For kernel dev it would be a disaster, there’s too much implicit action, and abstractions that have unknown runtime cost. The classic answer is that everyone uses 10% of its features over C, but nobody can agree on which 10%.
As someone forced to get up to date with C++ recently, at this point it’s a language in full identity crisis. It wants so badly to be Rust, but it’s got decades of baggage it’s dragging along.
I write C++, but I don’t like it. I don’t think any one should be claiming Linus doesn’t like it because he doesn’t know it. If he wants to contribute to this project, it must be C++.
I don’t think any one should be claiming Linus doesn’t like it because he doesn’t know it
I didn’t think anyone was making that claim? Either way I’m certainly not trying to.
If he wants to contribute to this project, it must be C++.
Yes, this is true, it’s just that he was also included in the decision making process to switch to it in the first place, and I feel like his continued use of it makes me think it doesn’t have to be as awful as everyone makes it out to be, or he wouldn’t use it at all.
Yea but all that function pointer indirection can actually hurt performance (especially caching), some things in C++ actually can be faster just because the compiler is better at optimizing for that.
Technically you’re right, but I don’t think that changes what I said about optimization. There are still cases where equivalent C++ code can be faster than the C version merely due to different optimizations used.
Meh, I’m unconvinced. If it’s any kind of hot spot, in either, you can optimize the hell out of it. C++ is often more bloated is it’s just a harder language pretending to be an easier one.
That’s fine, you don’t have to agree. Personally I do like to use just a few features of C++ without going too crazy, like simple classes and maybe one level of inheritance, but I don’t really get into templates or exceptions or other really complex/controversial stuff. I prefer having the stronger typing and better readability of this kind of C++, and I think it helps me make less mistakes, but I realize not everyone agrees, and that’s ok.
I think that’s the thing, C++ is so broad. It’s like many languages together. It’s complex with lots of implicitness yet unsafe. There is loads of support in compilers and tools to mitigate that, but that’s treatment not cure.
I think the same could be said about C now too, it is continuing to evolve itself with newer standards too just like C++. People choose to only use C features that they want, same goes for C++.
And that is true of any language, but C++ is without doubt one of the broadest. There are very different ways of working with it that compile very differently.
Yes. Too much for people to learn, so they make mistakes. I’ve seen compilers get confused with C++. Though it was MS’s…
If a language isn’t tight, it should at least be safe. C++ is neither. You can do anything with it, but I don’t think it’s ever really a good tool for a particular job.
I feel the OOP debate got a bit out of hand. I hate OOP as well, as a paradigm.
But I love objects. An object is just a struct that can perform operations on itself. It’s super useful. So many problems lend themselves to the use of objects.
I’ve been writing a mix of C and C++ for so long I don’t even know where the line is supposed to be. It’s “C with objects”. I probably use only 1% of the functionality of C++, but that 1% is a huge upgrade from bare C IMO.
I think the problem with OOP is something you can see whenever legislation is linked with prestige (it happens a lot in real life). The number of good possible rules is quite limited, and the number of people who want to make a name for themselves by championing them seems to be infinite. If you can’t find a good rule to claim as your own, you have to pick a bad rule and try to gaslight people into thinking it’s a necessary and beneficial. Enough people do that, and we end up with modern OOP.
Agreed. Objects are nice and a great way to program. Composition is great. Traits/interfaces are great. Namespaces are great. Objects are a really nice way to reap the benefits of principles like these.
But then there are aspects of OOP that absolutely suck, like inheritance. I hate inheritance. The rules get very confusing very quickly. For example, try understanding overriding of methods. Do I need to call the superclass method or not? If not, does it get called automatically? If so, in what order? How do these rules change for the constructor? Now repeat this exercise for every OOP language you use and try not to mix them up… Java, C++, Python, etc.
Fortunately, it feels like we rely on inheritance less and less these days. As an example, I really like how Java allows you to implement Runnable these days. Before, if you wanted to run a thread, you needed a separate object that inherited Thread. And what if that object needs to inherit from another one too? Things would get out of hand quickly. (This is a very old example, but with lambdas and other new features, things are getting even better now.)
Anyway, long story short, I think OOP is a complicated way to achieve good principles, and there are simpler ways to achieve those principles than a full OOP implementation.
I’ve seen this thing where people dislike inheritance a lot, and I have to admit that I kind of struggle with seeing the issue when it’s used appropriately. I write a bunch of models that all share a large amount of core functionality, so of course I write an abstract base class in which a couple methods are overridden by derived models. I think it’s beautiful in the way that I can say “This model will do X, Y, Z, as long as there exists an implementation of methods A, B, C, which have these signatures”, then I can inherit that base class and implement A, B, and C for a bunch of different cases. In short, I think it’s a very useful way to express the purpose of the code, without focusing on the implementation of specific details, and a very natural way of expressing that two classes are closely related models, with the same functionality, as expressed by the base class.
I honestly have a hard time seeing how not using inheritance would make such a code base cleaner, but please tell me, I would love to learn.
What you’re describing is an interface. An interface is a contract that ensures you can do something, but doesn’t care how.
Abstract classes can have abstract functions. When you do this, you’re basically just creating a base class with an interface on top; you’re saying “all my children must implement this interface of mine” without having to actually make a separate interface.
Abstract classes also offer additional functionality though, such as the ability to define properties, and default implementations of methods. You can even utilize the base class implementation of the method in your child class, in order to perform extra steps or format your input before you do whatever it is you were doing in the first place.
So, an interface is a contract that allows you to call a method, without having to know the specific class or implementation.
Inheritance is more like “it does everything that X does, but it also does Y and Z.” If you’re ever finding yourself writing an abstract class with purely abstract methods, you probably want to write an interface instead. That way, you get all the same functionality, but it’s more loosely coupled
Epecially when you think in “real” OOP terms:
Abstract classes are “child is a parent”, fx “duck is a bird”. Bird describes all the traits that all birds have in common. But not all birds fly, so flight must come from an interface. This interface can be passed around to any number of objects, and they’re not as tightly coupled because unlike an abstract class, an interface doesnt imply that “duck is a flight”. The interface is just something we know the duck can do.
As you can probably tell, I work with OOP on a daily basis and have for years. There are a lot of valid criticisms of the OOP philosophy, and I have heard a lot of good points for the record. I am just educating on the OOP principles because you said you were interested and to clear up any misconceptions.
Well yes, I get the differerence between an interface and a class, and what I write is typically a class, which contains properties and functionality that may or may not be overridden in derived classes.
For example, calling a parent class implementation can be useful when I have a derived model that needs to validate its input in some specific way, but otherwise does the same as the base class.
What I don’t understand is why this makes OOP bad?
Torvalds just really dislikes C++. He’s gone on the record saying that he thinks it’s just not a good language. In his own words “C++ is just a waste, there is no design at all, just adding some scum on top of C.”
The STD is maybe the only good thing C++ has over C, and even that is awful compared to other language’s standard libraries.
I can’t name another good thing C++ has. Maybe templates. C++’s reliance on inheritance for polymorphism is awful (should’ve gone with interfaces/traits).
Not to mention the mess with all the different types of constructors that must always be implemented.
It’s just a bunch of bad design choices added on top of an old outdated language.
I would have agreed with that before C++11. But since then, C++ has improved a lot. Its like the vision of what C++ suddenly became more clear. So I wonder if Linus would still say that today. (Unfortunately, there have been a lot of missteps in the development of C++ though, and so there is a lot of cruft that everyone wishes was not there…)
Lots of core UNIX and Linux projects are. C++ is not liked by a lot of low level FOSS community. I think Rust is going to get further into these areas. I know C++ well but prefer C. I know plenty of others who feel the same.
In that post, his critiques were around the problems with the STL and everyone using Boost. The STL has improved significantly since then, and it would be a limited subset of c++ if it was ever allowed
There have been mailing list conversations earlier this year, citing that clang/gcc now allowing c++ in their own code might mean they’ve taken care of the issues that made it unusable for kernel code
I’m not saying it will happen, but it’s not being shot down as an absolute insanity anymore, and I wouldn’t have expected Rust to be allowed in the kernel, either
Oh interesting. I didn’t realize boost was the main issue. Most people I’ve talked to were complaining about VTables introducing a bunch of indirection and people blindly using associative containers.
Vtable equivalents are used extensively in the kernel
You’ll find structs all over the place setting them up, e.g. every driver sets up a .probe function that the core will call, since it doesn’t know what driver it’s loading
Right the issue was more because they’re so easy to throw in without thinking about it so people overuse them. That may just be older devs complaining about newbies though.
That’s my guess, but there was a conversation on the mailing list a few months ago that wasn’t just immediately shut down, even by other prolific developers
Ts’o seems skeptical, but is at least asking whether c++ has improved
Take a look at what even the proposer is saying wouldn’t be allowed in:
<span style="color:#323232;"> (1) new and delete. There's no way to pass GFP_* flags in.
</span><span style="color:#323232;">
</span><span style="color:#323232;"> (2) Constructors and destructors. Nests of implicit code makes the code less
</span><span style="color:#323232;"> obvious, and the replacement of static initialisation with constructor
</span><span style="color:#323232;"> calls would make the code size larger.
</span><span style="color:#323232;">
</span><span style="color:#323232;"> (3) Exceptions and RTTI. RTTI would bulk the kernel up too much and
</span><span style="color:#323232;"> exception handling is limited without it, and since destructors are not
</span><span style="color:#323232;"> allowed, you still have to manually clean up after an error.
</span><span style="color:#323232;">
</span><span style="color:#323232;"> (4) Operator overloading (except in special cases).
</span><span style="color:#323232;">
</span><span style="color:#323232;"> (5) Function overloading (except in special inline cases).
</span><span style="color:#323232;">
</span><span style="color:#323232;"> (6) STL (though some type trait bits are needed to replace __builtins that
</span><span style="color:#323232;"> don't exist in g++).
</span><span style="color:#323232;">
</span><span style="color:#323232;"> (7) 'class', 'private', 'namespace'.
</span><span style="color:#323232;">
</span><span style="color:#323232;"> (8) 'virtual'. Don't want virtual base classes, though virtual function
</span><span style="color:#323232;"> tables might make operations tables more efficient.
</span>
C++ without class, constructors, destructors, most overloading and the STL? Wow.
I’ve only worked on a few embedded systems where C++ was even an option, but they allowed 2, 4, 5, and 7. Though, for the most part most classes were simple interfaces to some sort of SPI/I2C/CAN/EtherCAT device, most of which were singletons.
”C++ is a horrible language. It’s made more horrible by the fact that a lot of substandard programmers use it, to the point where it’s much much easier to generate total and utter crap with it. Quite frankly, even if the choice of C were to do nothing but keep the C++ programmers out, that in itself would be a huge reason to use C.”
”C++ is a horrible language. It’s made more horrible by the fact that a lot of substandard programmers use it, to the point where it’s much much easier to generate total and utter crap with it. Quite frankly, even if the choice of C were to do nothing but keep the C++ programmers out, that in itself would be a huge reason to use C.”
Add comment