I write mostly like I would in C, but use C++ features as needed. It ends up looking similar to Rust if you squint. All these "I write games in C" people complain about C++ features, and then end up reimplementing virtual interfaces manually with struct headers or massive switch statements, just to feel better about themselves. Writing games in C is not harder, you just have to implement modern language features by hand.
Complaining about a language having features you don't want is silly. C++ doesn't take longer to compile if you don't abuse templates.
> Complaining about a language having features you don't want is silly.
It might be silly if you're working on your own. Software that delivers a lot of value is usually developed and evolved not only by team, but by a team with changing members and changing leadership over the project's lifetime. The features used will be the union of all features used over the years, and while it's easy for team leads to allow the use of more features than their predecessors, it's quite hard to reduce them.
Also, you may be forced to use language features you don't want if they're used by libraries whose functionality you do want. For example, when doing low-level programming, I don't like implicit calls that I can't clearly see on the page (e.g. destructors or overloaded operators). But if libraries I want use them, then I'll have those implicit calls. But if the language doesn't have those features, libraries obviously won't use them.
> It might be silly if you're working on your own.
That's exactly the case when it's easiest. If you don't need a feature, just don't use it and case closed. With a team it's harder - you have to force/enforce others not to use a given feature.
> if they're used by libraries whose functionality you do want
If you're using C++ you can just use the C library you would've used otherwise, no?
> Also, you may be forced to use language features you don't
This is quite important and often overlooked. An annoying fallacy is that people think some features are optional, but once they get heavily used, they aren't anymore. Often they quickly become an requirement and if you don't follow suit, your code is legacy or you are an idiot. But well, I guess I create legacy code upfront and the cargo cultists create modern code that turns into legacy next day.
> C++ doesn't take longer to compile if you don't abuse templates.
It actually does though, unless you also drop C++ stdlib usage completely (have you looked at how many lines of code just <vector> alone pulls into each source file? - it's upward of 20kloc and growing with each new C++ version).
And at that point you get into discussions with various C++ camps about why you don't use the C++ stdlib and instead prefer to reinvent the wheel (and this friction with other C++ coders is the main problem of carving out your own subset - it works ok in complete isolation, but software development work hardly happens in splendid isolation and even then you'd might to want to use C++ libraries written by other people from time to time...)
And once you've been dragged into such C++-subset-discussion month after month, year after year, at that point it is much less exhausting to just write plain C. And the C community (if it can be called that) seems to be much less concerned about coding style dogma and generally a nicer bunch to interact with.
FWIW, I switched around 2017 and each time I have to interact with a C++ library for lack of alternatives it's usually not a pleasant experience (with the notable exception of Dear ImGui - but even there I started to prefer the C bindings so that I don't need to strictly separate the UI code from the rest of the code base, which sometimes makes sense, but often not, especially with an immediate mode UI framework).
> then end up reimplementing virtual interfaces manually
C++ dynamic dispatch (your "virtual interfaces") is achieved by welding a vtable onto every type and providing a pointer to that vtable for instances of the type. If in 90% of your code you deal with specific types like Goose or Swan or Duck or Seagull, and only 10% needs to work with the broad Bird category well, too bad, every Goose, Swan, Duck and Seagull carries around that vtable pointer even if it goes nowhere near that 10% of the system. This way your Bird code "just works" in C++.
That's not the only way to crack this nut. Idiomatic Rust approach uses vtables only in the Bird code, elsewhere they don't exist, and thus don't take up space in a Duck or whatever that's always a Duck, but in exchange now you're spending more time thinking, because by default there aren't any vtables and so dynamic dispatch isn't possible at all.
So while that C programmer has to implement features by hand, they are at least able to specifically implement the feature they wanted, not whatever was easiest for Bjarne Stroustrup last century.
C++ nudges you to think in terms of single elements. Operator overloading, ctors/dtors, references, etc. you pay that cost all over the place.
C programs tend to nudge you into thinking in terms of arrays of data.
For game development you generally want to think this way. The cost of vtables and all the cache misses doesn’t have to be paid. A game has to stream bytes. Many things at once. Rarely single elements at a time.
I agree, there is a nice language buried in all the features trying to escape.
The struggle is that if you choose a subset of the language to use and you have dependencies… well you’re including whatever features they use into your subset.
I don’t think C++ is a bad language at all or anything. Just that it nudges developers into certain modes of thinking based on the features it chooses to focus on. Might explain why some developers still choose C.
The context you snipped is about dynamic dispatch, that is what you - as someone who has "learned C++ properly" apparently call "classes with at least one virtual method" and indeed the earlier comment calls "virtual interfaces".
The convenience of having regular generics and dyn generics handled automatically is a great feature of Rust, sure, however you can write a template in C++ that directly calls a method, f.e. obj.Object::method(), which skips the vtable, achieving the same thing. Or you can keep manually writing everything in C because you refuse to learn C++.
Except that C++ provides the tools to do just like C, Rust, or whatever one feels like doing for dispatching, even if it requires a few pages of template metaprogramming mixed with compile time executions, or writing exactly the same C code on the common subset across both languages.
Now with reflection even more tools will be available.
Which is why despite all its warts and security flaws, many inherited from C source code compatibility, many domains will keep using it, because they will complain about their missing 1% that no one else uses.
Because these are General Purpose languages you can do the same things, but the contrast here is what's provided in the box and how it is used idiomatically, because in practice that's what gets used, and that's what I explained above.
You can write C++ style OOP hierarchy code in Rust but that's not idiomatic, and you can write Rust style explicit dynamic dispatch in C++ but again it isn't idiomatic.
For great examples of the above, see the classic Scientific and Engineering C++: An Introduction with Advanced Techniques and Examples by Barton & Nackman (1994).
What you're describing is needing to use specific verbose patterns to opt out of the defaults that do more complex things under the hood, whereas they're describing how C and Rust do not do those things by default and instead let you opt into them. It's not disingenuous to point that out.
What i am pointing out are neither complex nor any magic under-the-hood. They are simply techniques in the C++ repertoire well-known since the early 90's and used in all high-performance C++ libraries.
tialaramex made a big deal out of overheads in C++ dynamic dispatch (which incidentally are pretty minimal) using a trivial example, when performance focused (both time and size) C++ programmers do not use it that way at all. Modeling in C++ can be done in any number of ways and is driven by specific perspectives of the end goal.
> C++ doesn't take longer to compile if you don't abuse templates.
Surprisingly, this is not true. I've written a C++ file only to realize at the end that I did not use any C++ features. Renaming the file to .c halved the compilation time.
On some compiler toolchains (IIRC MSVC was the main offender) you get a lot more code pulled into your source file when including a C stdlib header (like <stdio.h>) in C++ mode versus C mode. Basically a couple hundred lines in C mode versus thousands of lines in C++ mode.
I agree it shouldn't really matter if there's no C++ features in play, but I suppose third party headers could bite you if they use #ifdef __cplusplus to guard optional C++ extensions on top of their basic C interface. In that case the compiler could be dealing with dramatically more complex code when you build in C++ mode.
Maybe it is similar for the same compiler (but one should check, I suspect C could still be faster), but then there are much more C compilers. For example, TCC is a lot faster than GCC.
tcc is 8x faster, twice as fast isn't doing it justice.
As for the header thing, that'd could potentially be true if the compile time was something like 450ms -> 220ms, but why bother saying it when you're only saving a few hundred milliseconds
Going from 220 to 450 ms would be a disaster in my project. It has many thousands of files. Recompilation of almost everything happens from time to time.
If those made-up numbers were true, they would be very significant and an argument in favor of keeping the code in C
A 200ms difference is adding or removing 200lines lines of implementation, and spliting it up into a file can make it slower because of include overhead. You completely made up C being twice as fast as C++.
I measured once and to my surprise templates aren't (directly) the reason for long compile times. It's function bodies in headers, and obviously templates are in headers and they call other templated functions/classes which explodes code generation and time. But if it's only a few lines and doesn't call other templated functions it's likely fine. I wrote about it here https://bolinlang.com/wheres-my-compile-time
After writing that, I wrote my own standard library (it has data structs like vector, hashmap and sets; slices, strings, rng, print, some io functions, and more) which uses a lot of templates, and it compiles in <200ms on both clang and gcc. Many standard library headers take much longer to compile than that. It's not a terrible idea to have your own standard lib if you need quick compile times.
Oh? I use a boring old $5 VPS (using nginx for https). There's no cloudflare or anything else involved. I wonder why that's happening. I may look into that this week
I’ve seen this play out a lot. People say they “write games in C” and then quietly rebuild half of C++ anyway with vtables in structs or giant switch statements, just without the compiler helping. That’s fine if it makes you happier, but it’s not obviously simpler or safer. Also, C++ compile times are mostly a self-inflicted wound via templates and metaprogramming, not some inherent tax you pay for having virtual functions.
A switch statement is how you do ad-hoc polymorphism in C -- i dont thinks an own against C developers to point that out. If they wanted to adopt the C++ style that immediately requires the entire machinery of OOP, which is an incredibly heavy price to avoid a few switch statements in the tiny number of places ad-hoc poly is actually needed
You don't usually do C++ subsets if you want the full shebang.
I have a "mini-std" headerfile that's about 500 LoC implementing lightweight variants of std::vector, std::function, a stack-local std::function (unsafe as hell and useful as hell to avoid allocations), a shared-ptr, qsort and some other nifty stuff.
That does a lot of things, but even then I use other patterns that brings a lot of bang for the buck without having to go full C (hint: the stack-local function equivalent gets a lot of mileage).
C++ reimplements a lot of the things we do in C with function pointers, while hiding what's actually happening behind topheavy syntax that implies a 1990s object oriented paradigm that's dead now.
Contrary to popular belief C++ isn't really object–oriented. I mean, you can write object–oriented code, but the language doesn't make assumptions about what goes into a class. It's really just a struct with associated functions.
> It's really just a struct with associated functions.
If that were actually true C++ would be a lot easier to accept as a 'C successor'. Instead you have the implicit this-pointer, complicated rules for which constructor or operator overload is actually called, a hidden vtable pointer (not to mention multiple inheritance), then you have public/private/protected, override vs final, const methods (which wouldn't be needed as a separate syntax feature if the this arg wouldn't be implicit) etc etc etc... that a lot of OOP-baggage which a lot of C++ coders probably don't even notice anymore.
A plain C struct with function pointers does indeed make a lot more sense than all the OOP-ism that C++ hides from you ;)
It has lots of optional features, none of which make it an object–oriented language. It's an everything–oriented language. It has lambda syntax for functors, but you're not calling it an impure–functional programming language. Virtual functions are one feature, which is useful in pseudo–object–oriented code.
> Complaining about a language having features you don't want is silly.
If your criteria for a good language is "how many features does it have", then sure, C++ wins. OTOH, if you criteria is "How many footguns does the language have" then C++ loses to almost every other mainstream language, which includes C.
Surely your criteria should be some combination of the two (plus other factors). C may have fewer footguns than C++, but it still has many, whilst also lacking many useful features
> Surely your criteria should be some combination of the two (plus other factors).
Sure, but the weighting would be different for different people.
> C may have fewer footguns than C++, but it still has many, whilst also lacking many useful features
We are not talking "2 fewer footguns", or "5 fewer footguns"; we are talking "dozens fewer footguns".
When I need a language with more features than C, I don't choose C++, because the choice is not "Use C for simplicity, and use C++ to trade simplicity off against features", it's usually "Use C for simplicity, and use Java/C#/Go/Rust for features".
Of course people do" virtual functions" in C, but I think this is not an argument despite C.
I noticed that making virtual in C++ is sooo easy that people start abusing it. This making reading/understanding/debugging code much harder (especially if they mess this up with templates).
And here C is a way - it allow but complicates "virtual". So, you will think twice before using it
Yeah, you could argue that choosing C is just choosing a particular subset of C++.
The main difference from choosing a different subset, e.g. “Google C++” (i.e. writing C++ according to the Google style guide), is that the compiler enforces that you stick to the subset.
When I developed D, a major priority was string handling. I was inspired by Basic, which had very straightforward, natural strings. The goal was to be as good as Basic strings.
And it wasn't hard to achieve. The idea was to use length delimited strings rather than 0 terminated. This meant that slices of strings being strings is a superpower. No more did one have to constantly allocate memory for a slice, and then keep track of that memory.
Length-delimited also super speeded string manipulation. One no longer had to scan a string to find its length. This is a big deal for memory caching.
Static strings are length delimited too, but also have a 0 at the end, which makes it easy to pass string literals to C functions like printf. And, of course, you can append a 0 to a string anytime.
I agree on the former two (std::string and smart pointers) because they can't be nicely implemented without some help from the language itself.
The latter two (hash maps and vectors), though, are just compound data types that can be built on top of standard C. All it would need is to agree on a new common library, more modern than the one designed in the 70s.
I think a vec is important for the same reason a string is… because being able to properly get the length, and standardized ways to push/pop from them that don’t require manual bounds checking and calls to realloc.
Hash maps are mostly only important because everyone ought to standardize on a way of hashing keys.
But I suppose they can both be “bring your own”… to me it’s more that these types are so fundamental and so “table stakes” that having one base implementation of them guaranteed by the language’s standard lib is important.
You can surely create a std::string-like type in C, call it "newstring", and write functions that accept and return newstrings, and re-implement the whole standard library to work with newstrings, from printf() onwards. But you'll never have the comfort of newstring literals. The nice syntax with quotes is tied to zero-terminated strings. Of course you can litter your code with preprocessor macros, but it's inelegant and brittle.
Because C wants to run on bare metal, an allocating type like C++ std::string (or Rust's String) isn't affordable for what you mean here.
I think you want the string slice reference type, what C++ called std::string_view and Rust calls &str. This type is just two facts about some text, where it is in memory and how long it is (or equivalently where it ends, storing the length is often in practice slightly faster in real machines so if you're making a new one do that)
In C++ this is maybe non-obvious because it took until 2020 for C++ to get this type - WG21 are crazy, but this is the type you actually want as a fundamental, not an allocating type like std::string.
Alternatively, if you're not yet ready to accept that all text should use UTF-8 encoding, -- and maybe C isn't ready for that yet - you don't want this type you just want byte slice references, Rust's &[u8] or C++ std::span<char>
Automatic memory accounting — construct/copy/destruct. You can't abstract these in C. You always have to call i_copied_the_string(&string) after copying the string and you always have to call the_string_is_out_of_scope_now(&string) just before it goes out of scope
For many string operations such as appending, inserting, overwriting etc. the memory management can be made automatic as well in C, and I think this is the main advantage. Just automatic free at scope end does not work (without extensions).
You can make strings (or bignums or matrices) more convenient than the C default but you can never make them as convenient as ints, while in C++ you can.
Yes, but I do not think this is a good thing. A programming language has to fulfill many requirements, and convenience for the programmer is not the most important.
The C++ std::string is both very complicated mechanically and underspecified, which is why Raymond Chen's article about std::string has to explain three different types (one for each of the three popular C++ stdlib implementations) and still got some details wrong resulting in a cycle of corrections.
So that wouldn't really fit C very well and I'd suggest that Rust's String, which is essentially just Vec<u8> plus a promise that this is a UTF-8 encoded string, is closer.
It is when compared with C89, also the ISO C++ requires inclusion of ISO C standard library.
The differences are the usual that occur with guest languages, in this case the origin being UNIX and C at Bell Labs, eventually each platform goes its own merry way and compatibility slowly falls apart with newer versions.
In regards to C89 the main differences are struct and unions naming rules, () means void instead of anything goes, ?: precedent rules, implicit casts scenarios are reduced like from void pointers.
For every person who says on the internet that you can just use a C++ subset, there's another who insists that C is the bad C++ subset. So compiling C code with a C++ compiler promotes your code from "good C code" to "bad C++ code" (most C code isn't "exception safe," for example).
It's arguably irrational to evaluate a language based on this, but you can think of "this code could be better" as a sort of mild distraction. C++ is chock full of this kind of distraction.
I feel like, for me, it’s that I am more familiar with writing in C and switching to C++ seems rather difficult. So, sure I am reimplementing features that already exist in anoter language, it just so happens in this case is C++. Why not use python if you want to avoid reimplementing the wheel as much as possible. And sure python is not suited for game development but I just wanted to make a point with it. I think in the end ising a language you are most familiar with results in the most amount of enjoyable coding.
For a solo dev, it's not difficult. C++ is nearly a superset of C. You don't have to adopt all of C++ to start using it and to get immediate benefits from it (for example, unique_ptr, shared_ptr, and vector would all be things that I think any C dev would really appreciate).
A reason I can think of to not move to C++ is that it is a vast language and, if you are working on a team, it can be easy for team members ultimately forcing the whole team to become an expert in C++ simply because they all will be familiar with a different set of C++ features.
But for a solo dev? No reason not to use it, IMO. It's got a much nicer standard library with a rich set of datastructures that just make it easier to write correct code even if you keep a C style for everything.
Exactly, not even do you need to religiously need stick to your subset, separate modules can be using supetsets that import useful libraries and if they're used for code that is seldomly changed (such as model importers) then the longer compile time will only matter for rebuilds and not quick tests.
It's possible to use only a subset of the language. You could write a Java program without classes if you really wanted to. Just put the whole thing in main().
A lot of smart people pick and choose what they want from the language, just like religion, they keep the good parts and discard the bad.
That's probably to do with exceptions — possibly the only thing that pervades C++ code even if you don't use it. The compiler has to write code so an exception at any point leaves the stack in a sensible way, etc. Try -fno-exceptions (and -fno-rtti might save some memory while you're at it)
Regrettably not every C++ feature is free if you don't use it. But there aren't many that aren't.
Complaining about a language having features you don't want is silly. C++ doesn't take longer to compile if you don't abuse templates.