5 ms·

A 2024 Discussion Whether to Convert the Linux Kernel from C to Modern C++

34 points by ashvardanian 3y ago

karmakaze 3y ago
I don't normally get into C/C++ or kernel chats, but I hope they choose something different than C++, maybe Zig. The problem with C++ is that like Scala, it's hard to stick to a subset without realizing you've used more than intended after the fact. Maybe the kernel team could do it since there's a culture of keeping things clean/simple.
- kreetx 3y ago
  I think they'll likely agree on some C++ subset when getting into it so I wouldn't be worried. Though, why not just evolve C itself with those features? C is evolving, too, and this could be used as a vehicle for it.
- teleforce 3y ago
  Perhaps they should just jump straight to D. The proposed C++ 20 is like D-- given how many D features have been adopted over the years. Now that D supports C native compilation and supported by GCC eco-system so the conversion can be done in peacemeal, it will be win-win situation for safety, convenience, power and flexibility. One of the main reasons of switching to C++ given is for templates, and D templating system is second to none. D also run and compile as fast as C, and the latter cannot be said to C++. For kernel developers fast compilation will be necessary not a luxury. Imagine kernel development in D without GC and user level development with D with the default GC, it really looks like a match made in heaven to me.
jerf 3y ago
Modern C++ has addressed none of the reasons Linus originally rejected it. Those reasons were widely misunderstood, but I tried to explain them at the time and it seems to have helped at least some people: https://news.ycombinator.com/item?id=1422160 https://news.ycombinator.com/item?id=1422160 Given the original objections, it should only be even more objectionable now as C++ has added even more subtleties and even more you have to understand in order to fully grok what a given bit of C++ code is doing. I'd also disagree with the idea that it's lower effort to add C++ than to use Rust. C and C++ have diverged wildly. Knowing C means you can mostly write C (and that only mostly) in C++ but it does not mean you know C++ anymore. Maybe literally 30 years ago you could show a C programmer how C++ compiles down into C and that would be enough for the C programmer to start bashing away at C++, but that day is long past. In terms of whether it is easier to ask a kernel developer to learn Rust or to learn C++, bearing in mind a kernel developer would be pretty much required to learn all of C++, every last quirk and every last detail, it honestly could still be advantage Rust on that front. Katamari Dama-C++ is a monster.
- bryanlarsen 3y ago
  They're talking about a very small strict subset of C++. For example, it excludes function and operator overloading, so your 2010 comment is handled.
  - pointlessone 3y ago
    I’m not familiar with C++ so might be a silly question but is it possible/easy to enforce usage of an arbitrary subset of the language features?
    - bryanlarsen 3y ago
      Easy? No. But the Linux kernel is one of the places where it is definitely possible. Kernel patches are well reviewed both manually and automatically.
hulitu 3y ago
> A 2024 Discussion Whether to Convert the Linux Kernel from C to Modern C++ They can call it Windows then. /s

34 points by ashvardanian 3y ago

I don't normally get into C/C++ or kernel chats, but I hope they choose something different than C++, maybe Zig. The problem with C++ is that like Scala, it's hard to stick to a subset without realizing you've used more than intended after the fact. Maybe the kernel team could do it since there's a culture of keeping things clean/simple.

kreetx 3y ago
I think they'll likely agree on some C++ subset when getting into it so I wouldn't be worried. Though, why not just evolve C itself with those features? C is evolving, too, and this could be used as a vehicle for it.
teleforce 3y ago
Perhaps they should just jump straight to D. The proposed C++ 20 is like D-- given how many D features have been adopted over the years. Now that D supports C native compilation and supported by GCC eco-system so the conversion can be done in peacemeal, it will be win-win situation for safety, convenience, power and flexibility. One of the main reasons of switching to C++ given is for templates, and D templating system is second to none. D also run and compile as fast as C, and the latter cannot be said to C++. For kernel developers fast compilation will be necessary not a luxury. Imagine kernel development in D without GC and user level development with D with the default GC, it really looks like a match made in heaven to me.

jerf 3y ago

Modern C++ has addressed none of the reasons Linus originally rejected it. Those reasons were widely misunderstood, but I tried to explain them at the time and it seems to have helped at least some people: https://news.ycombinator.com/item?id=1422160 https://news.ycombinator.com/item?id=1422160 Given the original objections, it should only be even more objectionable now as C++ has added even more subtleties and even more you have to understand in order to fully grok what a given bit of C++ code is doing. I'd also disagree with the idea that it's lower effort to add C++ than to use Rust. C and C++ have diverged wildly. Knowing C means you can mostly write C (and that only mostly) in C++ but it does not mean you know C++ anymore. Maybe literally 30 years ago you could show a C programmer how C++ compiles down into C and that would be enough for the C programmer to start bashing away at C++, but that day is long past. In terms of whether it is easier to ask a kernel developer to learn Rust or to learn C++, bearing in mind a kernel developer would be pretty much required to learn all of C++, every last quirk and every last detail, it honestly could still be advantage Rust on that front. Katamari Dama-C++ is a monster.

bryanlarsen 3y ago
They're talking about a very small strict subset of C++. For example, it excludes function and operator overloading, so your 2010 comment is handled.
- pointlessone 3y ago
  I’m not familiar with C++ so might be a silly question but is it possible/easy to enforce usage of an arbitrary subset of the language features?
  - bryanlarsen 3y ago
    Easy? No. But the Linux kernel is one of the places where it is definitely possible. Kernel patches are well reviewed both manually and automatically.

hulitu 3y ago

> A 2024 Discussion Whether to Convert the Linux Kernel from C to Modern C++ They can call it Windows then. /s