Hacker News Reader

I disagree with almost everything you said in that post; what's more, you're objectively wrong. Let's look at your claims.

>Every new C++ article about the latest "improvements" to the language tends to confirm my belief that C++ developers are masochists. They don't seem to be interested in actually improving the language and making in better and easier at creating useful software. [...] The entire ecosystem feels like a infinite pile of new features fixing old ones, bringing new flaws in the process.

Can you give a concrete example? C++ evolution can be divided in roughly two eras: C++98, and Modern C++ (C++11 and above). This latter introduces a new philosophy, an expansion of features that make it look like a completely new language indeed. From ubiquitous type deduction, to lambdas, to compile-time computation without TMP, the language has introduced powerful new abstractions, always with the trademark "zero cost" that gives C++ its power.

>[Code snippet]

>I don't want a language with magic.

What about that snippet is magic that leaves you so appalled? You got a great (new) feature: guaranteed compile-time conditionals with the same syntax as regular if statements. Depending on whether the object has serialize or not, the code will instantiate to each of the branches, guaranteeing elision of a potentially expensive runtime check and branch.

>Show me simplicity and elegance. [...] Show me how I don't have to choose between value, reference, raw pointer, unique_ptr, shared_ptr, etc, to pass a damn parameter to your function.

Show me a language that simpler and more elegant than C++, where you don't have to worry about choosing how to pass parameters, or manage memory, and that still is as fast as C++. You simply can't, because there's no such thing. C++ delivers ZERO-COST ABSTRACTIONS, that's its power, that's its raison d'être. No other language I'm aware of that delivers this, apart from maybe Rust or D.

--

I realize it's hip to hate on C++, but please keep it rational.

> Show me a language that simpler and more elegant than C++, where you don't have to worry about choosing how to pass parameters, or manage memory, and that still is as fast as C++.

As someone who built their career on memory constrained, fast code in C++ I'm gonna have to say that Rust is simpler and more elegant.

You've got your zero cost abstractions, heck I can decided static vs dynamic dispatch at compile time. Lots of things are more explicit leading to less foot-guns that you see with implicit type conversion and the like. Borrow-checker catches all those nasty dataraces and ownership problems up front. I've yet to run across anything I could do in C++ that I can't do in Rust.

Listen, I love C++ but I think you may want to expand your horizons a bit before declaring it the best language ever in this space.

Even if Rust is better than C++, that doesn’t mean C++ is bad.

Yes, actually, it does. Because C++'s mantra has always been zero cost abstractions. And people have defended the language because "you can't do better and still have the same performance". Well, it seems you can.

Languages are never compared across a single dimension (except to prove a point on the internet).

Rust does come with a different bag of complexity. I'm thrilled to see competition with C++ in this space finally and super excited about Rust, but let's not pretend Rust is easy, either. It has a notoriously monsterous learning curve (which is why there's a major focus to fix that curve: https://github.com/rust-lang/rust-roadmap/issues/17 )

And in this particular context Rust also still forces you to figure out how to pass parameters and manage memory. It "just" helps you with when to call free, which std::unique_ptr also helps you with. Rust makes it a compile error to have a bug, which is awesome, but the "you have to figure out memory allocation & layout" complexity is still mostly there.

> It has a notoriously monsterous learning curve

That learning curve exists in C++ too, it's just hidden under the iceberg of 'interesting' runtime behavior. There's nothing like trying to printf a data race only to have the print call insert a fence and make your data race disappear.

> And in this particular context Rust also still forces you to figure out how to pass parameters and manage memory.

This is a lot simpler in Rust. Everything is value type by default, Box<T> for heap allocations, which even then are managed mostly for you via RAII(yeah I know there's std::unique_ptr but you still also have new, delete, new[], delete[], etc). Parameter types are value, mut ref or const ref. Move semantics are implicit if you have a value type. You don't have const/volatile/etc function qualifiers.

implicit type conversion(esp constructors) can be a total bastard and trip you up in nasty ways. Even more annoying if you're trying to bit-pack or do clever math.

Rust got to learn from a lot of C++'s mistakes and is to me a better language for it.

> This is a lot simpler in Rust. Everything is value type by default, Box<T> for heap allocations, which even then are managed mostly for you via RAII(yeah I know there's std::unique_ptr but you still also have new, delete, new[], delete[], etc). Parameter types are value, mut ref or const ref. Move semantics are implicit if you have a value type. You don't have const/volatile/etc function qualifiers.

That's not simpler at all - it's basically the same. C++ is also value type by default and you have to ask for heap allocations. And just like in C++ you have types in Rust that look like stack allocations but aren't, like Vec.

So no, it's just value type otherwise Box<T>. That's not true in the slightest. And you need to specify in your function if it's taking ownership or not in addition to whether or not it's pass-by-value or pass-by-reference.

Other than new/delete it's the same shit, just a different syntax.

Rust got to break backwards compatibility which allowed it to remove a lot of C/C++'s old mistakes, yes, but the raw complexity is still entirely there, and very nearly identical.

> That's not simpler at all - it's basically the same

Oh, you mean like how you have rvalues, lvalues, xvalues, glvalues and prvalues[1]. In Rust you have value, ref and mut ref.

You also don't specify ownership of passed pointers. So you need to encode that into your function documentation, as opposed to Rust where only value types can change ownership.

> And just like in C++ you have types in Rust that look like stack allocations but aren't, like Vec.

Never said anything about that, we're just talking about parameter types here, of course you can have value types take ownership of stack allocations(Box<T>, std::unique_ptr and their variants).

> And you need to specify in your function if it's taking ownership or not in addition to whether or not it's pass-by-value or pass-by-reference.

Not in the case of Rust, value types are the only thing that can take ownership(even Arc/Rc return value types when you clone()). Refs just allow temporary access in a mutable/const way for the lifecycle of the function.

> Rust got to break backwards compatibility which allowed it to remove a lot of C/C++'s old mistakes

Thanks for proving my point. I've never seen a C++ codebase that didn't use some of C++'s mistakes.

[1] https://stackoverflow.com/questions/3601602/what-are-rvalues...

> Oh, you mean like how you have rvalues, lvalues, xvalues, glvalues and prvalues[1]. In Rust you have value, ref and mut ref.

That is not a reasonable comparison and you know it. rvalues & lvalues are the only things in C++ that you ever care about in the slightest unless you're writing a compiler, and if you're writing a compiler then Rust has things like xvalues, too (how do you think the Drop trait works?)

Answer me a simple question, why does std::move exist then?

(here's a hint, Rust doesn't need xvlaues because every value already has an associated lifetime)

I believe the only thing that can be said with certainty about Rust when compared to C++ is that it's safer.

Simplicity in the sense of easy to understand is an adjective I would apply to languages such as golang, Swift or even Java. C++ and Rust are complicated languages, because they allow low-level control of the implementation and they insist on the programmer thinking deeply about resource management.

Elegance is likewise a property I would not associate with Rust: resource management is clumsy and leaks into the syntax everywhere, it ends up defining how the language looks like.

> No other language I'm aware of that delivers this, apart from maybe Rust or D.

Reminds me of a great campaign speech from Donald Trump where he praises the border wall, saying there will be no way anyone could get down once they managed to climb it. Except maybe a rope. Good job proving your point by shooting yourself in the foot.

This is the paradox: Every new language feature in C++ has, as you say, a rational argument for inclusion. Yet when you combine them, you get a language that I will never use if I have a choice.

C++ has very little "magic" once you get an okay mental-model for how the compiler looks at things. And in fact you can write okay but sub-optimal/"unsafe" C++ that looks kinda like C without that much headache and without using things like unique_ptr (or even really refs to a certain extent).

However C++ does let you get as low-level as you want and micro-optimize things and write extremely precise types that other languages can really only dream about.

This said: the language is quite complicated which leads to it being "hard" to compile (and hard to learn if you care about some of the more interesting aspects).

> C++ developers are masochists

In my experience you have to have a very different mindset when writing production-quality C++. It's a lot "slower" to write (at least for me), but in the end you can end up writing some extremely expressive and efficient code that scripting languages can't even express very well (let alone run quickly).

(The C++ tooling around build systems and compilers and linkers / arcane black-magic is a real tragedy. I'm holding out for a sane C++ build-and-module system but the language-designers don't seem interested in standardizing these things.)

> Show me simplicity and elegance. Show me boring but powerful.

Many languages value these as design-goals and you may want to use those languages instead. C++ makes a different set of tradeoffs. It has its own sense of elegance.

This is a very romantic view of C++, but not (in my opinion) an accurate one. C++ feels like a language developed without a road map. I agree with the original comment. Many C++ updates are full of band aids to address issues so blatant that they had to fixed by others a long time ago (see boost).

> C++ has very little "magic" once you get an okay mental-model for how the compiler looks at things.

This is a meaningless statement. I could make the worlds most obtuse and complicated language and you could still say this. Just because after years, and I do mean years, of experience C++ stops regularly blindsiding you does not mean it is a well designed language.

There are parts of C++ that are good, beautiful even, but that does not give it a free pass from being among the messiest languages in widespread use.

I don't give C++ a free-pass for its warts, I just think it's important to realize where those warts actually live so that I can direct frustration appropriately :)

> C++ feels like a language developed without a road map.

Many of the warts are there due to the committee's very strong stance on requiring backward-compatibility. This does detract from the language sure, but the tradeoff is code that will continue to compile for many years which is important for huge codebases that want to continue to adopt the latest features.

Other languages tend to solve this with breaking-changes (python3, perl6, even java8 or ES* to some extent which arguably changed the whole paradigms of their languages).

Ultimately the roadmap is dictated by time-tested experience with existing features and what the community deems the most necessary and practical (see boost which is in many ways an incubator for language ideas).

Recent releases are explicitly around new roadmap items like better RAII-pointers and (finally) a standard threading API.

There are also new language-features that make the language easier (lambdas, more auto, etc) and faster (rvals/move operations). These are "optional" and can be ignored if you prefer.

> Just because ... C++ stops regularly blindsiding you does not mean it is a well designed language.

I'm still blind-sided when the compiler does crazy things I hadn't expected, but almost always I end up cursing the tooling not the language.

IMHO this is the fault of (1) the documentation being terse and separated from stdlib headers etc, (2) compilers/linkers being rather nasty and slow to work with, and (3) terrible error messages in general.

These are problems with the ecosystem and not so much with the language-proper. It's fair to say they sour the whole experience (and they really really do sour it), but the language itself is actually not that surprising once you get past the tooling.

> C++ developers .. don't seem to be interested in actually improving the language

In his talks at the 2014 and 2017 conferences for the D programming langue, Scott Meyers noted the difference between what individual C++ committee members say they value, and what the process actually produces. Will a new standard every three years make this better or worse? They do seem open to deprecating things that don't work.

You might enjoy the 2014 talk, called "The Last Thing D Needs." I don't want to spoil the punchline, which actually struck me as kind of bittersweet, but Scott talked about the distinction that Fred Brooks made in "No Silver Bullet" between essential and accidental complexity. C++ does have a lot of the latter.

Edit: link for the 2014 talk:

https://www.youtube.com/watch?v=KAWA1DuvCnQ

and for the 2017 talk:

https://www.youtube.com/watch?v=RT46MpK39rQ

> Show me how I don't have to choose between value, reference, raw pointer, unique_ptr, shared_ptr, etc, to pass a damn parameter to your function.

This really sounds like your problem isn't with C++, it's with, well, any language that's "low-level" enough to distinguish between values and references.

I grudgingly used C++ for years because at least it was better than C, but since C++11 it's actually been a great experience. Honestly my only complaint about modern C++ development is that the compile times are painfully slow. There are quirks, but overall it's now a wonderfully usable language.

Though obviously you shouldn't bother with C++ if you're working on stuff that could just as easily be written in Python or whatever.

I am working on a C++ project and I enjoy it.

The bread and butter of C++ is balancing relatively good performance with reasonably acceptable overhead and developer time.

I was personally tired of re-writing over and over.

    template <typename Type>
    struct PerformOperation {
      static void
      Op(int&) {} // noop
    };

    template <>
    struct PerformOperation<float> {
      static void
      Op(int& k) {
        // do something
      }
    };

And all of the specific things you're complaining about... well they're in your control. Do you or your company hold a standard for interfaces or memory tenancy? If not, maybe you need one.

I am a C++ dev and I kind of enjoy it. There is a lot of hate towards C++ and a good part of it is justified. Yet what vindicates the language in my opinion is that some of the most complex and interesting software in the world is written in C++ (browsers, search engines, machine learning engines, database management systems, game engines, you name it). So if you want to work on those you'd better know the difference between a reference and a shared_ptr.

As for the masochism... Well, after some time you stop noticing the warts and just work around them without expending much mental energy. Not a big deal.

I usually work in Java and .NET.

Occasionally I drop down to C++, either on private projects, or because I need to integrate some library on the former eco-systems.

The only thing I don't enjoy in C++ is the copy-paste compatibility that makes some devs use it as "C with C++" compiler, not even "C with Classes".

For devs that dislike magic there is always Go.

Agreed completely, speaking as another one whose worked on C++ projects and tried very hard to stop others on the team from cramming as many new features as they could into code they wrote just for the sake of using them. It's a similar attitude to "architecture astronautism", some people just seem to love creating complexity while others (like me) enjoy the complete opposite. IMHO C++ is good when used as a thin layer of syntactic sugar over C, the "C with classes" that so many C++ fans speak derogatorily of.

Then again, my language of choice for general use is C89.

C89 with -pedantic ,-Wall, and -O2 flags, I hope! O2, of course, not for the speed but or the small number of additional warnings it might generate.

There is no real reason not to be using C99 (or C11) unless you work with targets that only have an "ANSI" compiler. The additions that come with each new standard are either sorely needed (stdint.h, stdatomic.h, etc.) or conservative to a fault (_Generic) so you can't even complain about bloat.

Down with "ANSI" C fetishism.

Bad programmers don't like C++ because it often makes them feel stupid. Good programmers like C++ because it sometimes makes them feel stupid.

constexpr in this case seems like a nice and less magic replacement for SFINAE, which always seemed like a bit of a hack