[growlNark]

> Embracing Resource Acquisition Is Initialization (RAII) makes C++ the safest, most productive and fun C++ has ever been.

This seems like an extremely low bar.

Anyway, what use is there for C++ in 2025 aside from maintenance of legacy codebases and game engines? Off-hand I'd say C++ programmers are twice as expensive as rust programmers for the same semantics, and then you're still stuck with a C++ programmer you need to figure out how to socialize with.

[green7ea]

Author here, I do prefer a language like Rust that makes all of this automatic but that isn't always an option. There are other parts of Rust that can be hard as well.

I wrote this article for a few friends who have recently started working with an existing, large C++ code base. Some industries like the video game industry have also stuck to C++ (they have their reasons).

A good programmer can work within the given constraints to make a useful program — sometimes, one of those constraints is the choice of language.

[BoingBoomTschak]

Once Qt is properly accessible in Rust, I think your claim will be a lot more realistic. This goes for other important C/C++ libraries (e.g. libavcodec, vips/ImageMagick, VTK) without a solid (maintained, documented, etc...) Rust interface.

[tialaramex]

Qt is a C++ kitchen sink library - perhaps it feels very important to C++ programmers but I can't imagine Rust people are looking forward to having more redundant types that do exactly the same thing as a type they've used for years but like, now with a Q in their name.

The fact that lists like yours so often end up being "Look at all these C libraries" ie not actually about C++ at all is revealing. It's an endorsement of Bjarne's position that he needed that C compatibility, decades later C++ alternatives remain unpopular but it also tells you that you're never going to raise the bar this way. C++ is not a route out.

AFAIK there is no equivalent of rustls-openssl-compat for C++. The knowledge that this library (OpenSSL) is trash never spurred any C++ programmers to do better and provide the same ABI but with a C++ implementation.

[BoingBoomTschak]

What's your point? There's a lot to criticize about these libraries, but the fact remains that they're considered important and basically impossible to rewrite from scratch. Of course, Rust could get a rock solid portable GUI toolkit that doesn't rely on a webview (probably not much harder than wrapping Qt, to be honest), but it's not there yet.

I'm not in love with them or the abomination known as C++, if that's what you're implying.

[ogoffart]

> Rust could get a rock solid portable GUI toolkit that doesn't rely on a webview [...], but it's not there yet.

What would you say is missing in Slint?

> probably not much harder than wrapping Qt, to be honest

There are already Qt bindings for rust (cxx-qt, qmetaobject-rs). What's missing from these?

[tialaramex]

> basically impossible to rewrite from scratch

How so? This definitely needs at least a citation of somebody who can explain in detail why this code is "impossible to rewrite". Who wrote the one we have now, Martians?

[motorest]

> This seems like an extremely low bar.

Is it, though? Most mainstream languages fail to support anything resembling RAII, at least as first-class support. Do you actually have an example of a language that does a better job at resource management than C++?

[asyx]

Rust simply because the constructor being a static method on the type that can return basically anything including optional and result types and copy / move semantics are handled very easily through macros instead of constructor / assignment operator overrides.

[motorest]

> Rust simply because the constructor being a static method (...)

This assertion makes no sense. RAII is not defined by whether you use factory functions or special member functions to initialize an object. In fact, RAII is only incidentally related ro memory management. RAII is a strategy to manage any and all types of resources, ranging from memory management to even files and TCP connections, by leveraging assurances that the runtime provides regarding scopes and object life cycles. None of this changes if you employ factory functions to initialize resources.

[palata]

Not that I agree with the condescending parent (e.g. "you're still stuck with a C++ programmer you need to figure out how to socialize with"), but...

> Most mainstream languages fail to support anything resembling RAII

Wikipedia says: "RAII is associated most prominently with C++, where it originated, but also Ada,[3] Vala,[4] and Rust"

> Do you actually have an example of a language that does a better job at resource management than C++?

They don't necessarily offer a pattern like RAII, but what about "try-with-resources" in Java, or "use" in Kotlin that goes with `AutoCloseable`?

And what about "using" in C#?

What about "defer" in Swift?

I find those simpler than RAII.

[motorest]

> Wikipedia says: "RAII is associated most prominently with C++, where it originated, but also Ada,[3] Vala,[4] and Rust"

From those you listed, only Rust can be described as mainstream. Do you think that one out of a couple dozens refutes the statement that "most mainstream languages fail to support anything resembling RAII"?

> They don't necessarily offer a pattern like RAII, but what about "try-with-resources" in Java, or "use" in Kotlin that goes with `AutoCloseable`?

Try-with-resources and the disposable pattern offer similar features but they still fall a bit short. Unlike RAII, they are not thread-safe, require specialized syntax and boilerplate code, and require manually specifying scopes.

But even if you consider try-with-resources and Disposable pattern as a perfect replacement of RAII, now point out how many mainstream languages support them. You have Java and JVM languages, you have C# and .NET languages, Python, and...

> I find those simpler than RAII.

Arguably this boils down to personal taste, but RAII actually ensures your resources will be released, and you can tell exactly when this will happen. Disposable patterns don't, and screwing up a using with statement is all it takes to get your application to leak resources and fail silently.

[pjmlp]

One caveat though, you need to pair those with static analysis tooling, if you want to ensure developers don't forget to make use of them.

[znkr]

I agree, this is a real benefit of RAII compared to defer. That said, there are disadvantages of making the resource freeing part invisible. Not having to debug destructors that do network IO (e.g. closing a buffered writer that writes to the other side of the world) or expensive computation (e.g. buffered IO writing to a compressor) is a define plus. Don’t get me started on error handling…

[motorest]

> I agree, this is a real benefit of RAII compared to defer. That said, there are disadvantages of making the resource freeing part invisible.

RAII doesn't make resource freeing invisible. It makes it obvious, deterministic, and predictable. The runtime guarantees that, barring an abnormal program termination that's not handled well, your resource deallocation will take place. This behavior is not optional, or requires specifying ad-hoc blocks or sections.

[palata]

How does RAII solve that? Developers can "forget" to use RAII, right? Or are you saying that it's easier to spot because RAII requires quite a bunch of boilerplate (whereas a one-liner like "defer" is easier to forget)?

Not criticising, really trying to understand :-).

[pjmlp]

RAII when supported natively by the language, is done at the implementation level, not the usage point.

So in languages like Rust, D, Ada, Swift, C++, the compiler will do the rest unless you go out of your way to avoid the call to take place, like placing a value type on the heap using plain pointers.

With the other approach, even if you implement IDisposable, AutoCloseable, ContextManager and similar, you have to remember to manually write the code pattern that takes care of calling close(), or whatever the method/function happens to be called.

In languages with good support for FP patterns, like trailing lambdas, currying and such, there is another pattern, that is much safer, in case you don't want a static analysis tool to track resource usage, the with pattern.

You do something like withDBConnection connection (fun db -> all related db operations).

Assuming the lambda doesn't do naughty things to have db parameter escape the scope, the withDBConnection function will take care of handling the whole connection lifecycle.

[znkr]

When I switched from C++ to a bunch of other languages, I missed RAII initially. However, I quickly learned that other languages just do it differently and often even better (ever had to check an error when closing a resource?). Nowadays, I think that RAII is a solution to a C++ problem and other languages have a better solution for resource management (try-with-resources in Java, defer in Go, with in Python).

[motorest]

> However, I quickly learned that other languages just do it differently and often even better (ever had to check an error when closing a resource?).

I don't think that's even conceptually the same. The point of RAII is that resource deallocation is ensured, deterministic, and built into the happy path. Once you start to throw errors and relying on those to manage resources, you're actually dealing the consequences of not having RAII.

[oytis]

> try-with-resources in Java, defer in Go, with in Python

Or 'goto error8;' in C. Still RAII is much more convenient, especially for cases where you allocate a lot of interdependent resources at different time points. It keeps deallocation logic close to allocation logic (unlike, say, defer), makes sure deallocation always happens in the reverse order of allocations and doesn't force you to create a nested scope each time you allocate a resource

[tialaramex]

Obviously a language with linear types is just better at this, such as Austral.

In terms of languages you'd actually deploy today Rust is better both at this narrow feature and more broadly.

[pjmlp]

Ada/SPARK, with better ergonomics than Rust, but its use will stay niche in high integrity computing.

[oytis]

Where are C++ programmers paid better than Rust programmers? I thought Rust salaries are being driven high by all the crypto/fintech companies + scarcity

[bluGill]

productivity counts not hours. If you can get the same work done faster then you cost to the company is lower even - possible even if your salery is higher.

though it isn't clear how much of rust's increased productivity is caused by being a new language where the architecture mistakes of the past decades are not slowing you down. We will need several more decades to answer that.

[pjmlp]

For one, the compiler and runtime implementation of the language you might happen to use, that that includes Rust.

[flohofwoe]

> stuck with a C++ programmer you need to figure out how to socialize with.

You're obviously trolling, but:

IME Rust attracts the same 'difficult' characters that are also attracted to C++ (because both are 'puzzle solving languages' instead of 'problem solving languages') the typical Rust coder may be even worse because of the missionary fervor and 'holier than thou' attitude.

[oytis]

Puzzle-solving languages are rather Lisp or Haskell. Both C++ and Rust are very practical for their problem domain.

[sesm]

IDK about Haskell, but iterating in REPL with Lisp is the most practical form of programming I've experienced. In other mainstream practical languages this approach is reintroduced as a productivity tool like Quokka, etc.

C++ was practical some decades ago (hardware-friendly variant of OOP for GUI), but it failed as a library language and the domain where it's practical on modern hardware is much smaller. I will not say anything about Rust.

[delta_p_delta_x]

> but it failed as a library language

This is very inaccurate. Essentially every high-performance library, user-mode driver, desktop application, and more is written in nothing but C++. Give me any library you can think of, and I assure you it is written in C++ (or maybe C, but this is masochism on the part of the developers). Even libraries for other languages like numpy, pandas, pytorch, etc are written in C++.

[flohofwoe]

> or maybe C, but this is masochism on the part of the developers

C is the better choice when interoperability with other languages is needed (technically: a C API, the implementation language doesn't matter - but if a C++ implementation is wrapped in C API as an afterthought the result is usually a shitty C API). Personally I switched to C from C++ for writing libraries ca 2017 and don't regret that decision one bit.

Also, many C++ coders only have a foggy idea how convenient working in modern C can be, because their idea of C is usually the 'common C/C++ subset' that exists in C++, and this is stuck deep in the early 90s (for instance the designated-init feature in C++20 is a mere shadow of what C99 designated-init can do - to a point that the C++20 version is pretty much useless for real world usage).

[delta_p_delta_x]

> only have a foggy idea how convenient working in modern C can be

Here is a list of C++ features that C doesn't have, that are an immediate deal-breaker for me:

  - reference and move semantics
  - templates, and type constraints with concepts
  - namespaces
  - `constexpr`, `consteval`, and other compile-time magic
  - `auto` type deduction
  - trailing return types 
  - RAII (can't believe I put this this late, but eh)
  - a passable (although still not perfectly complete) standard library that blows the C standard library out of the water
  - improved semantics that allow programmers to reason about the logic in code better, than obsess over pointer arithmetic
  - built-in virtual functions, function pointer tables, etc

I can list more, but this is going to end up as a list of 'essentially every feature in C++ that isn't in C', which is the very reason for the former language to exist.

[oytis]

Repl is cool, but Python also has a repl, and has a much more intuitive programming model (and much fewer braces).

Totally agree about the domain - no one is writing enterprise applications in C++ these days, luckily. It still does have its domain though, where there is not much choice apart from C++ or Rust (or C if you are a dinosaur)

[djmips]

Oh no, C++ has template meta programming and the ability to mask a DSL as advanced architecture, and then you can even implement Lisp or pick your favorite while claiming you're coding in C...++

[ninkendo]

> the typical Rust coder may be even worse because of the missionary fervor and 'holier than thou' attitude

Anecdata, but the number of times I actually encounter these missionary Rust coders (the RIIR types) is utterly dwarfed by the number of times I hear people complaining about them. The memes making fun of the insufferable rust evangelists are at least 10x as prevalent as the evangelists themselves.