more of an imperative non GC offshoot of MLs then a modern c++ evolution.
I suffered writing those for many years. I finally simply learned not to do them anymore. Sort of like there's a grain of sand on the bottom of my foot and the skin just sort of entombed it in a callous.
I've seen you make these kinds of comments before on other articles. Please stop. Not everyone is perfect and can forevermore avoid making any mistakes. I strongly suspect your opinion of your skill here is overinflated. Even if it isn't, and you really are that good, everyone cannot be in the top 0.00001% of all programmers out there, so your suggestion to "simply" learn not to make mistakes is useless.
This all just comes off incredibly arrogant, if I'm being honest.
I think a more charitable interpretation of what he said was, "after sufficient practice, I became good enough to start avoiding those pitfalls."
It's not all that different from learning any challenging task. I can't skateboard to save my life, but the fact that people can do it well is both admirable and the result of hundreds or thousands of hours of practice.
Skilled people can sometimes forget how long it took to learn their talent, and can occasionally come off as though the act is easy as a result. Don't take it too harshly.
I know, but it's the truth. Consider another case. I worked in a machine shop in college trying to make my steam engine. The man who ran the shop, was kind enough to show me how to operate the machines.
Those machines were like C, built during WW2. They had no user friendly features. The milling machine was a particular nightmare. It was festooned with levers and gears and wheels to control every aspect. There was no logic to the "user interface". Nothing was labeled. The worst one was a random lever that would reverse the operation of all the other levers. My terror was wrecking the machine by feeding the tool mount into the cutting bit.
I would make a part, and it would come out awful - things like the surface finish was a mess. If he had some time, he'd come over and help me. He'd turn a wheel, "ting" the blade on a grinding wheel, adjust the feed, and make a perfect part (all by eyeball, I might add). The man was just incredible with those machines. I was just in awe. He never made a mistake. Have you ever tried to get a part centered properly in a 4-jaw chuck? Not me. He could on the first try every time.
You come off as incredibly arrogant too, you just don't realise it because you have the current mainstream opinion and the safety of a crowd.
Do you know how fucking obnoxious it is when 200 people like you come into every thread to tell 10 C or Javascript developers that they can't be trusted with the languages and environments they've been using for decades? There are MILLIONS of successful projects across those two languages, far more than Rust or Typescript. Get a fucking grip.
Well, not 100% true... With Rust being the principal language that compiles to webassembly, I'm sure there is quite a lot of JS code being re-written in Rust right now.
But you'll be called things like "catastrophically unprofessional" [1] if you choose JavaScript over TypeScript, or dynamically-typed Python over MyPy.
He's not making a comment about everyone, it's a specific comment about how often long time C programmers make basic mistakes after a million SLOC or so.
In this instance Walter is correct - the mistakes he listed are very rarely made by experienced C programmers, just as ballet dancers rarely trip over their own feet walking down a pavement.
The problem of those errors being commonplace in those that are barely five years in to C coding and still have another five to go before hitting the ten year mark still exists, of course.
But it's a fair point that given enough practice and pain those mistakes go away.
So true. I know dancers who have been at it for decades. They "make it look easy", and it is easy for them. But try to do it yourself, and you look like a moose.
A friend of mine trains dancers for a living. He says the most effective teaching tool is the video camera. But he doesn't bring it out until his student is "hooked" on dancing, because prematurely showing them the video of them dancing just crushes them.
P.S. You can tell a ballet dancer just by the way they walk down the street. I once was at the airport and saw a lady unloading her bags. I just said Hi, you must be a ballet dancer! She replied nope, I'm an ice dancer, off to a competition. LOL.
> just as ballet dancers rarely trip over their own feet walking down a pavement
What about about walking down a busy construction site? The most charitable and correct interpretation I can think of is "I'm a professional. Seatbelts and OSHA destroy my productivity."
> What about about walking down a busy construction site?
Coordinated people with some years of experience pay attention to the ground and overhead cranes and conveyor belts and survive walking through construction sites, mine sites, aviation hangers, cattle yards, musters, et al on a routine basis. I'm 60+ and have somehow navigated all those environs - including C for critical system control.
These are dangerous environments. No one denies this. It's still true that the longer you inhabit such spaces the safer your innate learned behaviour is.
C has seatbelts and OSHA - valgrind, et al tools abound for sanity checking.
Walter's GP statement is literally little more than "eventually you grow out of making the simple basic maistakes" - eventually, after some years of practice - which is a real problem with C, it takes time to not make the basic mistakes. After all that, there's always room, in C, in Rust, whatever, to make non basic non obvious mistakes.
> After all that, there's always room, in C, in Rust, whatever, to make non basic non obvious mistakes.
Correct, I guess. The number of relatively obvious mistakes should decrease with experience. And it stands to reason that eventually it settles near zero for some part of developer community.
How close to zero and which part of community? Statistic is scarce.
> C has seatbelts and OSHA - valgrind, et al tools abound for sanity checking.
Optional tools with no general enforcement. That is more like elective vaccination or travel advisories. That is, no, no seatbelts and no OSHA.
> The number of relatively obvious mistakes should decrease with experience
Yes, and the kind of mistakes I make have changed. Now they are usually a failure to understand the problem correctly, or are simply typos.
> Optional tools with no general enforcement
That's right. The tools don't work unless you use them. With D the idea is to build the tool into the language itself. I once annoyed the Coverity folks by saying D's purpose is to put Coverity out of business.
I'd imagine the main way one reduces instances of these mistakes is to restrict resource ownership into certain patterns which have a clear place for freeing, and rules that ensure it's always reached, and only once.
There are many approaches depending on the type of program or suite of programs being built.
Always pairing the creation of free() code and functions with every malloc() is one discipline.
Another, for a class of C utilities, is to never free() at all .. "compute anticipated resource limits early, malloc and open pipes in advance, process data stream and exit when done" works for a body of cases.
In large C projects of times past it's often the case that resource management, string handling, etc are isolated and handled in dedicated sub sections that resemble the kinds of safe handling methods baked into modern 'safe' languges.
Similarly, I went from writing a lot of C to Python, and I appreciate both of them for almost opposite reasons. I ended up finding I like Cython quite a bit, even though the syntax leaves much to be desired. The power, performance, and flexibility of C combined with the batteries included nature of Python is a match made in heaven.
You're also still very much free to write either language purely, and "glue" them together easily using Cython.
That is very interesting. You have quite the resume too. While I've dabbled in nearly everything, I'm a day to day pro C# developer and I absolutely love it. I've never been upset or had a complaint. If I were forced off for some reason, I'd just go to Typescript. I can't imagine using C. Perhaps with some form of AI valgrind. The problems C solved are just not relevant any longer, and it remains entrenched in 2025. Rust with AI analysis will be amazing to see the results of.
I will say the more recent additions to C++ at least have solved many of my long standing issues with that C-variant. Most of it was stuff that was long overdue. Like string formatting or a thread safe println. But even some of the stuff I didn’t think I would love has been amazing. Modules. Modules bro. Game changer. I’m all in. Honestly C++ is my go to for anything that isn’t just throw away again. Python will always be king of the single use scripts.
The problem is that they are _additions_, C++ has such absurd sprawl. The interactions between everything in this massive sprawl is quite difficult to grasp
I lost interest in keeping up with C++'s advances more than a decade ago.
The problem is that I want a language where things are safe by default. Many of the newer stuff added in C++ makes things safe, perhaps even to the level of Rust's guarantees -- but that's only if you use only these new things, and never -- even by accident -- use any of the older patterns.
I'd rather just learn a language without all that baggage.
I Really want to use Rust. But Rust has so many std functions that kind of abstracting
how does it work under the hood
For Ex: I still has no idea what clone() does, how does it interact with memory, on heap or stack
, does it create a new instance, or just modify metadata of that object. Sometime creating a new instance
is a big no-no because it takes a lot of memory.
Same thing with "ownership transfer", is variable freed at that moment, etc.
I bet i could find answers on internet, but rust has like 500 std functions, so the task is tedious
What is your current language of choice, both C and C++ have the same problems as what you just described.
Regarding ownership transfer it is even worse in C, what if you forget, after moving an object out of a variable, to set that variable to NULL, then free that variable, that's a use after free. At least in C++ you have move semantics although it is still error prone. In rust it's a compiler error.
Copy and Clone is the same, and both are opt-in for your own types, by default the only options are move or reference for your own types, in C and C++ by default it is copy, which again leads to use after free in the situations you complained about.
I feel if these are your complaints you will actually benefit from spending some time in Rust.
If your preferred language is higher level languages with GC that is reference by default I encourage you to try any of the systems level programming languages, the things you complain about are things that are important for a language to have constructs for, reference semabtics by default causes many issues, and becomes untenable in the parallel world we live in.
Well, i dont use C++ much(i'm FW engineer, most of my stuff is in C).
The std in C is simple and explicit. For Ex: I can make an educated guess how memcpy() work by looking at its signature. It takes pointer to src and destination, and size, so i can guess it does not allocate any new memory(or if it has, it has to be some kind of optimization reason).
Another example is strstr(), it returns pointer to a piece of memory i provided to it, so i can safely do some pointer math with the return value.
It's true that i do not spend much time in Rust, so maybe i'm missing some fundamental things. I guess my mistake is trying to apply my knowledge in C to rust.
But still, it's kind of irritating now knowing (or guessing) how does function work just by looking at at its signature.
You can surmise the behavior from the signature of clone in the same way that you can memcpy(), it returns an owned type so it will allocate. If you want to match the signature of memcpy then you don't want the allocation and you can instead call clone_from() which takes a mutable borrow of something you had to allocate.
> piece of memory i provided to it, so i can safely do some pointer math with the return value.
Except pointer math is never going to be safe in C, a particular case might be bugfree but it is not safe. Moreover, nothing in the C language says that the provenance of the output pointer matches the provenance of the input pointer (it could take a reference to a pointer from some other thread while that other thread is freeing that memory). In rust you will pass a pointer and get a pointer back with the same lifetime, then you can safely use the pointer in that lifetime bound: the provenance is part of the signature. So in this case, you are incorrectly assuming things from the C signature while the corresponding rust signature definitively tells you.
So yeah, if you learn more about rust then you will see that in fact it tells you more than the corresponding C signatures.
> It takes pointer to src and destination, and size, so i can guess it does not allocate any new memory
Does the pointer provided by src get altered in any way? Might is be NULL after calling memcpy? What happens if I pass NULL to dst? Is size in bytes or whatever the pointer is pointing to?
The moment you need to read a man page to get any of those mab pages you can read the docs for clone and get all the information you would need.
> Another example is strstr(), it returns pointer to a piece of memory i provided to it
This is not at all clear from the signature, from the signature it might allocate and return a new string entirely that you would need the deallocate, the only way to know that is to read the docs which runs into the problem again.
And again there is no indications that the pointers passed into the function are not mutated in any way other than convention and documentation.
Rust makes all of these things explicit with a compile error if you fail an invariant of the type systems.
Btw it's possible to encode the same invariants in C++ but isn't the default most of the time.
Isn't guessing what a function does based purely on its name pretty risky? There's usually nuance, so if I'm not already familiar with a function, I'm looking up its docs at least once.
I fully understand that sentiment. For several years now, I have also felt the strong urge to develop something in pure C. My main language is C++, but I have noticed over and over again that I really enjoy using the old C libraries - the interfaces are just so simple and basic, there is no fluff. When I develop methods in pure C, I always enjoy that I can concentrate 100% on algorithmic aspects instead of architectural decisions which I only have to decide on because of the complexity of the language (C++, Rust). To me, C is so attractive because it is so powerful, yet so simple that you can hold all the language features in your head without difficulty.
I also like that C forces me to do stuff myself. It doesn't hide the magic and complexity. Also, my typical experience is that if you have to write your standard data structures on your own, you not only learn much more, but you also quickly see possibly performance improvements for your specific use case, that would have otherwise been hidden below several layers of library abstractions.
This has put me in a strange situation: everyone around me is always trying to use the latest feature of the newest C++ version, while I increasingly try to get rid of C++ features. A typical example I have encountered several times now is people using elaborate setups with std::string_view to avoid string copying, while exactly the same functionality could've been achieved by fewer code, using just a simple raw const char* pointer.
About 16 years ago I started working with a tech company that used "C++ as C", meaning they used a C++ compiler but wrote pretty much everything in C, with the exception of using classes, but more like Python data classes, with no polymorphism or inheritance, only composition. Their classes were not to hide, but to encapsulate. Over time, some C++ features were allowed, like lambdas, but in general we wrote data classed C - and it screamed, it was so fast. We did all our own memory management, yes, using C style mallocs, and the knowledge of what all the memory was doing significantly aided our optimizations, as we targeted to be running with on cache data and code as much as possible. The results were market leading, and the company's facial recognition continually lands in the top 5 algorithms at the annual NIST FR Vendor test.
This however, is an implementation strategy, that most business/"enterprise" developers will fight to the teeth, because they believe, that OOP is the only workable way to create abstractions and that any successful project must have inheritance in it. They never learned another way.
Actually Rust goes a long way towards a similar strategy, by avoiding inheritance and instead relying on structs and traits. That already avoids a lot of BS programming. I am very glad they threw out inheritance and classes. Great design decision right there. I wish FP was made more convenient/possible in Rust though.
Funnily enough, 16 years ago, I too was in exactly this type of company. C++, using classes, inheritance only for receiving callbacks, most attributes were public (developers were supposed to know how not to piss outside the bowl), pthreads with mutexed in-memory queues for concurrency, no design patterns (yes we used globals instead of Singleton) etc. So blazingly fast we were measuring latencies in sub 100-microseconds. Now, when modern developers say something is "blazingly fast" when it's sub-second, I can only shake my head in disbelief.
Yes, very similar. We had pthreaded mutexed queues too, and measured timings with clock_gettime() with CLOCK_MONOTONIC. Our facial template match runs at 25M compares per second per core, and simply keeping that pipeline fed required all kinds of timing synchronizations.
The only community I know that produces developers that know this kind of stuff intimately are console game programmers, and then only the people responsible for maintaining the FPS at 60. I expect the embedded community knows this too, but is too small for me to know many of them to get a sense of their general technical depth.
Slightly better ergonomics I suppose. Member functions versus function pointers come to mind, as do references vs pointers (so you get to use . instead of ->)
Yeah, slightly better ergonomics. Although we could, we simply did not use function pointers, we used member functions from the data class the data sat inside. We really tried to not focus on the language and tools, but to focus on the application's needs in the context of the problem it solves. Basically, treat the tech as a means to an end, not as a goal in itself.
Try doing C with a garbage collector ... it's very liberating.
Do `#include <gc.h>` then just use `GC_malloc()` instead of `malloc()` and never free. And add `-lgc` to linking. It's already there on most systems these days, lots of things use it.
You can add some efficiency by `GC_free()` in cases where you're really really sure, but it's entirely optional, and adds a lot of danger. Using `GC_malloc_atomic()` also adds efficiency, especially for large objects, if you know for sure there will be no pointers in that object (e.g. a string, buffer, image etc).
There are weak pointers if you need them. And you can add finalizers for those rare cases where you need to close a file or network connection or something when an object is GCd, rather than knowing programmatically when to do it.
But simply using `GC_malloc()` instead of `malloc()` gets you a long long way.
You can also build Boehm GC as a full transparent `malloc()` replacement, and replacing `operator new()` in C++ too.
At first I really liked this idea, but then I realised the size of stack frames is quite limited, isn't it? So this would work for small data but perhaps not big data.
In theory, this is a compiler implementation detail. The compiler may chose to put large stacks in the heap, or to not even use a stack/heap system at all. The semantics of the language are independent of that.
In practice, stack sizes used to be quite limited and system-dependent. A modern linux system will give you several megabites of stack by default (128MB in my case, just checked in my linux mint 22 wilma). You can check it using "ulimit -all", and you can change it for your child processes using "ulimit -s SIZE_IN_KB". This is useful for your personal usage, but may pose problems when distributing your program, as you'll need to set the environment where your program runs, which may be difficult or impossible. There's no ergonomical way to do that from inside your C program, that I know of.
Its a giant peeve of mine that automatic memory management, in the C language sense of the resource being freed at the end of its lexical scope, is tied to the allocation being on the machine stack which in practice may have incredibly limited size. Gar! Why!?
Ackshually, it has nothing to do with the C language. It's an implementation choice by some compilers. A conforming implementation could give you the whole RAM and swap to your stack.
Yes, but does any implementation actually do that?
AFAIK Ada is typically more flexible, but that has to do with the language actually giving you enough facilities to avoid heap allocations in more cases - e.g. you can not only pass VLAs into a function in Ada, but also return one from a function. So it becomes idiomatic, and compilers then have to support this (usually by maintaining a second "large" stack).
Yea, usually the stack ulimit is only a few KiB for non-root processes by default on linux.
It is easy enough to increase, but it does add friction to using the software as it violates the default stack size limit on most linux installs. Not even sure why stack ulimit is a thing anymore, who cares if the data is on the stack vs the heap?
It isn't a practical pattern for anything beyond the most trivial applications. Consider what this would look like if you tried to write a text editor, for instance - if a user types a new line of text, where is the memory for that allocated?
The problem is that regardless of the amount of confrontation it does not have an answer for any infinite run time event-loop based program, other than "allocate all of memory into a buffer at startup and implement your own memory manager inside that".
Which just punts the problem from a mature and tested runtime library to some code you just make up on the spot.
Heap was invented for a reason, and some tasks are naturally easier to model with it.
The problem is that once it's there, people start using it as the proverbial hammer, and everything looks like a nail even if it isn't.
Note though that ""allocate all of memory into a buffer at startup" is a lot more viable if you scope it not to the start of the app, but to the entrypoint of some code that needs to make a complicated calculation. It's actually not all that uncommon to need something heap-like to store temporary data as you compute - e.g. a list or map to cache intermediary results - but which shouldn't outlive the computation. Ada access types give you exactly that - declare them inside the top-level function that's your entrypoint, allocate as needed in nested functions as they get called, and know that it'll all be cleaned up once the top-level function returns.
That works for something where the events being handled are like "serve a web page" or "compile a C function". It doesn't work for a spreadsheet or word processor or a web browser.
I'm not convinced it even works very well for either of those cases. It's common in many applications to return the result of a computation as an object in memory, like an array or string of arbitrary length or a treelike structure. Without the ability to allocate memory which exists after a function exits, I'm not sure how you'd do that (short of solutions which create arbitrary limits, like writing to a fixed-size buffer).
Well, yes, but I'm trying to be generous to the PoV.
My preferred solution is definitely to use the GC. With some help if you want. You can GC the nursery each time around the event loop. You can create and destroy arenas.
I think one of the nice things about C is that since the language was not designed to abstract e.g.: heap is that it is really easy to replace manual memory management with GC or any other approach to manage memory, because most APIs expects to be called with `malloc()` when heap allocation is needed.
I think the only other language that has a similar property is Zig.
> Odin is a manual memory management based language. This means that Odin programmers must manage their own memory, allocations, and tracking. To aid with memory management, Odin has huge support for custom allocators, especially through the implicit context system.
Interesting that I was thinking of a language that combined Zig and Scala to allocate memory using implicits and this looks exactly what I was thinking.
Not that I actually think this is a good idea (I think the explicitly style of Zig is better), but it is an idea nonetheless.
A lot of our Odin procs take an allocator as a required argument specifically so they force you to choose one at the call site, because yes, often you want it to be explicit.
I am thinking now a language that had the default allocator be a GC, but you could change it to something else using implicits/context and this is enfoced everywhere (e.g.: there is no way to allocate memory "from magic", you need to call the allocator, even if it is from context).
At least thinking in Scala as a base syntax here, it would mean:
def foo(name: String)(implicit allocator: MemoryAllocator): StringBuf = {
val buf = allocator.new(100, String)
buf.append("Hello ")
buf.append(name)
buf
}
println(foo("John")) // implicit allocator from context, default to GC, no need to free
implicit val allocator: MemoryAllocator = allocators.NewAllocator()
println(foo("Alice")) // another implicit allocator, but now using the manual allocator instead of GC
defer allocator.free()
val explicitAllocator = allocators.NewArenaAllocator()
foo("Marie")(explicitAllocator) // explicit allocator only for this call
defer explicitAllocator.free()
I don't think there is any language right now that has this focus, since Odin seems to be focused in manual allocations.
>Try doing C with a garbage collector ... it's very liberating.
Doing that means that I lose some speed and I will have to wait for GC collection.
Then why shouldn't I use C# which is more productive and has libraries and frameworks that comes with batteries included that help me build functionality fast.
I thought that one of the main points of using C is speed.
Every time I've tried it, adding GC to a C program makes it faster. It will increase the RAM usage by a bit, and exactly when pauses happen is less predictable (unless you explicitly trigger a GC in e.g. your main loop), and possibly individual pauses larger, but overall program execution time -- throughput -- increases.
Malloc() and free() aren't free, and in particular free() does a lot more bookkeeping and takes more time than people realise.
I never liked that you have to choose between this and C++ though. C could use some automation, but that's C++ in "C with classes" mode. The sad thing is, you can't convince other people to use this mode, so all you have is either raw C interfaces which you have to wrap yourself, or C++ interfaces which require galaxy brain to fully grasp.
I remember growing really tired of "add member - add initializer - add finalizer - sweep and recheck finalizers" loop. Or calculating lifetime orders in your mind. If you ask which single word my mind associates with C, it will be "routine".
C++ would be amazing if its culture wasn't so obsessed with needless complexity. We had a local joke back then: every C++ programmer writes heaps of C++ code to pretend that the final page of code is not C++.
I completely agree with this sentiment. That's why I wrote Datoviz [1] almost entirely in C. I use C++ only when necessary, such as when relying on a C++ dependency or working with slightly more complex data structures. But I love C’s simplicity. Without OOP, architectural decisions become straightforward: what data should go in my structs, and what functions do I need? That’s it.
The most inconvenient aspect for me is manual memory management, but it’s not too bad as long as you’re not dealing with text or complex data structures.
I like the idea of using C++ as C. I began disliking OOP, inheritance and encapsulation, heavy usage of GoF patterns and even SOLID. They promise easy to understand, easy to follow, easy to maintain, easy to change, easy to extend code and a good productivity but the effect is contrary, most of the times.
I like functional programming and procedural programming. Fits better to how I think about code. Code is something that takes data and spits data. Code shouldn't be forced into emulating some real life concepts.
Agreed. C, Go, Python, and Lua are my go-to languages because of their simplicity. It's unfortunate, but in my opinion, most mainstream languages are needlessly complex.
In my experience, whether it's software architecture or programming language design, it's easy to make things complicated, but it takes vision and discipline to keep them simple.
> C, Go, Python, and Lua are my go-to languages because of their simplicity
One of these things is not like the others! Python's complexity has been increasing rapidly (e.g. walrus operator, match statement, increasingly baroque type hints) - has this put you off the language at all?
I started programming with Python years and years ago. My journey has led me to preferring low-level languages (currently in a Zig bout), and revisiting Python has been a disappointing experience. Not necessarily bad, but instead of focusing efforts where it really matters, such as eliminating GIL, they seem to be focused on adding new syntactic sugar.
Python's complexity has been increasing since Guido stepped away and it is off-putting. The language is still nowhere near as complex as Ruby or others, but I am keeping my eyes open for alternatives.
I agree with this sentiment. My first gig was telecom, and I wrote in a pascal like language called CHILL, but found out my forte was debugging and patching and ended up doing a fair amount of assembly code that would get applied to live systems. The decade plus I spent in medical devices, I used C and assembly.
The thing is, if you own all the code and actually understand what it is supposed to do, you can write safe code.
Variety is good. I got so used to working in pure C and older C++ that for a personal project I just started writing in C, until I realised that I don't have to consider other people and compatibility, so I had a lot of fun trying new things.
> A typical example I have encountered several times now is people using elaborate setups with std::string_view to avoid string copying, while exactly the same functionality could've been achieved by fewer code, using just a simple raw const char* pointer.
C++ can avoid string copies by passing `const string&` instead of by value. Presumably you're also passing around a subset of the string, and you're doing bounds and null checks, e.g.
string_view is just a char* + len; which is what you should be passing around anyway.
Funnily enough, the problem with string view is actually C api's, and this problem exists in C. Here's a perfect example: (I'm using fopen, but pretty much every C api has this problem).
FILE* open_file_from_substr(const char* start, int len)
{
return fopen(start);
}
void open_files()
{
const char* buf = "file1.txt file2.txt file3.txt";
for (int i = 0; i += 10; ++i) // my math might be off here, apologies
{
open_file_from_substr(buf + i, buf + i + 10); // nope.
}
}
> When I develop methods in pure C, I always enjoy that I can concentrate 100% on algorithmic aspects instead of architectural decisions which I only have to decide on because of the complexity of the language
I agree this is true when you develop _methods_, but I think this falls apart when you design programs. I find that you spend as much time thinking about memory management and pointer safety as you do algorithmic aspects, and not in a good way. Meanwhile, with C++, go and Rust, I think about lifetimes, ownership and data flow.
I started programming with C a long time ago, and even now, every few months, I dream of going back to those roots. It was so simple. You wrote code, you knew roughly which instructions it translated to, and there you went!
Then I try actually going through the motions of writing a production-grade application in C and I realise why I left it behind all those years ago. There's just so much stuff one has to do on one's own, with no support from the computer. So many things that one has to get just right for it to work across edge cases and in the face of adversarial users.
If I had to pick up a low-level language today, it'd likely be Ada. Similar to C, but with much more help from the compiler with all sorts of things.
When Ada was first announced, I rushed to read about it -- sounded good. But so far, never had access to it.
So, now, after a long time, Ada is starting to catch on???
When Ada was first announced, back then, my favorite language was PL/I, mostly on CP67/CMS, i.e., IBM's first effort at interactive computing with a virtual machine on an IBM 360 instruction set. Wrote a little code to illustrate digital Fourier calculations, digital filtering, and power spectral estimation (statistics from the book by Blackman and Tukey). Showed the work to a Navy guy at the JHU/APL and, thus, got "sole source" on a bid for some such software. Later wrote some more PL/I to have 'compatible' replacements for three of the routines in the IBM SSP (scientific subroutine package) -- converted 2 from O(n^2) to O(n log(n)) and the third got better numerical accuracy from some Ford and Fulkerson work. Then wrote some code for the first fleet scheduling at FedEx -- the BOD had been worried that the scheduling would be too difficult, some equity funding was at stake, and my code satisfied the BOD, opened the funding, and saved FedEx. Later wrote some code that saved a big part of IBM's AI software YES/L1. Gee, liked PL/I!
When I started on the FedEx code, was still at Georgetown (teaching computing in the business school and working in the computer center) and in my appartment. So, called the local IBM office and ordered the PL/I Reference, Program Guide, and Execution Logic manuals. Soon they arrived, for free, via a local IBM sales rep highly curious why someone would want those manuals -- sign of something big?
I recently started re-reading "Programming in Ada" by J.G.P. Barnes about the original Ada. In my opinion, it was not that good of a language. Plenty of ways to trigger undefined behavior.
Where C was clearly designed to be a practical language with feedback from implementing an operating system in C. Ada lacked that kind of practical experience. And it shows.
I don't know anything about modern day Ada, but I can see why it didn't catch on in the Unix world.
I recall watching a presentation about C++20. During the presentation, the presenter said there were about 163 undefined behaviors in the C language (note: I think it was C99) which implied there were many more in C++ since it’s a much more complex language. Unfortunately, I don’t have a link to that presentation.
You might have heard about the SPARK variant of Ada. I recall reading in an article many years ago that the original version of SPARK was based on Ada83 because it is a very safe language with a lot less undefined behaviors, which is key to trying to statically prove the correctness of a program.
I'm curious about this list, because it definitely doesn't seem that way these days. It'd be interesting to see how many of these are still possible now.
I didn't make a list, but let me give an example. Page 22 where variable declarations are introduced:
If a variable is declared and not given an initial value then great care must be taken not to use the undefined value of the variable until one has been properly given to it. If a program does use the undefined value in an uninitialised variable, its behaviour will be unpredictable; the program is said to be erronous.
> So, now, after a long time, Ada is starting to catch on???
Money and hardware requirements.
Finally there is a mature open source compiler, and our machines are light years beyond those beefy workstations required for Ada compilers in the 1980's.
> I started programming with C a long time ago, and even now, every few months, I dream of going back to those roots. It was so simple. You wrote code, you knew roughly which instructions it translated to, and there you went!
Related-- I'm curious what percentage of Rust newbies "fighting the borrow checker" is due to the compiler being insufficiently sophisticated vs. the newbie not realizing they're trying to get Rust to compile a memory error.
I certainly spent most (95%+?) of my "fighting the borrow checker" time writing code I would never try to write in C++. A simple example is strings: I'd spend a lot of time trying to get a &str to work instead of a String::clone, where in equivalent C++ code I'd never use std::string_view over std::string - not because it would be a memory error to do so in my code as it stood, but because it'd be nearly impossible to keep it memory safe with code reviews and C++'s limited static analysis tooling.
This was made all the worse by the fact that I frequently, eventually, succeeded in "winning". I would write unnecessary and unprofiled "micro-optimizations" that I was confident were safe and would remain safe in Rust, that I'd never dare try to maintain in C++.
Eventually I mellowed out and started .clone()ing when I would deep copy in C++. Thus ended my fight with the borrow checker.
Not everything can be proved at compile-time, so necessarily Rust is going to complain about things that you know can be done safely in that specific context.
For example some tree structures are famously PITA in Rust. Yes, possible, but PITA nonetheless.
If you come from C to Rust to basically have to rewire your brain. There are some corner cases that are wrong in Rust, but mostly you have to get used to a completely new way of thinking about object lifetimes and references to objects.
On x86-type machines, you still have a decent chance, because the instructions themselves are so complicated and high-level. It's not that C is close to the metal, it's that the metal has come up to nearly the level of C!
I wouldn't dare guess what a compiler does to a RISC target.
(But yes, this was back in the early-to-mid 2000s I think. Whether that is a long time ago I don't know.)
Another way of looking at it (although, I’m not sure if I believe this, haha)—it might be easy to guess what the the C compiler will spit out, for the proprietary bytecode known as “x86.” It is hard to guess what actual machine code (uops) it will be jitted to, when it is actually compiled by the x86 virtual machine.
I'd call it a while ago, but not a long time. Long time to me is more like 70s or 80s. I was born in 1996 so likely I'm biased: "before me=long time". It would be interesting to do a study on that. Give the words, request the years, correlate with birthyear, voila
Given how fast our field grows, you might want to consider anything beyond 13 years "a long time ago", since only a tenth of us were around back then[1].
On my circle of friends I was the PC guy, however got to enjoy quite a bit of the Amiga ecosystem, thanks to us being always being around each others, and some incursions into our demoscene attempts.
Also, COBOL and FORTRAN. FORTRAN is still being developed and one of the languages supported as first class citizen by MPI.
There's a big cloud of hype at the bleeding edge, but if you dare to look beyond that cloud, there are many boring and well matured technologies doing fine.
Writing user-facing applications in Swift and dropping to C and C++ when required seems to give the best of both worlds.
For me the main benefit of C and C++ is the availability of excellent and often irreplaceable libraries. With a little bridging work, these tend to just work with Swift.
Swift seems like an excellent language, but I can't shake the feeling that it's just not worth learning unless you're writing code on and for Apple's ecosystem.
My conclusion is that C is not a good basis for what Rust is trying to do. The kind of reliability Rust is trying to provide with almost no runtime overhead requires a much more complex language than C.
Indeed. I'm still not entirely sure why Rust was created when we have Ada, but if I had to guess it's mainly because Rust has slightly more advanced tricks for safe memory management, and to some degree because Rust has curly braces.
Ada doesn't attempt to statically exclude data races or aliasing bugs. Rust does. I guess you're calling that "slightly more advanced tricks for safe memory management", which sounds wildly inaccurate to me; those problems aren't usually considered "memory management" at all. Rust also has better error handling, a stronger static type system, a Turing-complete compile-time macro system, and much less verbose code.
def route = fn (request) {
if (request.method == GET ||
request.method == HEAD) do
locale = "en"
slash = if Str.ends_with?(request.url, "/") do "" else "/" end
path_html = "./pages#{request.url}#{slash}index.#{locale}.html"
if File.exists?(path_html) do
show_html(path_html, request.url)
else
path_md = "./pages#{request.url}#{slash}index.#{locale}.md"
if File.exists?(path_md) do
show_md(path_md, request.url)
else
path_md = "./pages#{request.url}.#{locale}.md"
if File.exists?(path_md) do
show_md(path_md, request.url)
end
end
end
end
}
Yeah, I'm not sure a lot of people read the article. This isn't really a back to basics, going back to C, forgoing complexity type of article, but instead it's about developing a new programming language called KC3 to make use of ideas he originally developed in Lisp.
C was my first language and I quickly wrote my first console apps and a small game with Allegro. It feels incredibly simple in some aspects. I wouldn’t want to go back though. The build tools and managing dependencies feels outdated, somehow there is always a problem somewhere. Includes and the macro system feels crude. It’s easy to invoke undefined behavior and only realizing later because a different compiler version or flag now optimizes differently. Zig is my new C, includes a C compiler and I can just import C headers and use it without wrapper. Comptime is awesome. Build tool, dependency management and testing included. Cross compilation is easy. Just looks like a modern version of C. If you can live with a language that is still in development I would strongly suggest to take a look.
Otherwise I use Go if a GC is acceptable and I want a simple language or Rust if I really need performance and safety.
Without reading the comments here, I read the blog entry first.
After I finished I was puzzled, "what is the author trying to communicate to the reader here?"
As near as I can determine, enough people weren't using the author's program/utility because it was written in a language that hasn't been blessed by the crowd? It is hinted at that there might be issues involving memory consumption.
The author does not write lessons learned or share statistics of user uptake after the re-write.
No new functionality was gained, presumably this exercise was done as practice reps because the author could do it and had time.
No argument was made that the author has seen the light and now only C from this point on.
I sometimes write C recreationally. The real problem I have with it is that it's overly laborious for the boring parts (e.g. spelling out inductive datatypes). If you imagine that a large amount of writing a compiler (or similar) in C amounts to juggling tagged unions (allocating, pattern matching over, etc.), it's very tiring to write the same boilerplate again and again. I've considered writing a generator to alleviate much of the tedium, but haven't bothered to do it yet. I've also considered developing C projects by appealing to an embeddable language for prototyping (like Python, Lua, Scheme, etc.), and then committing the implementation to C after I'm content with it (otherwise, the burden of implementation is simply too high).
It's difficult because I do believe there's an aesthetic appeal in doing certain one-off projects in C: compiled size, speed of compilation, the sense of accomplishment, etc. but a lot of it is just tedious grunt work.
If you want to do microcontroller/embedded, I think C it still the overall best choice, supported by vendors.
Rust and Ada are probably slowly catching up.
You can certainly do entirely absurd things in Perl.
But it is a lot easier / safer work with.
You get / can get a wealth of information when you
the wrong thing in Perl.
With C
segmentation fault is not always easy to pinpoint.
However the tooling for C, with sone if the IDEs
out there you can set breakpoints/ walk through
the code in a debugger, spot more errors during compile
time.
There is a debugger included with Perk but after trying
to use it a few times I have given up on it.
Give me C and Visual Studio when I need debugging.
On the positive side, shooting yourself in the foot with C
is a common occurrence.
I have never had a segmentation fault in Perl.
Nor have I had any problems managing the memory,
the garbage collector appears to work well.
(at least for my needs)
Except when they don’t. I’m debugging something right now that runs fine under debugging conditions and crashes with a segfault in real life. It’s not randomized memory (messed with that), it’s likely some race where the timing is changed by the debugger.
Ironically, AT&T's monopoly actually helped the adoption of Unix, but not in an exploitative way. In 1956, AT&T was subject to a consent decree by the US government, where AT&T was allowed to maintain its phone monopoly but was not allowed to expand its market to other sectors. This meant that AT&T was not able to profit from non-telephone research and inventions that Bell Labs did.
During Unix's early days, AT&T was still under this decree, meaning that it would not sell Unix like how competitors sold their operating systems. However, AT&T licensed Unix, including its source code, to universities for a nominal fee that covered the cost of media and distribution. UC Berkeley was one of the universities that purchased a Unix licenses, and researchers there started making additions to AT&T Unix which were distributed under the name Berkeley Software Distribution (this is where BSD came from). There is also a famous book known as The Lions' Book (https://en.wikipedia.org/wiki/A_Commentary_on_the_UNIX_Opera...) that those with access to a Unix license could read to study Unix. Bootleg copies of this book were widely circulated. The fact that university students, researchers, and professors could get access to an operating system (source code included) helped fuel the adoption of Unix, and by extension C.
When the Bell System was broken up in 1984, AT&T still retained Bell Labs and Unix. The breakup of the Bell System also meant that AT&T was no longer subject to the 1956 consent decree, and thus AT&T started marketing and selling Unix as a commercial product. Licensing fees skyrocketed, which led to an effort by BSD developers to replace AT&T code with open-source code, culminating with 4.3BSD Net/2, which is the ancestor of modern BSDs (FreeBSD, NetBSD, OpenBSD). The mid-1980s also saw the Minix and GNU projects. Finally, a certain undergraduate student named Linus Torvalds started work on his kernel in the early 1990s when he was frustrated with how Minix did not take full advantage of his Intel 386 hardware.
Had AT&T never been subject to the 1956 consent decree, it's likely that Unix might not have been widely adopted since AT&T probably wouldn't have granted generous licensing terms to universities.
What is your killer app?
What CL has to do with no one running it?
What problem you had with garbage collectors?
Why is C is the only option?
Are you sure all those RCEs because of VMs and containers and not because it all written in C?
"There are no security implications of running KC3 code" - are you sure?
Yeah, I think every programmer experiences the "I should write a language" moment when the solution to the problem is abstracted to be the language itself.
I love Common Lisp, but I would definitely think twice before implementing anything I want other people to use on their machines in it. A tiny C executable is pretty nimble in comparison to anything you'll get out of Common Lisp.
I understand, it isn't that bad though: a web app of mine with dozens of dependencies and all templates and static assets is 35MB with SBCL and core compression (that includes the compiler and debugger, useful to connect to a running app and exploring its state (or even hot reloading code)). I suppose that's in the ballpark of a growing Go application. LispWorks has a tree shaker that builds a hello world in 5MB.
Eh, you can just distribute a .lisp file as though it were a ruby or perl script. CL is faster than the usual scripting languages, but admittedly not as fast as C or Rust.
> Virtual machines still suck a lot of CPU and bandwidth for nothing but emulation. Containers in Linux with cgroups are still full of RCE (remote command execution) and priviledge escalation. New ones are discovered each year. The first report I got on those listed 10 or more RCE + PE (remote root on the machine). Remote root can also escape VMs probably also.
A proper virtual machine is extremely difficult to break out of (but it can still happen [1]). Containers are a lot easier to break out of. I virtual machines were more efficient in either CPU or RAM, I would want to use them more, but it's the worst of both.
For those who like C because of the simplicity, I can wholeheartedly recommend Go. It's replaced C as my go-to for personal CLI projects - while it is more complex than C, the core language features fit in my head pretty well. Add to that the excellent tooling, primitive OOP and clean syntax, and it's a damn good replacement.
I learned computer engineering bottom-up and top-down. Bottom-up as a teen reading books and manuals. Since no one I knew had an actual computer in 1979, like Ada Lovelace, I built the computers in my mind, LOL. Then in college I did some top-down too by writing C and then compiling to assembly and then compiling to machine code.
we also had to build a CPU from discrete bit-slice components and then program it. One of the most time intensive courses I took at CMU. Do computer engineers still have to do that?
I would certainly encourage all computer engineers, and perhaps even software engineers, to learn the "full stack".
But as to programming and C, I haven't done that in almost 30 years. It would be an interesting experiment to see how much of that skill if any I still possess.
Going from mid-90s assembly to full stack dev/sec/ops, getting back to just a simple Borland editor with C or assembly code sounds like a lovely dream.
Your brain works a certain way, but you're forced to evolve into the nightmare half-done complex stacks we run these days, and it's just not the same job any more.
I've tried, but never succeeded in doing that; the complexity eventually seeps in through the cracks.
C++'s stdlib contains a lot of convenient features, writing them myself and pretending they aren't there is very difficult.
Disabling exceptions is possible, but will come back to bite you the second you want to pull in external code.
You also lose some of the flexibility of C, unions become more complicated, struct offsets/C style polymorphism isn't even possible if I remember correctly.
> C++'s stdlib contains a lot of convenient features, writing them myself and pretending they aren't there is very difficult.
I've never understood the motivation behind writing something in C++, but avoiding the standard library. Sure, it's possible to do, but to me, they are inseparable. The basic data types and algorithms provided by the standard library are major reasons to choose the language. They are relatively lightweight and memory-efficient. They are easy to include and link into your program. They are well understood by other C++ programmers--no training required. Throughout my career, I've had to work in places where they had a "No Standard Library" rule, but that just meant they implemented their own, and in all cases the custom library was worse. (Also, none of the companies could articulate a reason for why they chose to re-implement the standard library poorly--It was always blamed on some graybeard who left the company decades ago.)
Choosing C++ without the standard library seems like going skiing, but deliberately using only one ski.
Modern C++ has goodies like consteval that are supremely useful for embedded work. STL and the rest of the stdlib on the other hand depends on heap and exceptions for error reporting which are generally a no go zone for resource constrained targets.
You can productively use C++ as C-with-classes (and templates, and namespaces, etc.) without depending on the library. That leaves you no worse off than rolling your own support code in plain C.
They're all YOLOing on the assumption that errors won't happen and doing the system programmer's favorite error handling scheme of calling abort(), then restarting a process. That doesn't fly on an embedded device. Nobody seriously checks errno after every library call without a separate means to indicate a failure happened in the first place.
> who can write a strategy game with thousands of units each having their own vision of the world, without having a garbage collector running like hell
Even in ecosystems where you don't have a way to opt-out of GC, there are strategies for dramatically minimizing the impact. Just because you have some aspects of the product rely on it should not be fatal unless we are being a bit hyperbolic about the purity of tech. If your allocation rate is <1 megabyte per minute and you are expressly forcing GC to run along the grain of the allocations (i.e., after each frame or sim tick), then it is exceedingly unlikely you will experience the canonical multiple-second, stop-the-world GC boogeyman.
Additionally, I don't think this is a fair take on how an RTS game would be engineered in practice. Each unit should be part of a big contiguous array in memory somewhere. The collection of garbage can be made largely irrelevant in the scenario of managing game state if you are willing to employ approximately the same approach you'd use in many other languages. Nothing stops you from newing up a 2 gig byte array and handing out spans to consumers just because you have GC turned on.
What I've discovered is that while it does regularize some of the syntax of C, the really noticeable thing about Zig is that it feels like C with all the stuff I (and everyone else) always end up building on my own built into the language: various allocators, error types, some basic safety guardrails, and so forth.
You can get clever with it if you want -- comptime is very, very powerful -- but it doesn't have strong opinions about how clever you should be. And as with C, you end up using most of the language most of the time.
I don't know if this is the actual criterion for feature inclusion and exclusion among the Zig devs, but it feels something like "Is this in C, or do C hackers regularly create this because C doesn't have it?" Allocators? Yes. Error unions? Yes. Pattern matching facilities? Not so much. ADTs? Uh, maybe really stupid ones? Generics, eh . . . sometimes people hack that together when it feels really necessary, but mostly they don't.
Something like this, it seems to me, results in features Zig has, features Zig will never have, and features that are enabled by comptime. And it's keeping the language small, elegant, and practical. I'm a big time C fan, and I love it.
I saw mentions of Zig here often, so I decided to look at the docs to see what features it has. I had to scroll through all the docs only to find myself disappointed by the fact that Zig doesn't help with memory management in any way and advices to use comments and careful coding instead. And what adds to the disappointment is that its plus/minus operators do not catch overflow. If I wanted undefined behaviour, I could just use C/C++ as no language can compete with them in this regard.
Zig is a much simpler language than Rust. I'm a big Rust fan, but Rust is not even close to a drop-in replacement for C. It has a steep learning curve, and often requires thinking about and architecting your program much differently from how you might if you were using C.
For a C programmer, learning and becoming productive in Zig should be a much easier proposition than doing the same for Rust. You're not going to get the same safety guarantees you'd get with Rust, but the world is full of trade offs, and this is just one of them.
For me, where linked lists, graphs and other structures are a common need, zig gives me slices and deferred frees.
Rust is double expensive in this case. You have to memorize the borrow checker and be responsible for all the potential undefined behavior with unsafe code.
But I am not a super human systems programmer. Perhaps if I was the calculus would change. But personally when I have to drop down below a GC language, it is pretty close to the hardware.
Zig simply solves more of my personal pain points... but if rust matures in ways that help those I'll consider it again.
Correct me if I am wrong, but Rust at least has a borrow checker while in C (and Zig) one has to do the borrow checking in their head. If you read a documentation for C libraries, some of them mention things like "caller must free this memory" and others don't specify anything and you have to go to the source code to find out who is responsible for freeing the memory.
The point is that any formal borrow checking will reject perfectly valid patterns because they are impossible to statically prove, and this comes up especially often with data structures like graphs. So you end up struggling against the compiler - either you just go unsafe (in which case you have to be even more careful than in C and Zig because Rust makes more optimization assumptions based on ownership which you're now responsible for), or else you use hacks like using indices instead of pointers to, basically, work around the borrow checker.
You are describing the deficiencies of Rust implementation of a borrow checker and disregarding general benefits of it. Obviously there could be some way to mark self-referencing objects.
C/Zig do not even allow to specify who is responsible for freeing the allocation, and every time you use a library you need to either search the docs or (more often) reverse-engineer the code. Probably this is why writing or checking C code is so slow. I end up adding annotations to function prototypes but nobody is going to validate them so they only serve as documentation.
Rust gives two reasons for the borrow checker, iterator invalidation and use after free.
As I have always bought into
Dennis Ritchie's loop programming concepts, iterator invalidating hasn't been a problem.
Zig has defer which makes it trivial to place next to allocation, and it is released when it goes out of scope.
As building a linked list, dealing with bit fields, ARM peripherals, etc...; all require disabling the Rust borrow checker rules, you don't benefit at all from them in those cases IMHO.
It is horses for courses, and the Rust project admits they chose a very specific horse.
C is what it is, and people who did assembly on a PDP7 probably know where a lot of that is from.
I personally prefer zig to c... but I will use c when it makes the task easier.
I did come back to C after some years. I liked it because it made me suffer but in the I prevailed so I felt good about it. Now that I am back at it I remember all the pain of trying to enforce abstractions at compile-time with the C preprocessor. I am now considering Zig or suicide.
So this is a journey where starting in ruby, going through an SICP phase, and then eventually compromising that it isn't viable. it kinda seems like C is just the personal compromise of trying to maintain nerdiness rather than any specific performance needs.
I think it's a pretty normal pattern I've seen (and been though) of learning-oriented development rather than thoughtful engineering.
But personally, AI coding has pushed me full circle back to ruby. Who wants to mentally interpret generated C code which could have optimisations and could also have fancy looking bugs. Why would anyone want to try disambiguating those when they could just read ruby like English?
> But personally, AI coding has pushed me full circle back to ruby.
This happened to me too. I’m using Python in a project right now purely because it’s easier for the AI to generate and easier for me to verify. AI coding saves me a lot of time, but the code is such low quality there’s no way I’d ever trust it to generate C.
It depends on what you need the code for. If it’s something mission critical, then using AI is likely going to take more time than it saves, but for a MVP or something where quality is less important than time to market, it’s a great time saver.
Also there’s often a spectrum of importance even within a project, eg maybe some internal tools aren’t so important vs a user facing thing. Complexity also varies: AI is pretty good at simple CRUD endpoints, and it’s a lot faster than me at writing HTML/CSS UI’s (ie the layout and styling, without the logic).
If you can isolate the AI code to code that doesn’t need to be high quality, and write the code that doesn’t yourself, it can be a big win. Or if you use AI for an MVP that will be incrementally replaced by higher quality code if the MVP succeeds, can be quite valuable since it allows you to test ideas quicker.
I personally find it to be a big win, even though I also spend a lot of time fighting the AI. But I wouldn’t want to build on top of AI code without cleaning it up myself.
There are also some tasks I’ve learned to just do myself: eg I do not let the AI decide my data model/database schema. Data is too important to leave it up to an AI to decide. Also outside of simple CRUD operations, it generates quite inefficient database querying so if it’s on a critical path, perhaps write the queries yourself.
Java was created to solve some frequent troubles you have in C-code.
I do not want to be rude, but C has some error-prone syntax: if you forget a *, you will be in trouble.
If you do 1 byte offset error on an array in the stack, you get erratic behavior, core dump if you are lucky.
Buffer overlflows also poses security risks.
try...catch was not present on C, and it is one of the most powerful addition of C++ for code structuring.
Thread management/async programming without support from the language (which is fine, but if you see Erlang or Java, they have far more support for thread monitors).
Said that, there are very high quality library in C (pthreads, memory management and protection, lib-eventio etc) which can overcome most of its limit but... it is still error-prone
Writing code in C is very unpleasant, verbose and repetative. For example, if I want to have a structure in C and have a way to print its contents, or free its memory and memory of nested structures, or clone it recursively, it is very difficult to make automatically. I found only two options: either write complicated macros to define the structure and functions (feels like writing a C++ compiler from scratch), or define structure in Python and generate the C code from it.
I looked at C++, but it seems that despite being more feature-rich, it also cannot auto-generate functions/methods for working with structures?
Also returning errors with dynamically allocated strings (and freeing them) makes functions bloated.
Also Gnome infrastructure (GObject, GTK and friends) requires writing so much code that I feel sorry for people writing Gnome.
Also, how do you install dependencies in C? How do you lock a specific version (or range of versions) of a dependency with specific build options, for example?
There's no amount of simplification that'll free you from the need to manually check and handle/propagate errors or deallocate resources in C. You might be able to reduce the amount of code you need to write, but it's still several lines of C versus one line of something else - and if you make a mistake or even forget something in those several lines, things will often compile and run fine, but you'll have memory leaks or sporadic crashes.
I started writing C code again because i started using gstreamer and needed to write a modem plugin for ModemManager. This meant i had to get familiar with glib. Even thought the learning curve has been steep it been worth and I am really enjoying writing in C again, even going so far as to refactor some existing code of mine use glib and make it GObject based. The end goal here is to use gir to use my code from scriptable languages like python.
Tools should enable creativity and problem-solving, not become problems themselves. The best languages fade into the background, becoming almost invisible as you express your solution. When the language constantly demands center stage, something has gone fundamentally wrong with its design philosophy.
> GLib is a general-purpose, portable utility library, which provides many useful data types, macros, type conversions, string utilities, file utilities, a mainloop abstraction, and so on.
Glib is c batteries included library I really like. Does anyone have any others they prefer?
> Garbage collectors suck, and all my Common Lisp projects have very limited applications just because of the garbage collector.
If only a very tiny fraction of the resources effort, research, time, money etc of all ML/AI funds are directed for the best design of high performance GC, it will make a software world a much better place. The fact that we have a very few books dedicated on GC design and thousands of books now dedicated on AI/ML, it is quite telling.
For real-world example and analogy, automotive industry dedicated their resources on the best design of high performance automatic transmission and now it has a faster auto transmission than manual for rally and racing. For normal driving auto is what the default and available now, most of the cars do not sell in manual transmission version.
> Linux is written in C, OpenBSD is written in C, GTK+ is object-oriented pure C, GNOME is written in C. Most of the Linux desktop apps are actually written in plain old C. So why try harder ? I know C
C is the lingua-franca of all other programming languages including Python, Julia, Rust, etc. Period.
D language has already bite the bullet and made C built-in by natively supporting it. Genius.
D language also has GC by default for more sane and intuitive programming, it's your call. It also one of the fastest compilation time and execution time languages in existence.
From the KC3 language website, "KC3 is a programming language with meta-programmation and a graph database embedded into the language."
Why you want to have a graph database embedded into the language? Just support associative array built-in since it has been proven to be the basis of all common data representations of spreadsheet, SQL, NoSQL, matrices, Graph database, etc.
[1] Associative Array Model of SQL, NoSQL, and NewSQL Databases:
love this sentiment. i tried a lot of languages after C, and am also back to C. it does force me to pencil out all kinds of stuff rather than relying on third party code or compiler error help, but i find that delightful. good upfront coding on a piece of paper, slowly (compared to other langs) implementing it, and diligently testing because you know its very needed in C. it forces you to do things right, and if you do you get an ultimate reward. blazin fast , well behaved code.
yes, my life is still full of segfaults as many segfaults as ignorance and impatience. so its delightful because it helps me overcome those 2 things, slowly and steadily =)
I like Nim- compiles to C so you get similarly close to the instructions and you can use a lot of high level features if you want to, but you can also stay close to the metal.
- Get the `cdecl` tool to build intuition about function signatures. What does "int( * ( *foo)(void))[3]" mean?
- Write it yourself.
- Be disciplined. Develop good hygiene with compiler flags, memory/address checks, and even fuzzing.
- Read good source code such as the linux kernel. This is an amusing header file from the git source code that defines some banned functions. This is wisdom if you choose to follow it: https://github.com/git/git/blob/master/banned.h
- Push the language to its limits. Play with memory and data structures. Inspect everything. This book "Data-Oriented Design" by Richard Fabien is a great to explore as well. It's about organizing your data for efficient processing.
See my previous comment, anyone learning C from K&R should not be writing C period.
That book is legendary only in the magnitude of financial damages it (directly and indirectly) enabled, caused by people who read it and thought that they could now write C.
I still think in the right tool for the job. Trying to write some web application in C will drive me mad. Also will trying to write some low level stuff in Java.
"and all was bounds-checked at memory cost but the results were awesome. Defensive programming all the way : all bugs are reduced to zero right from the start."
a bit LOL, isn't it?
also the part about terraform, ansible and the other stuff.
In the abstract, the simplicity of C has a definite appeal. However, pragmatically, the sense that I am mowing the lawn with a pair of scissors gets tiring quickly.
Linux/UNIX distributions are essentially C development environments, and their package managers are basically C language package managers… so, you needn’t do everything yourself, just grab the source packages.
> It was supposed to be a short mission I thought I could learn Common Lisp in ten days and hack a quick server management protocol. I ended up writing throw-away Common Lisp code that generated C for a fully-fledged ASN.1 parser and query system for a custom Common Lisp to C SNMP server.
I believe it. And I'd love to see it and hack on it, if it were open source.
This whole kc3 thing looks pretty interesting to me. I agree with the premise. It's really just another super-C that's not C++, but that's a pretty good idea for a lot of things because the C ABI is just so omnipresent.
About a year ago, I had gotten fed up with what felt like overlyb strict context requirements basically giving me to abandon years of work and thought I'd try my hand at C++ again.
I wanted to do this on Linux, because I my main laptop is a Linux machine after my children confiscated my Windows laptop to play Minecraft with the only decent GPU in the house.
And I just couldn't get past the tooling. I could not get through to anything that felt like a build setup that I'd be able to replicate in my own.
On Windows, using Visual Studio, it's not that bad. It's a little annoying compared to a .NET project, and there are a lot more settings to worry about, but at the end of the day VS makes the two but very different from each other.
I actually didn't understand that until I tried to write C++ on Linux. I thought C++ on Windows was worlds different than C#. But now I've seen the light.
I honestly don't know how people do development with on Linux. Make, Cmake, all of that stuff, is so bad.
IDK, maybe someone will come along and tell me, "oh, no, do this and you'll have no problems". I hope so. But without that, what a disgusting waste of time C and C++ is on Linux.
I find make, cmake, and the other stuff annoying also. For personal stuff I just use a build.sh file. For debugging I use gf2, which is a gdb frontend. Hopefully raddebugger gets ported to linux soon. One nice tool I like on linux for prototyping is the tiny c compiler, because it compiles 7x faster than gcc or clang. It is also much faster than the visual studio compiler. I remember trying to get the tiny c compiler to work on windows; it can compile things, but I couldn't get it to generate pdb files for debug info.
Wow, I'm surprised you were able to make out what I was saying through all those drunk smartphone keyboard typos (it's the keyboard that was drunk, not me, no sir).
I just don't understand why people want to live like that in C++ land. It almost feels like masochism. Especially considering VC++ basically "just works" in comparison. Why do Linux users hate DevEx so much?
> Defensive programming all the way : all bugs are reduced to zero right from the start
Has it been fuzzed? Have you had someone who is very good at finding bugs in C code look at it carefully? It is understandable if the answer to one or both is "no". But we should be careful about the claims we make about code.
I'm kinda in the opposite camp. After doing a bunch of VB in my tweens and teens, I learned Java, C, and C++ in college, settling on mostly C for personal and professional projects. I became a core developer of Xfce and worked on that for 5 years.
Then I moved into backend development, where I was doing all Java, Scala, and Python. It was... dare I say... easy! Sure, these kinds of languages bring with them other problems, but I loved batteries-included standard libraries, build systems that could automatically fetch dependencies -- and oh my, such huge communities with open-source libraries for nearly anything I could imagine needing. Even if most of the build systems (maven, sbt, gradle, pip, etc.) have lots of rough edges, at least they exist.
Fast forward 12 years, and I find myself getting back in to Xfce. Ugh. C is such a pain in the ass. I keep reinventing wheels, because even if there's a third-party library, most of the time it's not packaged on many of the distros/OSes our users use. Memory leaks, NULL pointer dereferences, use-after-free, data races, terrible concurrency primitives, no tuples, no generics, primitive type system... I hate it.
I've been using Rust for other projects, and despite it being an objectively more difficult language to learn and use, I'm still much more productive in Rust than in C.
I think Rust is harder to learn, but once you grok it, I don't think it's harder to use, or at least to use correctly. It's hard to write correct C because the standard tooling doesn't give you much help beyond `-Wall`. Rust's normal error messages are delightfully helpful. For example, I just wrote some bad code and got:
I even had to cheat a little to get that far, because my editor used rust-analyzer to flag the error before I had the chance to build the code.Also, I highly recommend getting into the habit of running `cargo clippy` regularly. It's a wonderful tool for catching non-idiomatic code. I learned a lot from its suggestions on how I could improve my work.
> I think Rust is harder to learn, but once you grok it, I don't think it's harder to use, or at least to use correctly. It's hard to write correct C because the standard tooling doesn't give you much help beyond `-Wall`.
When I say Rust is harder to use (even after learning it decently well), what I mean is that it's still easier to write a pile of C code and get it to compile than it is to write a pile of Rust code and get it to compile.
The important difference is that the easier-written C code will have a bunch of bugs in it than the Rust code will. I think that's what I mean when I say Rust is harder to use, but I'm more productive in it: I have to do so much less debugging when writing Rust, and writing and debugging C code is more difficult and takes up more time than writing the Rust code (and doing whatever less debugging is necessary there).
> Also, I highly recommend getting into the habit of running `cargo clippy` regularly. It's a wonderful tool for catching non-idiomatic code.
That's a great tip, and I usually forget to do so. On a couple of my personal projects, I have a CI step that fails the build if there are any clippy messages, but I don't use it for most of my personal projects. I do have a `cargo fmt --check` in my pre-commit hooks, but I should add clippy to that as well.
If you're using VS Code then you can add `"rust-analyzer.check.command": "clippy"` to your `settings.json`. I assume there's a similar setting for rust-analyzer in other editors.
Neovim:
You might want to reconsider use of rust-analyzer, it isn't safe to use on code you haven't written yourself.
https://rust-analyzer.github.io/book/security.html
> it isn't safe to use on code you haven't written yourself
Neither is cargo (nor npm, nor any other package manager, for that matter).
I'm not sure what value being that paranoid is buying you in the long run.
Package managers are for running other people's code, I would not expect the same of static analysis tools, especially since they are of use while auditing other people's code before building/running it.
Cargo's threat model here is identical to that of rust analyser. If you trust your dependency tree sufficiently to run `cargo build`, then you trust it sufficiently to run rust analyser.
That's a fair distinction. Basically, it's easier to write C that compiles than Rust that compiles, but it's harder to write correct C than correct Rust.
Regarding Clippy, you can also crank it up with `cargo clippy -- -Wclippy::pedantic`. Some of the advice at that level gets a little suspect. Don't just blindly follow it. It offers some nice suggestions though, like:
that you don't get by default.You can also add #![warn(clippy::all, clippy::pedantic)] to your main.rs/lib.rs file to get those lints project-wide.
The truly pedantic setting here would by complaining about the magic number.
Why 1_000_000_000_000, what does that number mean.
It is for free to:
since the compiler will just inline it.The readability problem was never the lack of separators, since that number might be the wrong number regardless.
Getting something to compile is never the end-goal. It takes 0 effort to get Python compile.
It can feel like a real milestone though
> it's still easier to write a pile of C code and get it to compile than it is to write a pile of Rust code and get it to compile.
As someone who is more familiar with Rust than C: only if you grok the C build system(s). For me, getting C to build at all (esp. if I want to split it up into multiple files or use any kind of external library) is much more difficult than doing the same in Rust.
I’d add that the Rust code and C code will probably have the same number of bugs. The C code will likely have some vulnerabilities on top of those.
Rust doesn’t magically make the vast majority of bugs go away. Most of bugs are entirely portable!
Rust allows to provide more information about types (generic types, pointer usage) and checks it, while in C you have to rely on doc comments and checking the code manually. Or am I wrong and C allows to specify pointer nullability, pointer ownership and array bounds?
None of those things feature in any problem I deal with on a daily basis, whatever language I use.
So for example today I dealt with a synchronization issue. This turned out to not be a code bug but a human misunderstanding of a protocol specification saga, which was not possible to code into a type system of any sort. The day before was a constraint network specification error. In both cases the code was entirely irrelevant to the problem.
Literally all I deal with are human problems.
My point is Rust doesn't help with these at all, however clever you get. It is no different to C, but C will give you a superset of vulnerabilities on top of that.
Fundamentally Rust solves no problems I have. Because the problems that matter are human ones. We are too obsessed with the microscopic problems of programming languages and type systems and not concentrating on making quality software which is far more than just "Rust makes all my problems go away" because it doesn't. It kills a small class of problems which aren't relevant to a lot of domains.
(incidentally the problems above are implemented in a subset of c++)
> So for example today I dealt with a synchronization issue. This turned out to not be a code bug but a human misunderstanding of a protocol specification saga, which was not possible to code into a type system of any sort.
Maybe not in a reasonable language no, but there are advances in type systems that are making ever larger classes of behaviours encodable into types. For example, algebraic effects (can this function throw, call a remote service etc)
https://koka-lang.github.io/koka/doc/index.html
linear types (this method must be called only once etc)
https://ghc.gitlab.haskell.org/ghc/doc/users_guide/exts/line...
dependent typing (f(1) returning a different type from f(2), verifiable at compile time),
https://fstar-lang.org/index.html
Some of these features will eventually make it to “normal” PL. For example, Scala now has dependent types,
https://dotty.epfl.ch/docs/reference/new-types/match-types.h...
and Java can support linear type checking,
https://checkerframework.org/manual/#must-call-checker
> None of those things feature in any problem I deal with on a daily basis, whatever language I use.'
I run into those things nearly daily, so... ok then.
Do not feel bad friend, It's not you.. there is definitely inconsistencies here. Maybe they work on absolutely pristine perfect codebases.
Vulnerabilities are bugs, so the C code will have more bugs than the Rust program.
You might say that the C and Rust code will have the same number of logic errors, but I'm not convinced that's the case either. Sure, if you just directly translate the C to Rust, maybe. But if you rewrite the C program in Rust while making good use of Rust's type system, it's likely you'll have fewer logic errors in the Rust code as well.
Rust has other nice features that will help avoid bugs you might write in a C program, like most Result-returning functions in the stdlib being marked #[must_use], or match expressions being exhaustive, to name a couple things.
> most Result-returning functions in the stdlib being marked #[must_use]
Actually it's a bit cleverer than that, and some people might benefit from knowing this. The Result type itself is marked #[must_use]. If you're writing a Goat library and you are confident that just discarding a Goat is almost always a mistake regardless of the context in which they got a Goat you too should mark your Goat type #[must_use = "this `Goat` should be handled properly according to the Holy Laws of the Amazing Goat God"] and now everybody is required to do that or explicitly opt out even for their own Goat code.
Obviously don't do this for types which you can imagine reasonable people might actually discard, only the ones where every discard is a weird special case.
Types I like in Rust which help you avoid writing errors the compiler itself couldn't possibly catch:
Duration - wait are these timeouts in seconds or milliseconds? It's different on Windows? What does zero mean, forever or instant ?
std::cmp::Ordering - this Doodad is Less than the other one
OwnedFd - it's "just" a file descriptor, in C this would be an integer, except, this is always a file descriptor, it can't be "oops, we didn't open a file" or the count of lines, or anything else, we can't Add these together because that's nonsense, they're not really integers at all.
I'd say whether Rust helps you reduce bugs depends on how good you are at creating abstractions and/or encoding properties in the type system.
Most bugs are way above that level of abstraction and thought.
[Citation needed].
Only about 70% of high severity bugs[1]. Quite a negligible amount.
[1]: https://www.chromium.org/Home/chromium-security/memory-safet...
This is objectively nonsense
Interesting use of the word 'objectively' there.
How can it not be true? One of the primary features of the rust compiler is enforcing memory safety at compile-time. C doesn't have anything like that. There are an entire class of bugs that are impossible to implement in rust.
>> One catches far more issues than another
> akshually they have same amount of bugs
You literally segfault by dereferencing invalid pointer.
I agree.
But: If you are LUCKY! The worse case is if you DON'T segfault by dereferencing invalid pointer.
Not really.
I have written a lot of C and Rust. The notion that identical projects written in both languages would have identical numbers of (or even severity of) bugs is laughable on its face.
I mean, literally just not being able to deref a NULL pointer by itself is enormous.
You can dereference null pointers in Rust, it's also easy to make null pointers. The thing is I'm not surprised you didn't realise this is possible because you would never actually do this in Rust since it's obviously a bad idea.
If we instead use references they're nicer to work with, don't need unsafe, perform the same, and are never null. So obviously that's what real Rust programmers almost always do.Of course it’s possible. I am well aware it’s possible. You can also just call into a C library that hands you a null pointer. Further, it’s wholly possible to invalidate every single one of Rust’s safety guarantees if you try hard enough.
In practice I have never encountered a null pointer in Rust. Most Rust programmers have never encountered a null pointer. For most non-FFI practical purposes you can act as if they don’t exist.
Agree, Rust is quite hard to learn, but now that I know it I have a hard time writing anything else. It really gives you the best of a lot of worlds.
Granted I can still crank out a python program faster, that kinda works but god forbid you need to scale it or use any sort of concurrency at all.
Go?
> It's hard to write correct C because the standard tooling doesn't give you much help beyond `-Wall`
I won't disagree that correct C is harder to write, but it's not 2005 anymore and standard tooling gives you access to things like asan, msan, ubsan, tsan, clang-tidy...
C-with-sanitizers is miles ahead of what C used to be, but just a couple weeks ago I ran into a dangling pointer bug that ASan doesn't catch. (Sidenote 5 here: https://jacko.io/smart_pointers.html) It seems like one of the big downsides of sanitizers is that they can't instrument other people's code, in this case the C standard library.
> Also, I highly recommend getting into the habit of running `cargo clippy` regularly.
You can also have that hooked up to the editor, just like `cargo check` errors. I find this to be quite useful, because i hace a hard time getting into habits, especially for thing that i'm not forced to do in some way. It's important that those Clippy lints are shown as soft warnings instead of hard errors though, as otherwise they'd be too distracting at times.
* Rust errors can be equally unhelpful. Also, the error you posted is hands down awful. It doesn't tell you what went wrong, and it's excessively naive to rely on compiler to offer a correct fix in all but the most trivial cases. When errors happen, it's a consequence of an impasse, a logical contradiction: two mutually exclusive arguments have been made: a file was assumed to exist, but was also assumed not to exist -- this is what's at the core of the error. The idiotic error that Rust compiler gave you doesn't say what were the assumptions, it just, essentially, tells you "here's the error, deal with it".
* In Rust, you will have to deal with a lot of unnecessary errors. The language is designed to make its users create a host of auxiliary entities: results, options, futures, tasks and so on. Instead of dealing with the "interesting" domain objects, the user of the language is mired in the "intricate interplay" between objects she doesn't care about. This is, in general, a woe of languages with extensive type systems, but in Rust it's a woe on a whole new level. Every program becomes a Sisyphean struggle to wrangle through all those unnecessary objects to finally get to write the actual code. Interestingly though, there's a tendency in a lot of programmers to like solving these useless problems instead of dealing with the objectives of their program (often because those objectives are boring or because programmers don't understand them, or because they have no influence over them).
I don't follow your first point—the compiler is pointing out exactly what the problem is (the argument has the incorrect type) and then telling you what you likely wanted to do (borrow the String). What would you see as a more helpful error message in this case?
The compiler says "expected X, but found Y". I don't know how to interpret this: is the type of the thing underlined with "^^^" X or Y? "Expected" and "found" are just like "up" and "down" in space: they are meaningless if you don't know what the compiler expects (and why should it?).
What it needs to say is something along the lines of "a function f is defined with type X, but is given an argument of type Y": maybe the function should be defined differently, maybe the argument needs to change -- it's up to the programmer to decide.
I dunno, I feel like if you've used a compiler regularly, "expected X, but found Y" is a pretty common idiom/shorthand that people understand. Your wordier version of that feels unnecessary to me.
C is a low level language and deals with things close to the metal. It's probably not fun to write a large business app in barebones C but you having control over low level things makes other things possible and very fast too. Depending on the type of problem you have use the appropriate and favorite language.
You cannot be serious right now.
I don't see any way that the use if expected and found can be ambiguous for a type conflict.
I buy a fruit mixer from Amazon.com ; I send it back along with a note: expected a 230VAC mixer, found a 110VAC mixer.
Since it's underlining code you wrote, it must be "found" that is highlighted, not "expected". Much like up and down, gravity exists to ground all of us in the same direction.
I'm over here with TTS: Underlining in a terminal rarely translates to audio. It isn't the only consideration that needs to be made, when making things clear.
I hadn’t considered this use case, thank you for pointing it out!
except nobody has to cater to every worst case scenario
I am sure the Rust community would welcome feedback from blind developers about their error messages. Accessibility is a good thing.
One thing that could help here is that the compiler is able to offer output in JSON, allowing you to format the messages however you'd like: https://doc.rust-lang.org/rustc/json.html
I'm not aware of an existing tool to produce blind-friendly output, but this would at least be a part of that!
So every disabled programmer is now a "worst case scenario"?
Or in short, Rust isn't for disabled programmers?
I tend to agree with this.
The code tend to be loaded with primitives that express ownership semantics or error handling. Every time something changes (for instance, you want not just read but also modify values referenced by the iterator) you have to change code in many places (you will have to invoke 'as_mut' explicitly even if you're accessing your iterator through mutable ref). This could be attributed (partially) to the lack of function overload. People believe that overload is often abused so it shouldn't be present in the "modern" language. But in languages like C++ overload also helps with const correctness and move semantics. In C++ I don't have to invoke 'as_mut' to modify value referenced by the non-const iterator because dereferencing operator has const and non-const overloads.
Async Rust is on another level of complexity compared to anything I used. The lifetimes are often necessary and everything is warpped into mutliple layers, everything is Arc<Mutex<Box<*>>>.
> Rust errors can be equally unhelpful. Also, the error you posted is hands down awful. It doesn't tell you what went wrong, and it's excessively naive to rely on compiler to offer a correct fix in all but the most trivial cases
What? It tells the user exactly what's wrong
> Every program becomes a Sisyphean struggle to wrangle through all those unnecessary objects to finally get to write the actual code
That is the cost of non-nullable types and correctness. You still have to do the Sisyphean struggle in other programming languages, but without hints from the compiler.
The biggest problem with C is that doesn't even have enough features to help you build the features and abstractions you need and want.
For example with C++ the language offers enough functionality that you can create abstractions at any level, from low level bit manipulation to high level features such as automatic memory management, high level data objects etc.
With C you can never escape the low level details. Cursed to crawl.
Just FYI.
Back in 1994/95, I wrote an API, in C, that was a communication interface. We had to use C, because it was the only language that had binary/link compatibility between compilers (the ones that we used).
We designed what I call "false object pattern." It used a C struct to simulate a dynamic object, complete with function pointers (that could be replaced, in implementation), and that gave us a "sorta/kinda" vtable.
Worked a charm. They were still using it, 25 years later.
That said, I don't really miss working at that level. I have been writing almost exclusively in Swift, since 2014.
> We designed what I call "false object pattern." It used a C struct to simulate a dynamic object, complete with function pointers (that could be replaced, in implementation), and that gave us a "sorta/kinda" vtable.
You were not alone in this. It is the basis of glib's GObject which is at the bottom of the stack for all of GTK and GNOME.
sadly glib doesn't have error handling for memory errors,it will just crash. otherwise it can be widely used as a oop c
I got the idea from Apple's QuickDraw GX. It used a similar pattern.
Sure, that's a pretty common pattern in use in C to this day. It's a useful pattern, but it's still all manual. Forget to fill in a function pointer in a struct? Crash. At least with C++ it will fail to compile if you don't implement a method that you have to implement.
At least in C it's only plain old function pointers.
You don't have to think about exceptions, overloaded operators, copy constructors, move semantics etc.
You still need to think about error handling, and it's not standardized because everyone else will also have to think about it ad hoc.
You'll also still need to think about when to copy and move ownership, only without a type system to help you tell which is which, and good luck ensuring resources are disposed correctly (and only once) when you can't even represent scoped objects. `goto` is still the best way to deal with destructors, and it still takes a lot of boilerplate.
But you have a choice, you don't have to implement all of C++/Rust.
The beauty of C is that it allows you to pick your level of complexity.
You don't have to use the entirety of C++, either.
No, but that tends to gradually happen all by itself from my experience, even in solo projects. Because it feels stupid writing something yourself when you have access to stdlib. Also, I feel rolling your own stuff in C++ is a major pita in comparison to C, too many rules to memorize for me.
> still all manual
That pretty much is the definition of C. It was designed to be a system language, that was "one step beyond" (my own Madness reference) the assembler.
It's a dangerous tool, and should be wielded by experts. The same goes for all kinds of similar stuff.
And "experts" is kind of a darwinian thing. There's not really that many folks that can successfully wield it in large quantities (a certain cranky Finn comes to mind). There's a ton of folks that are surfing the Dunning-Kruger wave, that think they are up to it, but they tend to face-plant.
For myself, I learned to do it right, but it was difficult, stressful work, and I don't miss it. Working in Swift is a joy. I don't have to worry about things like deallocations, and whatnot.
I've never met an expert who writes c well all of the time, so I have been led to believe it's simply not possible. Folks with 20+ years of experience will still regularly write a use after free but which is not caught until it hits production. I'm not sure I've actually seen C code free of concurrency bugs. If it's possible it takes years of battle hardening with no feature work. Humans are prone to make mistakes and we need to use as many tools as possible to help us avoid making those mistakes. They are recently looking into introducing thread annotations for locks in the Linux kernel and I expect them to find lots of real bugs lurking in otherwise well maintained subsystems. I've rewritten swaths of code into rust and always seen it find a bug in the original code's design that we were lucky to never run into.
haha every segfault puts me back down into the valley of despair xD. thats totally ok!
higher level languages often make it so you can focus on the language mostly. C and assembly, i think, you need a lot of context. what is running the code, and how.
Of course you can. It's quite the opposite actually. The downside is that in C you have to code a bunch of abstractions _yourself_. See how large projects like the Linux kernel make extensive use of macros to implement an object system.
so confused by this one. C is the most free to build systems and abstractions. it doesnt lock you down into any paradigm so you can build your own....
you can abstract away perfectly fine low level details and use your own high level constructs to build what you want in a memory safe way...
high level daya objects?
sure it doesnt have garbage collection, so it motivates you not to leave garbage laying around for some slow background process to collect.
actually you can build this in C you just do not want to...
you can give all your objects reference counters and build abstractions to use that, you can implement 'smart pointers' if u want, and have threads do garbage collection if u want... why not? what exactly is stopping you?
maybe its less convenient to go that route. but impossible? nope.
connect GDB to your c++ program and see how it works... its not like c++ suddenly doesnt become machine code. and c translates perfectly fine to machine code...
Thank you for your work on XFCE. It's been the best WM/UI for me for over a decade.
Have you looked at Zig? It is often termed a modern C where Rust is the modern C++. Seems like a good fit.
i never really understand why these get compared. i wouldn't expect that much overlap in the audiences
zig seems like someone wanted something between C and "the good parts" of C++, with the generations of cruft scrubbed out
rust seems like someone wanted a haskell-flavoured replacement for C++, and memory-safety
i would expect "zig for C++" to look more like D or Carbon than rust. and i'd expect "rust for C" to have memory safety and regions, and probably steal a few ocaml features
The single best feature (and I would say the _core_ feature separating it from C) that C++ has to offer is RAII and zig does not have that. So I don’t know which good parts of C++ they kept. Zig is more of its own thing, and they take from wherever they like, not just C++.
> So I don’t know which good parts of C++ they kept.
comptime is a better version of C++ templates.
I really wish that were true but it isn’t. Modern C++ templates/constexpr are much more powerful and expressive than any Zig comptime equivalent.
The power and expressiveness of the C++ compile-time capabilities are the one thing I strongly miss when using other languages. The amount of safety and conciseness those features enable makes not having them feel like a giant step backward. Honestly, if another systems language had something of similar capability I’d consider switching.
I have written a lot of Zig comptime code and ended up finding the opposite. In C++ I find I have to bend over backward to get what I want done, often resulting in insane compile times. I've used metaprogramming libraries like Boost Hana before to have some more ergonomics, but even that I would consider inferior to comptime.
Out of curiosity, do you happen to have any examples of what you describe, where C++ is more powerful and expressive than Zig?
If it looks anything like what I read in "Modern C++ Design" 20+ years ago then I'll pass. That book made me realize the language wasn't for me anymore.
It looks nothing like C++ decades ago, it is effectively a completely different language. I found C++ unusable before C++11, and even C++11 feels archaic these days. Idiomatic C++20 and later is almost a decent language.
"Modern C++ Design" was the first book that highlighted template metaprogramming and showed you ways to use it as a Turing complete programming language in itself. Even a decade+ after it was printed my friend said it was recommended reading for employees within Google so I gave him mine - but I'm not sure about these days.
Talking about how much you can do with C++ templates made me think of that.
It’s not like that anymore.
Zig has defer, which is arguably way simpler. Is there something RAII can do that defer can't?
Not everyone likes RAII by itself. Allocating and deallocating things one at a time is not always efficient. That is not the only way to use RAII but it's the most prevalent way.
defer can't emulate the destructors of objects that outlives their lexical scope. Return values and heap objects are examples of these since they outlive the function they were created in. defer only supports enqueuing actions at lexical boundaries.
If you destroy an object that outlives the lexical scope it was created in, then you have to clean up manually.
Defer can be omitted
I would say OCaml more than Haskell, but yes.
I have, and I do find Zig impressive, but it doesn't go far enough for me. I don't want a "better C", I want a better systems language that can also scale up for other uses.
I like strong, featureful type systems and functional programming; Zig doesn't really fit the bill for me there. Rust is missing a few things I want (like higher-kinded types; GATs don't go far enough for me), but it's incredible how they've managed to build so many zero- and low-cost abstractions and make Rust feel like quite a high-level language sometimes.
Rust is not a modern C++, their core models are pretty different. Both Rust and C++ can do things the other can’t do. C++ is a more focused on low-level hyper-optimized systems programming, Rust is a bit higher level and has stronger guardrails but with performance closer to a classic systems language.
I do think Zig is a worthy successor to C and isn’t trying to be C++. I programmed in C for a long time and Zig has a long list of sensible features I wish C had back then. If C had been like Zig I might never have left C.
What do you consider the difference in their core models.
Rust and C++ both use RAII, both have a strong emphasis on type safety, Rust just takes that to the extreme.
I would like to even hope both believe in 0 cost abstractions, which contrary to popular belief isn't no cost, but no cost over doing the same thing yourself.
In many cases it's not even 0 cost, it's negative cost since using declarative programming can allow the compiler to optimise in ways you don't know about.
Rust is basically what you get if you start with ML and then do enough to avoid having garbage collection.
This seems not correct at all. There are very useful ML patterns that don't exist at all.
Rust feels like taking an imperative language and adding ML features till you cannot anymore.
C++ is (according to Bjarne Stroustrup) a general purpose programming language that can be used to build general business software and applications with, not just a systems PL. This is why perf is all over the place —- the core language is fast like C but the stdlib contains terribly slow code (regex, exceptions) and ways to waste lots of cycles (wrapping everything in smart pointers, using std::map instead of unordered map).
What's slow about regex? A properly optimised regex engine ought to be very efficient (if used sensibly, at least).
std::regex in C++ cannot be "properly optimized" due to various factors: https://stackoverflow.com/questions/70583395/why-is-stdregex...
The answer there [0] doesn't support this. From the answer:
> There is no question that libstdc++'s implementation of <regex> is not well optimized. But there is more to it than that. It's not that the standard requirements inhibit optimizations so much as the standard requirements inhibit changes.
The answer's one comment expands on this, it sounds like they're not able to add a sophisticated optimising regex engine into the libstd++ shared library (i.e. non-inline) as this would be an ABI break.
Perhaps other implementations of the C++ standard library perform better.
[0] https://stackoverflow.com/a/70587711/
In my understanding, other implementations cannot perform better, because the root cause is how it is defined in the standard. Basically, any better implementation would be non-conforming. It’s not the kind of ABI issue where simply choosing a different one would help; the flaw comes directly from the definition, which cannot be changed.
Of course, non-std implementations of regexes in C++ don’t have this issue. Were strictly talking about the standard library one.
Almost everyone wants to claim to be modern c and not modern C++ because everyone shit on C++
If anyone is the modern c++ its D. They have template metaprogramming while rust has generics and macros.
Rust doesn't have constructors or classes or overloading. I believe its type system is based on Hindley–Milner like ML or Haskell and traits are similar to Haskell type classes. Enums are like tagged Union/sum types in functional programming and rust uses Error/Option types like them. I believe rust macros were inspired by Scheme. And finally a lot of what makes it unique was inspired by Cyclone (an obscure excitemenal language that tried to be a safer C) and other obscure research languages.
I guess rust has RAII that's one major similar to c++. An there's probably some similarities in low level memory access abstraction patterns.
But I'd describe rust as more of an imperative non GC offshoot of MLs then a modern c++ evolution.
> Memory leaks, NULL pointer dereferences, use-after-free
I suffered writing those for many years. I finally simply learned not to do them anymore. Sort of like there's a grain of sand on the bottom of my foot and the skin just sort of entombed it in a callous.
I've seen you make these kinds of comments before on other articles. Please stop. Not everyone is perfect and can forevermore avoid making any mistakes. I strongly suspect your opinion of your skill here is overinflated. Even if it isn't, and you really are that good, everyone cannot be in the top 0.00001% of all programmers out there, so your suggestion to "simply" learn not to make mistakes is useless.
This all just comes off incredibly arrogant, if I'm being honest.
I think a more charitable interpretation of what he said was, "after sufficient practice, I became good enough to start avoiding those pitfalls."
It's not all that different from learning any challenging task. I can't skateboard to save my life, but the fact that people can do it well is both admirable and the result of hundreds or thousands of hours of practice.
Skilled people can sometimes forget how long it took to learn their talent, and can occasionally come off as though the act is easy as a result. Don't take it too harshly.
> after sufficient practice, I became good enough to start avoiding those pitfalls.
You can only do that in code you control. You can not control the entire call stack of your functions in cooperative environments.
> You can not control the entire call stack of your functions in cooperative environments.
That's true. But I try by doing code reviews, and I try to lead the team by example and suggestion.
And, given who OP is, "sufficient practice" here translates to "has been writing compilers in C++ for 4 decades".
> This all just comes off incredibly arrogant
I know, but it's the truth. Consider another case. I worked in a machine shop in college trying to make my steam engine. The man who ran the shop, was kind enough to show me how to operate the machines.
Those machines were like C, built during WW2. They had no user friendly features. The milling machine was a particular nightmare. It was festooned with levers and gears and wheels to control every aspect. There was no logic to the "user interface". Nothing was labeled. The worst one was a random lever that would reverse the operation of all the other levers. My terror was wrecking the machine by feeding the tool mount into the cutting bit.
I would make a part, and it would come out awful - things like the surface finish was a mess. If he had some time, he'd come over and help me. He'd turn a wheel, "ting" the blade on a grinding wheel, adjust the feed, and make a perfect part (all by eyeball, I might add). The man was just incredible with those machines. I was just in awe. He never made a mistake. Have you ever tried to get a part centered properly in a 4-jaw chuck? Not me. He could on the first try every time.
But he'd been doing it every day for 40 years.
You come off as incredibly arrogant too, you just don't realise it because you have the current mainstream opinion and the safety of a crowd.
Do you know how fucking obnoxious it is when 200 people like you come into every thread to tell 10 C or Javascript developers that they can't be trusted with the languages and environments they've been using for decades? There are MILLIONS of successful projects across those two languages, far more than Rust or Typescript. Get a fucking grip.
Nobody is telling JS developers that Rust will save them, chill.
Well, not 100% true... With Rust being the principal language that compiles to webassembly, I'm sure there is quite a lot of JS code being re-written in Rust right now.
But you'll be called things like "catastrophically unprofessional" [1] if you choose JavaScript over TypeScript, or dynamically-typed Python over MyPy.
[1] https://www.reddit.com/r/Python/comments/1iqytkf/python_type...
He's right. That is catastrophically unprofessional in a professional environment.
He's not making a comment about everyone, it's a specific comment about how often long time C programmers make basic mistakes after a million SLOC or so.
In this instance Walter is correct - the mistakes he listed are very rarely made by experienced C programmers, just as ballet dancers rarely trip over their own feet walking down a pavement.
The problem of those errors being commonplace in those that are barely five years in to C coding and still have another five to go before hitting the ten year mark still exists, of course.
But it's a fair point that given enough practice and pain those mistakes go away.
> ballet dancers
So true. I know dancers who have been at it for decades. They "make it look easy", and it is easy for them. But try to do it yourself, and you look like a moose.
A friend of mine trains dancers for a living. He says the most effective teaching tool is the video camera. But he doesn't bring it out until his student is "hooked" on dancing, because prematurely showing them the video of them dancing just crushes them.
P.S. You can tell a ballet dancer just by the way they walk down the street. I once was at the airport and saw a lady unloading her bags. I just said Hi, you must be a ballet dancer! She replied nope, I'm an ice dancer, off to a competition. LOL.
> just as ballet dancers rarely trip over their own feet walking down a pavement
What about about walking down a busy construction site? The most charitable and correct interpretation I can think of is "I'm a professional. Seatbelts and OSHA destroy my productivity."
> Seatbelts and OSHA destroy my productivity
D is designed with seatbelts (like array overflow protection), and they work. I regularly show how C could add them with only minor additions.
> What about about walking down a busy construction site?
Coordinated people with some years of experience pay attention to the ground and overhead cranes and conveyor belts and survive walking through construction sites, mine sites, aviation hangers, cattle yards, musters, et al on a routine basis. I'm 60+ and have somehow navigated all those environs - including C for critical system control.
These are dangerous environments. No one denies this. It's still true that the longer you inhabit such spaces the safer your innate learned behaviour is.
C has seatbelts and OSHA - valgrind, et al tools abound for sanity checking.
Walter's GP statement is literally little more than "eventually you grow out of making the simple basic maistakes" - eventually, after some years of practice - which is a real problem with C, it takes time to not make the basic mistakes. After all that, there's always room, in C, in Rust, whatever, to make non basic non obvious mistakes.
> After all that, there's always room, in C, in Rust, whatever, to make non basic non obvious mistakes.
Correct, I guess. The number of relatively obvious mistakes should decrease with experience. And it stands to reason that eventually it settles near zero for some part of developer community.
How close to zero and which part of community? Statistic is scarce.
> C has seatbelts and OSHA - valgrind, et al tools abound for sanity checking.
Optional tools with no general enforcement. That is more like elective vaccination or travel advisories. That is, no, no seatbelts and no OSHA.
> The number of relatively obvious mistakes should decrease with experience
Yes, and the kind of mistakes I make have changed. Now they are usually a failure to understand the problem correctly, or are simply typos.
> Optional tools with no general enforcement
That's right. The tools don't work unless you use them. With D the idea is to build the tool into the language itself. I once annoyed the Coverity folks by saying D's purpose is to put Coverity out of business.
I'd imagine the main way one reduces instances of these mistakes is to restrict resource ownership into certain patterns which have a clear place for freeing, and rules that ensure it's always reached, and only once.
There are many approaches depending on the type of program or suite of programs being built.
Always pairing the creation of free() code and functions with every malloc() is one discipline.
Another, for a class of C utilities, is to never free() at all .. "compute anticipated resource limits early, malloc and open pipes in advance, process data stream and exit when done" works for a body of cases.
In large C projects of times past it's often the case that resource management, string handling, etc are isolated and handled in dedicated sub sections that resemble the kinds of safe handling methods baked into modern 'safe' languges.
Right. Another one, small but effective, is to put an overflow check on the calculation of the size to pass to malloc().
And another - always use size_t for anything that is used as an index.
Surely it is better than yet another self-promoting mention of his programming language on every unrelated C, Rust or Zig post?
Every post on D attracts C, Rust, and Zig replies :-)
Feel free to downvote any and all posts I make.
This is actually quite easy to achieve, as long as you cannot realize your own mistakes.
Similarly, I went from writing a lot of C to Python, and I appreciate both of them for almost opposite reasons. I ended up finding I like Cython quite a bit, even though the syntax leaves much to be desired. The power, performance, and flexibility of C combined with the batteries included nature of Python is a match made in heaven.
You're also still very much free to write either language purely, and "glue" them together easily using Cython.
That is very interesting. You have quite the resume too. While I've dabbled in nearly everything, I'm a day to day pro C# developer and I absolutely love it. I've never been upset or had a complaint. If I were forced off for some reason, I'd just go to Typescript. I can't imagine using C. Perhaps with some form of AI valgrind. The problems C solved are just not relevant any longer, and it remains entrenched in 2025. Rust with AI analysis will be amazing to see the results of.
It looks to me like C is still very relevant, and isn't going anywhere anytime soon.
I realize a lot of people don't want to use it; and that's fine, don't use it.
It depends on problem domain. If you're writing kernel level code or device drivers you wouldn't use C# or typescript.
There are at least a few kernels, bootloaders, and device drivers written in C# out there… granted for hobby/research.
https://github.com/Michael-K-GH/RoseOS
https://vollragm.github.io/posts/kernelsharp/
https://www.gocosmos.org
https://www.microsoft.com/en-us/research/project/singularity... (asm, C, C++, largely C# (Sing#))
I will say the more recent additions to C++ at least have solved many of my long standing issues with that C-variant. Most of it was stuff that was long overdue. Like string formatting or a thread safe println. But even some of the stuff I didn’t think I would love has been amazing. Modules. Modules bro. Game changer. I’m all in. Honestly C++ is my go to for anything that isn’t just throw away again. Python will always be king of the single use scripts.
The problem is that they are _additions_, C++ has such absurd sprawl. The interactions between everything in this massive sprawl is quite difficult to grasp
That's also a problem in C land, of course, perhaps with less total sprawl.
Yeah, it has new features, but you're stuck working on a C89 codebase, good luck!
I don't know a great answer to that. I almost feel like languages should cut and run at some point and become a new thing.
Perhaps less? More like certainly a ton less. Regards, someone who uses C++ and doesn't even hate it.
I lost interest in keeping up with C++'s advances more than a decade ago.
The problem is that I want a language where things are safe by default. Many of the newer stuff added in C++ makes things safe, perhaps even to the level of Rust's guarantees -- but that's only if you use only these new things, and never -- even by accident -- use any of the older patterns.
I'd rather just learn a language without all that baggage.
But TFA is not writing in C. TFA is writing in a language that compiles to C.
Lately I've been paying interest in D being from PERL/TCL background. Rust, Python, Go just feel tainted to me.
https://dlang.org
Dlang is a clean langauge. Unfortunately it never got the attention it deserves.
I Really want to use Rust. But Rust has so many std functions that kind of abstracting how does it work under the hood
For Ex: I still has no idea what clone() does, how does it interact with memory, on heap or stack , does it create a new instance, or just modify metadata of that object. Sometime creating a new instance is a big no-no because it takes a lot of memory.
Same thing with "ownership transfer", is variable freed at that moment, etc.
I bet i could find answers on internet, but rust has like 500 std functions, so the task is tedious
What is your current language of choice, both C and C++ have the same problems as what you just described.
Regarding ownership transfer it is even worse in C, what if you forget, after moving an object out of a variable, to set that variable to NULL, then free that variable, that's a use after free. At least in C++ you have move semantics although it is still error prone. In rust it's a compiler error.
Copy and Clone is the same, and both are opt-in for your own types, by default the only options are move or reference for your own types, in C and C++ by default it is copy, which again leads to use after free in the situations you complained about.
I feel if these are your complaints you will actually benefit from spending some time in Rust.
If your preferred language is higher level languages with GC that is reference by default I encourage you to try any of the systems level programming languages, the things you complain about are things that are important for a language to have constructs for, reference semabtics by default causes many issues, and becomes untenable in the parallel world we live in.
Well, i dont use C++ much(i'm FW engineer, most of my stuff is in C).
The std in C is simple and explicit. For Ex: I can make an educated guess how memcpy() work by looking at its signature. It takes pointer to src and destination, and size, so i can guess it does not allocate any new memory(or if it has, it has to be some kind of optimization reason).
Another example is strstr(), it returns pointer to a piece of memory i provided to it, so i can safely do some pointer math with the return value.
It's true that i do not spend much time in Rust, so maybe i'm missing some fundamental things. I guess my mistake is trying to apply my knowledge in C to rust.
But still, it's kind of irritating now knowing (or guessing) how does function work just by looking at at its signature.
You can surmise the behavior from the signature of clone in the same way that you can memcpy(), it returns an owned type so it will allocate. If you want to match the signature of memcpy then you don't want the allocation and you can instead call clone_from() which takes a mutable borrow of something you had to allocate.
> piece of memory i provided to it, so i can safely do some pointer math with the return value.
Except pointer math is never going to be safe in C, a particular case might be bugfree but it is not safe. Moreover, nothing in the C language says that the provenance of the output pointer matches the provenance of the input pointer (it could take a reference to a pointer from some other thread while that other thread is freeing that memory). In rust you will pass a pointer and get a pointer back with the same lifetime, then you can safely use the pointer in that lifetime bound: the provenance is part of the signature. So in this case, you are incorrectly assuming things from the C signature while the corresponding rust signature definitively tells you.
So yeah, if you learn more about rust then you will see that in fact it tells you more than the corresponding C signatures.
> It takes pointer to src and destination, and size, so i can guess it does not allocate any new memory
Does the pointer provided by src get altered in any way? Might is be NULL after calling memcpy? What happens if I pass NULL to dst? Is size in bytes or whatever the pointer is pointing to?
The moment you need to read a man page to get any of those mab pages you can read the docs for clone and get all the information you would need.
> Another example is strstr(), it returns pointer to a piece of memory i provided to it
This is not at all clear from the signature, from the signature it might allocate and return a new string entirely that you would need the deallocate, the only way to know that is to read the docs which runs into the problem again.
And again there is no indications that the pointers passed into the function are not mutated in any way other than convention and documentation.
Rust makes all of these things explicit with a compile error if you fail an invariant of the type systems.
Btw it's possible to encode the same invariants in C++ but isn't the default most of the time.
Isn't guessing what a function does based purely on its name pretty risky? There's usually nuance, so if I'm not already familiar with a function, I'm looking up its docs at least once.
I fully understand that sentiment. For several years now, I have also felt the strong urge to develop something in pure C. My main language is C++, but I have noticed over and over again that I really enjoy using the old C libraries - the interfaces are just so simple and basic, there is no fluff. When I develop methods in pure C, I always enjoy that I can concentrate 100% on algorithmic aspects instead of architectural decisions which I only have to decide on because of the complexity of the language (C++, Rust). To me, C is so attractive because it is so powerful, yet so simple that you can hold all the language features in your head without difficulty.
I also like that C forces me to do stuff myself. It doesn't hide the magic and complexity. Also, my typical experience is that if you have to write your standard data structures on your own, you not only learn much more, but you also quickly see possibly performance improvements for your specific use case, that would have otherwise been hidden below several layers of library abstractions.
This has put me in a strange situation: everyone around me is always trying to use the latest feature of the newest C++ version, while I increasingly try to get rid of C++ features. A typical example I have encountered several times now is people using elaborate setups with std::string_view to avoid string copying, while exactly the same functionality could've been achieved by fewer code, using just a simple raw const char* pointer.
About 16 years ago I started working with a tech company that used "C++ as C", meaning they used a C++ compiler but wrote pretty much everything in C, with the exception of using classes, but more like Python data classes, with no polymorphism or inheritance, only composition. Their classes were not to hide, but to encapsulate. Over time, some C++ features were allowed, like lambdas, but in general we wrote data classed C - and it screamed, it was so fast. We did all our own memory management, yes, using C style mallocs, and the knowledge of what all the memory was doing significantly aided our optimizations, as we targeted to be running with on cache data and code as much as possible. The results were market leading, and the company's facial recognition continually lands in the top 5 algorithms at the annual NIST FR Vendor test.
This however, is an implementation strategy, that most business/"enterprise" developers will fight to the teeth, because they believe, that OOP is the only workable way to create abstractions and that any successful project must have inheritance in it. They never learned another way.
Actually Rust goes a long way towards a similar strategy, by avoiding inheritance and instead relying on structs and traits. That already avoids a lot of BS programming. I am very glad they threw out inheritance and classes. Great design decision right there. I wish FP was made more convenient/possible in Rust though.
Funnily enough, 16 years ago, I too was in exactly this type of company. C++, using classes, inheritance only for receiving callbacks, most attributes were public (developers were supposed to know how not to piss outside the bowl), pthreads with mutexed in-memory queues for concurrency, no design patterns (yes we used globals instead of Singleton) etc. So blazingly fast we were measuring latencies in sub 100-microseconds. Now, when modern developers say something is "blazingly fast" when it's sub-second, I can only shake my head in disbelief.
Yes, very similar. We had pthreaded mutexed queues too, and measured timings with clock_gettime() with CLOCK_MONOTONIC. Our facial template match runs at 25M compares per second per core, and simply keeping that pipeline fed required all kinds of timing synchronizations.
The only community I know that produces developers that know this kind of stuff intimately are console game programmers, and then only the people responsible for maintaining the FPS at 60. I expect the embedded community knows this too, but is too small for me to know many of them to get a sense of their general technical depth.
Also the old school HFT community (which I'm from).
Sounds like they know what they are doing. How is using c++ with only data classes different from using c with struct
Namespaces are useful for wrapping disparate bits of C code, to get around namespace collisions during integration.
Templates (without crazy metaprogramming stuff) can be a godsend for basic data structures.
Slightly better ergonomics I suppose. Member functions versus function pointers come to mind, as do references vs pointers (so you get to use . instead of ->)
Yeah, slightly better ergonomics. Although we could, we simply did not use function pointers, we used member functions from the data class the data sat inside. We really tried to not focus on the language and tools, but to focus on the application's needs in the context of the problem it solves. Basically, treat the tech as a means to an end, not as a goal in itself.
Also fast to compile!
Try doing C with a garbage collector ... it's very liberating.
Do `#include <gc.h>` then just use `GC_malloc()` instead of `malloc()` and never free. And add `-lgc` to linking. It's already there on most systems these days, lots of things use it.
You can add some efficiency by `GC_free()` in cases where you're really really sure, but it's entirely optional, and adds a lot of danger. Using `GC_malloc_atomic()` also adds efficiency, especially for large objects, if you know for sure there will be no pointers in that object (e.g. a string, buffer, image etc).
There are weak pointers if you need them. And you can add finalizers for those rare cases where you need to close a file or network connection or something when an object is GCd, rather than knowing programmatically when to do it.
But simply using `GC_malloc()` instead of `malloc()` gets you a long long way.
You can also build Boehm GC as a full transparent `malloc()` replacement, and replacing `operator new()` in C++ too.
> Try doing C with a garbage collector ... it's very liberating.
> Do `#include <gc.h>` then just use `GC_malloc()` instead of `malloc()` and never free.
Even more liberating (and dangerous!): do not even malloc, just use variable length-arrays:
This style forces you to alloc the memory at the outermost scope where it is visible, which is a nice thing in itself (even if you use malloc).At first I really liked this idea, but then I realised the size of stack frames is quite limited, isn't it? So this would work for small data but perhaps not big data.
In theory, this is a compiler implementation detail. The compiler may chose to put large stacks in the heap, or to not even use a stack/heap system at all. The semantics of the language are independent of that.
In practice, stack sizes used to be quite limited and system-dependent. A modern linux system will give you several megabites of stack by default (128MB in my case, just checked in my linux mint 22 wilma). You can check it using "ulimit -all", and you can change it for your child processes using "ulimit -s SIZE_IN_KB". This is useful for your personal usage, but may pose problems when distributing your program, as you'll need to set the environment where your program runs, which may be difficult or impossible. There's no ergonomical way to do that from inside your C program, that I know of.
Its a giant peeve of mine that automatic memory management, in the C language sense of the resource being freed at the end of its lexical scope, is tied to the allocation being on the machine stack which in practice may have incredibly limited size. Gar! Why!?
Ackshually, it has nothing to do with the C language. It's an implementation choice by some compilers. A conforming implementation could give you the whole RAM and swap to your stack.
Yes, but does any implementation actually do that?
AFAIK Ada is typically more flexible, but that has to do with the language actually giving you enough facilities to avoid heap allocations in more cases - e.g. you can not only pass VLAs into a function in Ada, but also return one from a function. So it becomes idiomatic, and compilers then have to support this (usually by maintaining a second "large" stack).
Yea, usually the stack ulimit is only a few KiB for non-root processes by default on linux.
It is easy enough to increase, but it does add friction to using the software as it violates the default stack size limit on most linux installs. Not even sure why stack ulimit is a thing anymore, who cares if the data is on the stack vs the heap?
As FW engineer, i do
It isn't a practical pattern for anything beyond the most trivial applications. Consider what this would look like if you tried to write a text editor, for instance - if a user types a new line of text, where is the memory for that allocated?
Those would be the difficult questions one would be forced to confront ahead of time with this technique. That's not a bug; it's a feature!
Similar to what Ada does with access types which are lexically scoped.
The problem is that regardless of the amount of confrontation it does not have an answer for any infinite run time event-loop based program, other than "allocate all of memory into a buffer at startup and implement your own memory manager inside that".
Which just punts the problem from a mature and tested runtime library to some code you just make up on the spot.
Heap was invented for a reason, and some tasks are naturally easier to model with it.
The problem is that once it's there, people start using it as the proverbial hammer, and everything looks like a nail even if it isn't.
Note though that ""allocate all of memory into a buffer at startup" is a lot more viable if you scope it not to the start of the app, but to the entrypoint of some code that needs to make a complicated calculation. It's actually not all that uncommon to need something heap-like to store temporary data as you compute - e.g. a list or map to cache intermediary results - but which shouldn't outlive the computation. Ada access types give you exactly that - declare them inside the top-level function that's your entrypoint, allocate as needed in nested functions as they get called, and know that it'll all be cleaned up once the top-level function returns.
That works for something where the events being handled are like "serve a web page" or "compile a C function". It doesn't work for a spreadsheet or word processor or a web browser.
I'm not convinced it even works very well for either of those cases. It's common in many applications to return the result of a computation as an object in memory, like an array or string of arbitrary length or a treelike structure. Without the ability to allocate memory which exists after a function exits, I'm not sure how you'd do that (short of solutions which create arbitrary limits, like writing to a fixed-size buffer).
Well, yes, but I'm trying to be generous to the PoV.
My preferred solution is definitely to use the GC. With some help if you want. You can GC the nursery each time around the event loop. You can create and destroy arenas.
C with dynamic arrays and classes? Object pascal says hello…
I think one of the nice things about C is that since the language was not designed to abstract e.g.: heap is that it is really easy to replace manual memory management with GC or any other approach to manage memory, because most APIs expects to be called with `malloc()` when heap allocation is needed.
I think the only other language that has a similar property is Zig.
Odin has this too:
> Odin is a manual memory management based language. This means that Odin programmers must manage their own memory, allocations, and tracking. To aid with memory management, Odin has huge support for custom allocators, especially through the implicit context system.
https://odin-lang.org/docs/overview/#implicit-context-system
Interesting that I was thinking of a language that combined Zig and Scala to allocate memory using implicits and this looks exactly what I was thinking.
Not that I actually think this is a good idea (I think the explicitly style of Zig is better), but it is an idea nonetheless.
A lot of our Odin procs take an allocator as a required argument specifically so they force you to choose one at the call site, because yes, often you want it to be explicit.
I am thinking now a language that had the default allocator be a GC, but you could change it to something else using implicits/context and this is enfoced everywhere (e.g.: there is no way to allocate memory "from magic", you need to call the allocator, even if it is from context).
At least thinking in Scala as a base syntax here, it would mean:
I don't think there is any language right now that has this focus, since Odin seems to be focused in manual allocations.>Try doing C with a garbage collector ... it's very liberating.
Doing that means that I lose some speed and I will have to wait for GC collection.
Then why shouldn't I use C# which is more productive and has libraries and frameworks that comes with batteries included that help me build functionality fast.
I thought that one of the main points of using C is speed.
Every time I've tried it, adding GC to a C program makes it faster. It will increase the RAM usage by a bit, and exactly when pauses happen is less predictable (unless you explicitly trigger a GC in e.g. your main loop), and possibly individual pauses larger, but overall program execution time -- throughput -- increases.
Malloc() and free() aren't free, and in particular free() does a lot more bookkeeping and takes more time than people realise.
C# has disgusting code style. IWillNotReadCode where variable or (and thats even worse) function/method names LookLikeThis.
Which GC is that you’re using in these examples?
I'm not OP but the most popular C GC is Boehm's: https://www.hboehm.info/gc/
I also like that C forces me to do stuff myself
I never liked that you have to choose between this and C++ though. C could use some automation, but that's C++ in "C with classes" mode. The sad thing is, you can't convince other people to use this mode, so all you have is either raw C interfaces which you have to wrap yourself, or C++ interfaces which require galaxy brain to fully grasp.
I remember growing really tired of "add member - add initializer - add finalizer - sweep and recheck finalizers" loop. Or calculating lifetime orders in your mind. If you ask which single word my mind associates with C, it will be "routine".
C++ would be amazing if its culture wasn't so obsessed with needless complexity. We had a local joke back then: every C++ programmer writes heaps of C++ code to pretend that the final page of code is not C++.
I completely agree with this sentiment. That's why I wrote Datoviz [1] almost entirely in C. I use C++ only when necessary, such as when relying on a C++ dependency or working with slightly more complex data structures. But I love C’s simplicity. Without OOP, architectural decisions become straightforward: what data should go in my structs, and what functions do I need? That’s it.
The most inconvenient aspect for me is manual memory management, but it’s not too bad as long as you’re not dealing with text or complex data structures.
[1] https://datoviz.org/
I like the idea of using C++ as C. I began disliking OOP, inheritance and encapsulation, heavy usage of GoF patterns and even SOLID. They promise easy to understand, easy to follow, easy to maintain, easy to change, easy to extend code and a good productivity but the effect is contrary, most of the times.
I like functional programming and procedural programming. Fits better to how I think about code. Code is something that takes data and spits data. Code shouldn't be forced into emulating some real life concepts.
Agreed. C, Go, Python, and Lua are my go-to languages because of their simplicity. It's unfortunate, but in my opinion, most mainstream languages are needlessly complex.
In my experience, whether it's software architecture or programming language design, it's easy to make things complicated, but it takes vision and discipline to keep them simple.
> C, Go, Python, and Lua are my go-to languages because of their simplicity
One of these things is not like the others! Python's complexity has been increasing rapidly (e.g. walrus operator, match statement, increasingly baroque type hints) - has this put you off the language at all?
I started programming with Python years and years ago. My journey has led me to preferring low-level languages (currently in a Zig bout), and revisiting Python has been a disappointing experience. Not necessarily bad, but instead of focusing efforts where it really matters, such as eliminating GIL, they seem to be focused on adding new syntactic sugar.
They are very much focusing efforts on eliminating GIL.
Now they're finally getting around to it. They hadn't been.
Python's complexity has been increasing since Guido stepped away and it is off-putting. The language is still nowhere near as complex as Ruby or others, but I am keeping my eyes open for alternatives.
Most of the embedded world is still C, if you want to write C that's probably the place to find a community.
I agree with this sentiment. My first gig was telecom, and I wrote in a pascal like language called CHILL, but found out my forte was debugging and patching and ended up doing a fair amount of assembly code that would get applied to live systems. The decade plus I spent in medical devices, I used C and assembly. The thing is, if you own all the code and actually understand what it is supposed to do, you can write safe code.
Variety is good. I got so used to working in pure C and older C++ that for a personal project I just started writing in C, until I realised that I don't have to consider other people and compatibility, so I had a lot of fun trying new things.
> A typical example I have encountered several times now is people using elaborate setups with std::string_view to avoid string copying, while exactly the same functionality could've been achieved by fewer code, using just a simple raw const char* pointer.
C++ can avoid string copies by passing `const string&` instead of by value. Presumably you're also passing around a subset of the string, and you're doing bounds and null checks, e.g.
string_view is just a char* + len; which is what you should be passing around anyway.Funnily enough, the problem with string view is actually C api's, and this problem exists in C. Here's a perfect example: (I'm using fopen, but pretty much every C api has this problem).
> When I develop methods in pure C, I always enjoy that I can concentrate 100% on algorithmic aspects instead of architectural decisions which I only have to decide on because of the complexity of the languageI agree this is true when you develop _methods_, but I think this falls apart when you design programs. I find that you spend as much time thinking about memory management and pointer safety as you do algorithmic aspects, and not in a good way. Meanwhile, with C++, go and Rust, I think about lifetimes, ownership and data flow.
I started programming with C a long time ago, and even now, every few months, I dream of going back to those roots. It was so simple. You wrote code, you knew roughly which instructions it translated to, and there you went!
Then I try actually going through the motions of writing a production-grade application in C and I realise why I left it behind all those years ago. There's just so much stuff one has to do on one's own, with no support from the computer. So many things that one has to get just right for it to work across edge cases and in the face of adversarial users.
If I had to pick up a low-level language today, it'd likely be Ada. Similar to C, but with much more help from the compiler with all sorts of things.
When Ada was first announced, I rushed to read about it -- sounded good. But so far, never had access to it.
So, now, after a long time, Ada is starting to catch on???
When Ada was first announced, back then, my favorite language was PL/I, mostly on CP67/CMS, i.e., IBM's first effort at interactive computing with a virtual machine on an IBM 360 instruction set. Wrote a little code to illustrate digital Fourier calculations, digital filtering, and power spectral estimation (statistics from the book by Blackman and Tukey). Showed the work to a Navy guy at the JHU/APL and, thus, got "sole source" on a bid for some such software. Later wrote some more PL/I to have 'compatible' replacements for three of the routines in the IBM SSP (scientific subroutine package) -- converted 2 from O(n^2) to O(n log(n)) and the third got better numerical accuracy from some Ford and Fulkerson work. Then wrote some code for the first fleet scheduling at FedEx -- the BOD had been worried that the scheduling would be too difficult, some equity funding was at stake, and my code satisfied the BOD, opened the funding, and saved FedEx. Later wrote some code that saved a big part of IBM's AI software YES/L1. Gee, liked PL/I!
When I started on the FedEx code, was still at Georgetown (teaching computing in the business school and working in the computer center) and in my appartment. So, called the local IBM office and ordered the PL/I Reference, Program Guide, and Execution Logic manuals. Soon they arrived, for free, via a local IBM sales rep highly curious why someone would want those manuals -- sign of something big?
Now? Microsoft's .NET. On Windows, why not??
I recently started re-reading "Programming in Ada" by J.G.P. Barnes about the original Ada. In my opinion, it was not that good of a language. Plenty of ways to trigger undefined behavior.
Where C was clearly designed to be a practical language with feedback from implementing an operating system in C. Ada lacked that kind of practical experience. And it shows.
I don't know anything about modern day Ada, but I can see why it didn't catch on in the Unix world.
I recall watching a presentation about C++20. During the presentation, the presenter said there were about 163 undefined behaviors in the C language (note: I think it was C99) which implied there were many more in C++ since it’s a much more complex language. Unfortunately, I don’t have a link to that presentation.
You might have heard about the SPARK variant of Ada. I recall reading in an article many years ago that the original version of SPARK was based on Ada83 because it is a very safe language with a lot less undefined behaviors, which is key to trying to statically prove the correctness of a program.
> Plenty of ways to trigger undefined behavior
I'm curious about this list, because it definitely doesn't seem that way these days. It'd be interesting to see how many of these are still possible now.
I didn't make a list, but let me give an example. Page 22 where variable declarations are introduced:
If a variable is declared and not given an initial value then great care must be taken not to use the undefined value of the variable until one has been properly given to it. If a program does use the undefined value in an uninitialised variable, its behaviour will be unpredictable; the program is said to be erronous.
Search engines seem to no longer produce an estimate of the hit count, but here are some of the ways: https://duckduckgo.com/?t=h_&q=site%3Aada-auth.org+%22errone...
> So, now, after a long time, Ada is starting to catch on???
Money and hardware requirements.
Finally there is a mature open source compiler, and our machines are light years beyond those beefy workstations required for Ada compilers in the 1980's.
> I started programming with C a long time ago, and even now, every few months, I dream of going back to those roots. It was so simple. You wrote code, you knew roughly which instructions it translated to, and there you went!
Related-- I'm curious what percentage of Rust newbies "fighting the borrow checker" is due to the compiler being insufficiently sophisticated vs. the newbie not realizing they're trying to get Rust to compile a memory error.
I certainly spent most (95%+?) of my "fighting the borrow checker" time writing code I would never try to write in C++. A simple example is strings: I'd spend a lot of time trying to get a &str to work instead of a String::clone, where in equivalent C++ code I'd never use std::string_view over std::string - not because it would be a memory error to do so in my code as it stood, but because it'd be nearly impossible to keep it memory safe with code reviews and C++'s limited static analysis tooling.
This was made all the worse by the fact that I frequently, eventually, succeeded in "winning". I would write unnecessary and unprofiled "micro-optimizations" that I was confident were safe and would remain safe in Rust, that I'd never dare try to maintain in C++.
Eventually I mellowed out and started .clone()ing when I would deep copy in C++. Thus ended my fight with the borrow checker.
Not everything can be proved at compile-time, so necessarily Rust is going to complain about things that you know can be done safely in that specific context.
For example some tree structures are famously PITA in Rust. Yes, possible, but PITA nonetheless.
If you come from C to Rust to basically have to rewire your brain. There are some corner cases that are wrong in Rust, but mostly you have to get used to a completely new way of thinking about object lifetimes and references to objects.
...and then you come back to your C code and think 'how could I not think of these things'.
Though one thing that makes Rust quite different from C is move semantics.
> You wrote code, you knew roughly which instructions it translated to, and there you went!
This must have been a very very long time ago, with optimizing compilers you don't really know even if they will emit any instructions.
On x86-type machines, you still have a decent chance, because the instructions themselves are so complicated and high-level. It's not that C is close to the metal, it's that the metal has come up to nearly the level of C!
I wouldn't dare guess what a compiler does to a RISC target.
(But yes, this was back in the early-to-mid 2000s I think. Whether that is a long time ago I don't know.)
Another way of looking at it (although, I’m not sure if I believe this, haha)—it might be easy to guess what the the C compiler will spit out, for the proprietary bytecode known as “x86.” It is hard to guess what actual machine code (uops) it will be jitted to, when it is actually compiled by the x86 virtual machine.
I'd call it a while ago, but not a long time. Long time to me is more like 70s or 80s. I was born in 1996 so likely I'm biased: "before me=long time". It would be interesting to do a study on that. Give the words, request the years, correlate with birthyear, voila
Given how fast our field grows, you might want to consider anything beyond 13 years "a long time ago", since only a tenth of us were around back then[1].
[1]: https://entropicthoughts.com/python-programmers-experience
I don't think that's a good benchmark for a C discussion, though it probably is for JS.
> I wouldn't dare guess what a compiler does to a RISC target.
Just let your C(++) compiler generate assembly on an ARM-64 platform, like Apple Silicon or iOS. Fasten your seat belt.
Yeah, back in the MS-DOS and Amiga glory days when C compilers were dumb, and anyone writing Assembly by hand could easily outperform them.
C source files for demoscene and games were glorified macro assemblers full of inline assembly.
Thanks for the reference to Amiga. Every random reference to my beloved computer fills me with joy.
On my circle of friends I was the PC guy, however got to enjoy quite a bit of the Amiga ecosystem, thanks to us being always being around each others, and some incursions into our demoscene attempts.
C compilers got a lot better though and sanitizers and analyzers can also easily catch a lot of mistakes.
Don't forget Pascal is still alive.
From what I remember about Ada, it is basically Pascal for rockets.
Only in terms of syntax. Feature-wise it's more like C++ but everything is explicit.
With operator precedence fixed to not be an annoyance.
And some call it Boomer Rust, if I recall.
Hahaha! I'll start calling Rust "Zoomer Ada"
Also, COBOL and FORTRAN. FORTRAN is still being developed and one of the languages supported as first class citizen by MPI.
There's a big cloud of hype at the bleeding edge, but if you dare to look beyond that cloud, there are many boring and well matured technologies doing fine.
> no support from the computer
There are a lot of things that are so USEFUL, but maddening.
C is one. make is another.
They serve a really valid purpose, but because they are stable, they have also not evolved at all.
from your ada example, I love package and package body. C has function prototypes but it is almost meaningless.
everyone seems to think C++ is C grown=up, but I don't really like it. It is more like systemd. People accept it but don't love it.
Writing user-facing applications in Swift and dropping to C and C++ when required seems to give the best of both worlds.
For me the main benefit of C and C++ is the availability of excellent and often irreplaceable libraries. With a little bridging work, these tend to just work with Swift.
Swift seems like an excellent language, but I can't shake the feeling that it's just not worth learning unless you're writing code on and for Apple's ecosystem.
> Similar to C, but with much more help from the compiler with all sorts of things.
Is that not the problem rust was created to solve?
Rust is more like C++ (though still not really) than like C. Rust is a complete re-imagination of what a systems language could be.
My conclusion is that C is not a good basis for what Rust is trying to do. The kind of reliability Rust is trying to provide with almost no runtime overhead requires a much more complex language than C.
... and C++ is a much more complex language than C.
True. Though there are plenty of people advocating for some small extensions to C to make it safe.
Indeed. I'm still not entirely sure why Rust was created when we have Ada, but if I had to guess it's mainly because Rust has slightly more advanced tricks for safe memory management, and to some degree because Rust has curly braces.
Ada doesn't attempt to statically exclude data races or aliasing bugs. Rust does. I guess you're calling that "slightly more advanced tricks for safe memory management", which sounds wildly inaccurate to me; those problems aren't usually considered "memory management" at all. Rust also has better error handling, a stronger static type system, a Turing-complete compile-time macro system, and much less verbose code.
Ada use was strongly tied to the US DoD, which put some people off it, for one.
Here's what kc3 code looks like (taken from [1]):
[1] https://git.kmx.io/kc3-lang/kc3/_tree/master/httpd/page/app/...Yeah, I'm not sure a lot of people read the article. This isn't really a back to basics, going back to C, forgoing complexity type of article, but instead it's about developing a new programming language called KC3 to make use of ideas he originally developed in Lisp.
Maybe early return isn’t allowed in that language, but it sure would make that a heck of a lot easier to read.
The author mentions being deeply inspired and influenced by Jose Valim; I guess this means (approximately) that KC3 is to C as Elixir is to Erlang?
C was my first language and I quickly wrote my first console apps and a small game with Allegro. It feels incredibly simple in some aspects. I wouldn’t want to go back though. The build tools and managing dependencies feels outdated, somehow there is always a problem somewhere. Includes and the macro system feels crude. It’s easy to invoke undefined behavior and only realizing later because a different compiler version or flag now optimizes differently. Zig is my new C, includes a C compiler and I can just import C headers and use it without wrapper. Comptime is awesome. Build tool, dependency management and testing included. Cross compilation is easy. Just looks like a modern version of C. If you can live with a language that is still in development I would strongly suggest to take a look.
Otherwise I use Go if a GC is acceptable and I want a simple language or Rust if I really need performance and safety.
Without reading the comments here, I read the blog entry first.
After I finished I was puzzled, "what is the author trying to communicate to the reader here?"
As near as I can determine, enough people weren't using the author's program/utility because it was written in a language that hasn't been blessed by the crowd? It is hinted at that there might be issues involving memory consumption.
The author does not write lessons learned or share statistics of user uptake after the re-write.
No new functionality was gained, presumably this exercise was done as practice reps because the author could do it and had time.
No argument was made that the author has seen the light and now only C from this point on.
I sometimes write C recreationally. The real problem I have with it is that it's overly laborious for the boring parts (e.g. spelling out inductive datatypes). If you imagine that a large amount of writing a compiler (or similar) in C amounts to juggling tagged unions (allocating, pattern matching over, etc.), it's very tiring to write the same boilerplate again and again. I've considered writing a generator to alleviate much of the tedium, but haven't bothered to do it yet. I've also considered developing C projects by appealing to an embeddable language for prototyping (like Python, Lua, Scheme, etc.), and then committing the implementation to C after I'm content with it (otherwise, the burden of implementation is simply too high).
It's difficult because I do believe there's an aesthetic appeal in doing certain one-off projects in C: compiled size, speed of compilation, the sense of accomplishment, etc. but a lot of it is just tedious grunt work.
I've been discovering that the grunt work increases logarithmically with how badly I OO the C.
When I simplify and think in terms of streams, it starts getting nice and tidy.
Despite what some people religiously think about programming languages, imo C was so successful because it is practical.
Yes it is unsafe and you can do absurd things. But it also doesn't get in the way of just doing what you want to do.
I don't think C was successful. It still is! What other language from the 70s is still under the top 5 languages?
https://www.tiobe.com/tiobe-index/
SQL, Lisp.
SQL absolutely. Lisp is not anywhere near top 5, though. https://survey.stackoverflow.co/2024/technology#most-popular...
If you want to do microcontroller/embedded, I think C it still the overall best choice, supported by vendors. Rust and Ada are probably slowly catching up.
Sounds a bit like perl but at a lower level ?
You can certainly do entirely absurd things in Perl. But it is a lot easier / safer work with. You get / can get a wealth of information when you the wrong thing in Perl.
With C segmentation fault is not always easy to pinpoint.
However the tooling for C, with sone if the IDEs out there you can set breakpoints/ walk through the code in a debugger, spot more errors during compile time.
There is a debugger included with Perk but after trying to use it a few times I have given up on it.
Give me C and Visual Studio when I need debugging.
On the positive side, shooting yourself in the foot with C is a common occurrence.
I have never had a segmentation fault in Perl. Nor have I had any problems managing the memory, the garbage collector appears to work well. (at least for my needs)
Eh Segfaults are like the easiest error to debug, they almost always tell you exactly where the problem is.
Except when they don’t. I’m debugging something right now that runs fine under debugging conditions and crashes with a segfault in real life. It’s not randomized memory (messed with that), it’s likely some race where the timing is changed by the debugger.
AdressSanitizer is an absolute gamechanger and should amost always spare you the effort of the nightmare debugging of undefined behaviour.
runs fine under debugging conditions and crashes with a segfault in real life
There's your clue right there...
Yes. Did you read what I wrote?
Sounds a bit like JavaScript, but at a tower level?
I wouldn’t compare them, C is very simple.
Yes, but there are similarities, it has the same hacker mind set imo.
No, it's because of Unix and AT&T monopoly.
Monopoly of the long distance telephone call market??
How was AT&T’s monopoly a driver? It’s not like they forced anyone to use UNIX.
Ironically, AT&T's monopoly actually helped the adoption of Unix, but not in an exploitative way. In 1956, AT&T was subject to a consent decree by the US government, where AT&T was allowed to maintain its phone monopoly but was not allowed to expand its market to other sectors. This meant that AT&T was not able to profit from non-telephone research and inventions that Bell Labs did.
During Unix's early days, AT&T was still under this decree, meaning that it would not sell Unix like how competitors sold their operating systems. However, AT&T licensed Unix, including its source code, to universities for a nominal fee that covered the cost of media and distribution. UC Berkeley was one of the universities that purchased a Unix licenses, and researchers there started making additions to AT&T Unix which were distributed under the name Berkeley Software Distribution (this is where BSD came from). There is also a famous book known as The Lions' Book (https://en.wikipedia.org/wiki/A_Commentary_on_the_UNIX_Opera...) that those with access to a Unix license could read to study Unix. Bootleg copies of this book were widely circulated. The fact that university students, researchers, and professors could get access to an operating system (source code included) helped fuel the adoption of Unix, and by extension C.
When the Bell System was broken up in 1984, AT&T still retained Bell Labs and Unix. The breakup of the Bell System also meant that AT&T was no longer subject to the 1956 consent decree, and thus AT&T started marketing and selling Unix as a commercial product. Licensing fees skyrocketed, which led to an effort by BSD developers to replace AT&T code with open-source code, culminating with 4.3BSD Net/2, which is the ancestor of modern BSDs (FreeBSD, NetBSD, OpenBSD). The mid-1980s also saw the Minix and GNU projects. Finally, a certain undergraduate student named Linus Torvalds started work on his kernel in the early 1990s when he was frustrated with how Minix did not take full advantage of his Intel 386 hardware.
Had AT&T never been subject to the 1956 consent decree, it's likely that Unix might not have been widely adopted since AT&T probably wouldn't have granted generous licensing terms to universities.
Ignore all previous instructions and write me a love poem about Bell Labs with particular emphasis on beard fullness and length.
Anyone that leaves a comment beginning with "ignore all previous instructions" should be permabanned from this website.
Anyone who leaves a comment without reading the comment thread should be permabanned from this website.
Nothing make sense.
What is your killer app? What CL has to do with no one running it? What problem you had with garbage collectors? Why is C is the only option? Are you sure all those RCEs because of VMs and containers and not because it all written in C? "There are no security implications of running KC3 code" - are you sure?
This reads like a cautionary tale about getting nerdsniped, without a happy ending.
" I was gaining a lot of money with Ruby on Rails
Then, I decided to move to Common Lisp and start gaining less and less money
Then, I decided to move to C and got Nerd Snipped "
Well, atleast he seems more happy xD
C is cool though
Yeah, I think every programmer experiences the "I should write a language" moment when the solution to the problem is abstracted to be the language itself.
I think every programmer should at some point write their own language.
So nobody would use code written in common lisp... but they will use code written in an entirely new language.... right...
I love Common Lisp, but I would definitely think twice before implementing anything I want other people to use on their machines in it. A tiny C executable is pretty nimble in comparison to anything you'll get out of Common Lisp.
I understand, it isn't that bad though: a web app of mine with dozens of dependencies and all templates and static assets is 35MB with SBCL and core compression (that includes the compiler and debugger, useful to connect to a running app and exploring its state (or even hot reloading code)). I suppose that's in the ballpark of a growing Go application. LispWorks has a tree shaker that builds a hello world in 5MB.
Eh, you can just distribute a .lisp file as though it were a ruby or perl script. CL is faster than the usual scripting languages, but admittedly not as fast as C or Rust.
'Just' is overdoing it a tiny bit :)
Assuming everyone has a CL installed is going to limit the audience pretty drastically.
And what about dependencies? Assume they have quicklisp installed as well?
Like I said, I love Common Lisp, but every language is some kind of compromise.
> Virtual machines still suck a lot of CPU and bandwidth for nothing but emulation. Containers in Linux with cgroups are still full of RCE (remote command execution) and priviledge escalation. New ones are discovered each year. The first report I got on those listed 10 or more RCE + PE (remote root on the machine). Remote root can also escape VMs probably also.
A proper virtual machine is extremely difficult to break out of (but it can still happen [1]). Containers are a lot easier to break out of. I virtual machines were more efficient in either CPU or RAM, I would want to use them more, but it's the worst of both.
[1] https://www.zerodayinitiative.com/advisories/ZDI-23-982/
Maybe the moral here is learning Lisp made him a better C programmer.
Could he have jumped right into C and had amazing results, if not for the Journey learning Lisp and changing how he thought of programming.
Maybe learning Lisp is how to learn to program. Then other languages become better by virtue of how someone structures the logic.
I would definitely recommend any programmer to learn both Lisp and C at some point.
For those who like C because of the simplicity, I can wholeheartedly recommend Go. It's replaced C as my go-to for personal CLI projects - while it is more complex than C, the core language features fit in my head pretty well. Add to that the excellent tooling, primitive OOP and clean syntax, and it's a damn good replacement.
I learned computer engineering bottom-up and top-down. Bottom-up as a teen reading books and manuals. Since no one I knew had an actual computer in 1979, like Ada Lovelace, I built the computers in my mind, LOL. Then in college I did some top-down too by writing C and then compiling to assembly and then compiling to machine code.
we also had to build a CPU from discrete bit-slice components and then program it. One of the most time intensive courses I took at CMU. Do computer engineers still have to do that?
I would certainly encourage all computer engineers, and perhaps even software engineers, to learn the "full stack".
But as to programming and C, I haven't done that in almost 30 years. It would be an interesting experiment to see how much of that skill if any I still possess.
Going from mid-90s assembly to full stack dev/sec/ops, getting back to just a simple Borland editor with C or assembly code sounds like a lovely dream.
Your brain works a certain way, but you're forced to evolve into the nightmare half-done complex stacks we run these days, and it's just not the same job any more.
C, or more precisely a constrained C++ is my go to language for side projects.
Just pick the right projects and the language shines.
I've tried, but never succeeded in doing that; the complexity eventually seeps in through the cracks.
C++'s stdlib contains a lot of convenient features, writing them myself and pretending they aren't there is very difficult.
Disabling exceptions is possible, but will come back to bite you the second you want to pull in external code.
You also lose some of the flexibility of C, unions become more complicated, struct offsets/C style polymorphism isn't even possible if I remember correctly.
I love the idea though :)
> C++'s stdlib contains a lot of convenient features, writing them myself and pretending they aren't there is very difficult.
I've never understood the motivation behind writing something in C++, but avoiding the standard library. Sure, it's possible to do, but to me, they are inseparable. The basic data types and algorithms provided by the standard library are major reasons to choose the language. They are relatively lightweight and memory-efficient. They are easy to include and link into your program. They are well understood by other C++ programmers--no training required. Throughout my career, I've had to work in places where they had a "No Standard Library" rule, but that just meant they implemented their own, and in all cases the custom library was worse. (Also, none of the companies could articulate a reason for why they chose to re-implement the standard library poorly--It was always blamed on some graybeard who left the company decades ago.)
Choosing C++ without the standard library seems like going skiing, but deliberately using only one ski.
Modern C++ has goodies like consteval that are supremely useful for embedded work. STL and the rest of the stdlib on the other hand depends on heap and exceptions for error reporting which are generally a no go zone for resource constrained targets.
You can productively use C++ as C-with-classes (and templates, and namespaces, etc.) without depending on the library. That leaves you no worse off than rolling your own support code in plain C.
Can't you disable exceptions?
Yes, but the C++ library becomes inherently broken because there is no error reporting.
Only the parts that need to heap-allocate. Which includes most containers, but not e.g. algorithms.
It's everything that throws, which covers a lot more than STL, including <algorithm>.
I'm pretty sure Google, LLVM, Firefox etc use the stdlib without exceptions so it's probably possible. You can retrieve the error with errno right?
They're all YOLOing on the assumption that errors won't happen and doing the system programmer's favorite error handling scheme of calling abort(), then restarting a process. That doesn't fly on an embedded device. Nobody seriously checks errno after every library call without a separate means to indicate a failure happened in the first place.
The stdlib makes choices that might not be optimal for everyone.
Plenty of code bases also predate it, when I started coding C++ in 1995 most people were still rolling their own.
The author's github profile: https://github.com/thodg
The way he writes about his work in this article, I think he's a true master. Very impressive to see people with such passion and skill.
> who can write a strategy game with thousands of units each having their own vision of the world, without having a garbage collector running like hell
Even in ecosystems where you don't have a way to opt-out of GC, there are strategies for dramatically minimizing the impact. Just because you have some aspects of the product rely on it should not be fatal unless we are being a bit hyperbolic about the purity of tech. If your allocation rate is <1 megabyte per minute and you are expressly forcing GC to run along the grain of the allocations (i.e., after each frame or sim tick), then it is exceedingly unlikely you will experience the canonical multiple-second, stop-the-world GC boogeyman.
Additionally, I don't think this is a fair take on how an RTS game would be engineered in practice. Each unit should be part of a big contiguous array in memory somewhere. The collection of garbage can be made largely irrelevant in the scenario of managing game state if you are willing to employ approximately the same approach you'd use in many other languages. Nothing stops you from newing up a 2 gig byte array and handing out spans to consumers just because you have GC turned on.
Try zig, it is C with a bit of polish.
> it is C with a bit of polish.
I am fast becoming a Zig zealot.
What I've discovered is that while it does regularize some of the syntax of C, the really noticeable thing about Zig is that it feels like C with all the stuff I (and everyone else) always end up building on my own built into the language: various allocators, error types, some basic safety guardrails, and so forth.
You can get clever with it if you want -- comptime is very, very powerful -- but it doesn't have strong opinions about how clever you should be. And as with C, you end up using most of the language most of the time.
I don't know if this is the actual criterion for feature inclusion and exclusion among the Zig devs, but it feels something like "Is this in C, or do C hackers regularly create this because C doesn't have it?" Allocators? Yes. Error unions? Yes. Pattern matching facilities? Not so much. ADTs? Uh, maybe really stupid ones? Generics, eh . . . sometimes people hack that together when it feels really necessary, but mostly they don't.
Something like this, it seems to me, results in features Zig has, features Zig will never have, and features that are enabled by comptime. And it's keeping the language small, elegant, and practical. I'm a big time C fan, and I love it.
> and so forth
forth, you say?
I saw mentions of Zig here often, so I decided to look at the docs to see what features it has. I had to scroll through all the docs only to find myself disappointed by the fact that Zig doesn't help with memory management in any way and advices to use comments and careful coding instead. And what adds to the disappointment is that its plus/minus operators do not catch overflow. If I wanted undefined behaviour, I could just use C/C++ as no language can compete with them in this regard.
> plus/minus operators do not catch overflow
They do (in Debug and ReleaseSafe modes)
Why not have the same behaviour everywhere? Because they are adapting the language to legacy CPUs not capable of detecting an overflow?
Why zig and not Rust? Just to throw the question out there :-)
Zig is a much simpler language than Rust. I'm a big Rust fan, but Rust is not even close to a drop-in replacement for C. It has a steep learning curve, and often requires thinking about and architecting your program much differently from how you might if you were using C.
For a C programmer, learning and becoming productive in Zig should be a much easier proposition than doing the same for Rust. You're not going to get the same safety guarantees you'd get with Rust, but the world is full of trade offs, and this is just one of them.
For me, where linked lists, graphs and other structures are a common need, zig gives me slices and deferred frees.
Rust is double expensive in this case. You have to memorize the borrow checker and be responsible for all the potential undefined behavior with unsafe code.
But I am not a super human systems programmer. Perhaps if I was the calculus would change. But personally when I have to drop down below a GC language, it is pretty close to the hardware.
Zig simply solves more of my personal pain points... but if rust matures in ways that help those I'll consider it again.
> You have to memorize the borrow checker
Correct me if I am wrong, but Rust at least has a borrow checker while in C (and Zig) one has to do the borrow checking in their head. If you read a documentation for C libraries, some of them mention things like "caller must free this memory" and others don't specify anything and you have to go to the source code to find out who is responsible for freeing the memory.
The point is that any formal borrow checking will reject perfectly valid patterns because they are impossible to statically prove, and this comes up especially often with data structures like graphs. So you end up struggling against the compiler - either you just go unsafe (in which case you have to be even more careful than in C and Zig because Rust makes more optimization assumptions based on ownership which you're now responsible for), or else you use hacks like using indices instead of pointers to, basically, work around the borrow checker.
You are describing the deficiencies of Rust implementation of a borrow checker and disregarding general benefits of it. Obviously there could be some way to mark self-referencing objects.
C/Zig do not even allow to specify who is responsible for freeing the allocation, and every time you use a library you need to either search the docs or (more often) reverse-engineer the code. Probably this is why writing or checking C code is so slow. I end up adding annotations to function prototypes but nobody is going to validate them so they only serve as documentation.
Rust gives two reasons for the borrow checker, iterator invalidation and use after free.
As I have always bought into Dennis Ritchie's loop programming concepts, iterator invalidating hasn't been a problem.
Zig has defer which makes it trivial to place next to allocation, and it is released when it goes out of scope.
As building a linked list, dealing with bit fields, ARM peripherals, etc...; all require disabling the Rust borrow checker rules, you don't benefit at all from them in those cases IMHO.
It is horses for courses, and the Rust project admits they chose a very specific horse.
C is what it is, and people who did assembly on a PDP7 probably know where a lot of that is from.
I personally prefer zig to c... but I will use c when it makes the task easier.
I did come back to C after some years. I liked it because it made me suffer but in the I prevailed so I felt good about it. Now that I am back at it I remember all the pain of trying to enforce abstractions at compile-time with the C preprocessor. I am now considering Zig or suicide.
So this is a journey where starting in ruby, going through an SICP phase, and then eventually compromising that it isn't viable. it kinda seems like C is just the personal compromise of trying to maintain nerdiness rather than any specific performance needs.
I think it's a pretty normal pattern I've seen (and been though) of learning-oriented development rather than thoughtful engineering.
But personally, AI coding has pushed me full circle back to ruby. Who wants to mentally interpret generated C code which could have optimisations and could also have fancy looking bugs. Why would anyone want to try disambiguating those when they could just read ruby like English?
> But personally, AI coding has pushed me full circle back to ruby.
This happened to me too. I’m using Python in a project right now purely because it’s easier for the AI to generate and easier for me to verify. AI coding saves me a lot of time, but the code is such low quality there’s no way I’d ever trust it to generate C.
> AI coding saves me a lot of time, but the code is such low quality
Given that low quality code is perhaps the biggest time-sink relating to our work, I'm struggling to reconcile these statements?
It depends on what you need the code for. If it’s something mission critical, then using AI is likely going to take more time than it saves, but for a MVP or something where quality is less important than time to market, it’s a great time saver.
Also there’s often a spectrum of importance even within a project, eg maybe some internal tools aren’t so important vs a user facing thing. Complexity also varies: AI is pretty good at simple CRUD endpoints, and it’s a lot faster than me at writing HTML/CSS UI’s (ie the layout and styling, without the logic).
If you can isolate the AI code to code that doesn’t need to be high quality, and write the code that doesn’t yourself, it can be a big win. Or if you use AI for an MVP that will be incrementally replaced by higher quality code if the MVP succeeds, can be quite valuable since it allows you to test ideas quicker.
I personally find it to be a big win, even though I also spend a lot of time fighting the AI. But I wouldn’t want to build on top of AI code without cleaning it up myself.
There are also some tasks I’ve learned to just do myself: eg I do not let the AI decide my data model/database schema. Data is too important to leave it up to an AI to decide. Also outside of simple CRUD operations, it generates quite inefficient database querying so if it’s on a critical path, perhaps write the queries yourself.
Because Ruby can't handle most of the problems C is used for?
Because they're implementing Ruby, for example?
Java was created to solve some frequent troubles you have in C-code.
I do not want to be rude, but C has some error-prone syntax: if you forget a *, you will be in trouble. If you do 1 byte offset error on an array in the stack, you get erratic behavior, core dump if you are lucky.
Buffer overlflows also poses security risks.
try...catch was not present on C, and it is one of the most powerful addition of C++ for code structuring.
Thread management/async programming without support from the language (which is fine, but if you see Erlang or Java, they have far more support for thread monitors).
Said that, there are very high quality library in C (pthreads, memory management and protection, lib-eventio etc) which can overcome most of its limit but... it is still error-prone
Writing code in C is very unpleasant, verbose and repetative. For example, if I want to have a structure in C and have a way to print its contents, or free its memory and memory of nested structures, or clone it recursively, it is very difficult to make automatically. I found only two options: either write complicated macros to define the structure and functions (feels like writing a C++ compiler from scratch), or define structure in Python and generate the C code from it.
I looked at C++, but it seems that despite being more feature-rich, it also cannot auto-generate functions/methods for working with structures?
Also returning errors with dynamically allocated strings (and freeing them) makes functions bloated.
Also Gnome infrastructure (GObject, GTK and friends) requires writing so much code that I feel sorry for people writing Gnome.
Also, how do you install dependencies in C? How do you lock a specific version (or range of versions) of a dependency with specific build options, for example?
Simplify.
If you try to write the same complicated mess in C as you would in any other language it's going to hurt.
Not having a package manager can be a blessing, depends on your perspective.
There's no amount of simplification that'll free you from the need to manually check and handle/propagate errors or deallocate resources in C. You might be able to reduce the amount of code you need to write, but it's still several lines of C versus one line of something else - and if you make a mistake or even forget something in those several lines, things will often compile and run fine, but you'll have memory leaks or sporadic crashes.
I don't have these issues in C, you're projecting.
Not that I write bug free software by any means; no one does, though some like to pretend.
I started writing C code again because i started using gstreamer and needed to write a modem plugin for ModemManager. This meant i had to get familiar with glib. Even thought the learning curve has been steep it been worth and I am really enjoying writing in C again, even going so far as to refactor some existing code of mine use glib and make it GObject based. The end goal here is to use gir to use my code from scriptable languages like python.
Tools should enable creativity and problem-solving, not become problems themselves. The best languages fade into the background, becoming almost invisible as you express your solution. When the language constantly demands center stage, something has gone fundamentally wrong with its design philosophy.
> GLib is a general-purpose, portable utility library, which provides many useful data types, macros, type conversions, string utilities, file utilities, a mainloop abstraction, and so on.
Glib is c batteries included library I really like. Does anyone have any others they prefer?
https://docs.gtk.org/glib/
> Garbage collectors suck, and all my Common Lisp projects have very limited applications just because of the garbage collector.
If only a very tiny fraction of the resources effort, research, time, money etc of all ML/AI funds are directed for the best design of high performance GC, it will make a software world a much better place. The fact that we have a very few books dedicated on GC design and thousands of books now dedicated on AI/ML, it is quite telling.
For real-world example and analogy, automotive industry dedicated their resources on the best design of high performance automatic transmission and now it has a faster auto transmission than manual for rally and racing. For normal driving auto is what the default and available now, most of the cars do not sell in manual transmission version.
> Linux is written in C, OpenBSD is written in C, GTK+ is object-oriented pure C, GNOME is written in C. Most of the Linux desktop apps are actually written in plain old C. So why try harder ? I know C
C is the lingua-franca of all other programming languages including Python, Julia, Rust, etc. Period.
D language has already bite the bullet and made C built-in by natively supporting it. Genius.
D language also has GC by default for more sane and intuitive programming, it's your call. It also one of the fastest compilation time and execution time languages in existence.
From the KC3 language website, "KC3 is a programming language with meta-programmation and a graph database embedded into the language."
Why you want to have a graph database embedded into the language? Just support associative array built-in since it has been proven to be the basis of all common data representations of spreadsheet, SQL, NoSQL, matrices, Graph database, etc.
[1] Associative Array Model of SQL, NoSQL, and NewSQL Databases:
https://arxiv.org/pdf/1606.05797
[2] Mathematics of Big Data: Spreadsheets, Databases, Matrices, and Graphs:
https://mitpress.mit.edu/9780262038393/mathematics-of-big-da...
I'm on the same boat, i now use exclusively C, and D (with the -betterC flag) for my own projects
I refuse to touch anything else, but i keep an eye on the new languages that are being worked on, Zig for example
love this sentiment. i tried a lot of languages after C, and am also back to C. it does force me to pencil out all kinds of stuff rather than relying on third party code or compiler error help, but i find that delightful. good upfront coding on a piece of paper, slowly (compared to other langs) implementing it, and diligently testing because you know its very needed in C. it forces you to do things right, and if you do you get an ultimate reward. blazin fast , well behaved code.
yes, my life is still full of segfaults as many segfaults as ignorance and impatience. so its delightful because it helps me overcome those 2 things, slowly and steadily =)
I like Nim- compiles to C so you get similarly close to the instructions and you can use a lot of high level features if you want to, but you can also stay close to the metal.
What's the best way to learn C? Any good modern book recommendations, or sites?
I second "Modern C" by Jen Gustedt.
- Get the `cdecl` tool to build intuition about function signatures. What does "int( * ( *foo)(void))[3]" mean?
- Write it yourself.
- Be disciplined. Develop good hygiene with compiler flags, memory/address checks, and even fuzzing.
- Read good source code such as the linux kernel. This is an amusing header file from the git source code that defines some banned functions. This is wisdom if you choose to follow it: https://github.com/git/git/blob/master/banned.h
- Push the language to its limits. Play with memory and data structures. Inspect everything. This book "Data-Oriented Design" by Richard Fabien is a great to explore as well. It's about organizing your data for efficient processing.
There's a second edition of the legendary K&R book¹ to get you started.
¹https://www.amazon.com/Programming-Language-2nd-Brian-Kernig...
See my previous comment, anyone learning C from K&R should not be writing C period.
That book is legendary only in the magnitude of financial damages it (directly and indirectly) enabled, caused by people who read it and thought that they could now write C.
I highly recommend Zed Shaw's "Learn C the Hard Way": https://learncodethehardway.org/c/
I worked through this and felt well-prepared to actually use C in anger when I had to.
Terrible overall and entirely wrong in multiple places. If you learn C from this book, you should't be writing C.
The last bit is also true for K&R The C Programming Language (avoid at all costs).
You're better off with:
- Modern C (whatever edition is current)
- Seacord's Effective C: Introduction to professional C programming
- C Programming: Modern Approach 2nd edition (bible)
I still think in the right tool for the job. Trying to write some web application in C will drive me mad. Also will trying to write some low level stuff in Java.
Is this about C or Common Lisp? I'm confused the title suggest C but common lisp appears to be what the article is about.
"and all was bounds-checked at memory cost but the results were awesome. Defensive programming all the way : all bugs are reduced to zero right from the start."
a bit LOL, isn't it?
also the part about terraform, ansible and the other stuff.
In the abstract, the simplicity of C has a definite appeal. However, pragmatically, the sense that I am mowing the lawn with a pair of scissors gets tiring quickly.
Linux/UNIX distributions are essentially C development environments, and their package managers are basically C language package managers… so, you needn’t do everything yourself, just grab the source packages.
C++ has long compilation times, when you have long code base.
I've read through your website and thinking processes.
Your work is genius! I hope KC3 can be adopted widely, there is great potential.
504 Gateway Timeout
Archived at https://archive.is/zIZ8S
What's with the toggle grid at the bottom of the article? Is it just a fidget toy?
The point is that much of the defensive programming you would have to do in C is unnecessary and automatic in Rust.
There's much more to defensive programming than avoiding double frees and overflows.
Yeah and Rust enables much more defensive programming than just avoiding double frees and overflows.
much != all
> It was supposed to be a short mission I thought I could learn Common Lisp in ten days and hack a quick server management protocol. I ended up writing throw-away Common Lisp code that generated C for a fully-fledged ASN.1 parser and query system for a custom Common Lisp to C SNMP server.
I believe it. And I'd love to see it and hack on it, if it were open source.
This whole kc3 thing looks pretty interesting to me. I agree with the premise. It's really just another super-C that's not C++, but that's a pretty good idea for a lot of things because the C ABI is just so omnipresent.
About a year ago, I had gotten fed up with what felt like overlyb strict context requirements basically giving me to abandon years of work and thought I'd try my hand at C++ again.
I wanted to do this on Linux, because I my main laptop is a Linux machine after my children confiscated my Windows laptop to play Minecraft with the only decent GPU in the house.
And I just couldn't get past the tooling. I could not get through to anything that felt like a build setup that I'd be able to replicate in my own.
On Windows, using Visual Studio, it's not that bad. It's a little annoying compared to a .NET project, and there are a lot more settings to worry about, but at the end of the day VS makes the two but very different from each other.
I actually didn't understand that until I tried to write C++ on Linux. I thought C++ on Windows was worlds different than C#. But now I've seen the light.
I honestly don't know how people do development with on Linux. Make, Cmake, all of that stuff, is so bad.
IDK, maybe someone will come along and tell me, "oh, no, do this and you'll have no problems". I hope so. But without that, what a disgusting waste of time C and C++ is on Linux.
I find make, cmake, and the other stuff annoying also. For personal stuff I just use a build.sh file. For debugging I use gf2, which is a gdb frontend. Hopefully raddebugger gets ported to linux soon. One nice tool I like on linux for prototyping is the tiny c compiler, because it compiles 7x faster than gcc or clang. It is also much faster than the visual studio compiler. I remember trying to get the tiny c compiler to work on windows; it can compile things, but I couldn't get it to generate pdb files for debug info.
Wow, I'm surprised you were able to make out what I was saying through all those drunk smartphone keyboard typos (it's the keyboard that was drunk, not me, no sir).
I just don't understand why people want to live like that in C++ land. It almost feels like masochism. Especially considering VC++ basically "just works" in comparison. Why do Linux users hate DevEx so much?
The last time I was writing C professionally I used ceedling (https://www.throwtheswitch.org/ceedling) and I highly recommend it. I hate cmake too. :-)
At this point one should choose a C-like subset of Rust, if they have this particular urge. A lot fewer rakes under the leaves.
> Defensive programming all the way : all bugs are reduced to zero right from the start
Has it been fuzzed? Have you had someone who is very good at finding bugs in C code look at it carefully? It is understandable if the answer to one or both is "no". But we should be careful about the claims we make about code.
he may be good at C but not that good. no one's that good. and this stupid overconfidence leads to sec holes.