Hacker News Reader

Comment by kllrnohj | Hacker News Reader

> C will always have a use for critical performance in the embedded world.

Based off of what reasoning? It's not like C is any faster or more efficient than Rust or C++. Heck you can often go faster in C++/Rust than you can in C thanks to the ease of using abstractions that just make things go-faster for you automatically (such as SSO in things like strings, or constexpr for compile-time resolution).

The embedded world isn't using C because it's a good fit, they are using it because of legacy inertia and lack of motivation to move to anything else. IoT is starting to change that, as all those C memory vulnerability issues turn into product support costs.

And also realistically embedded isn't even restricted to C already. Very popular microcontrollers like the esp8266 or esp32 also have javascript runtimes.

View on HN

It's not like C is any faster or more efficient than Rust or C++

Huh? C is certainly faster than C++.

go-faster for you automatically (such as SSO in things like strings, or constexpr for compile-time resolution).

That is a terrible example. You shouldn't be using the STL (i.e. std::string copying) for performance critical sections to begin with.

Very popular microcontrollers like the esp8266 or esp32 also have javascript runtimes.

First, esps are not low end microcontrollers at all. Second, no one uses JS runtimes for performance critical code on microcontrollers (e.g. JS will not give you direct memory access). What an absurd claim.

In what benchmarks?!?

Virtual functions, runtime type information, STL, larger executable code size, etc...

Memory footprint will be smaller on C without using C++ abstractions. Which is important for embedded systems.

You mean like a Commodore 64?

https://youtu.be/zBkNBP00wJE

Yawn.

Yes. But even C was considered too slow for something like a Motorola 68000 back in the day. Of course it's possible.

Besides, C++ can be just as performant if you ignore a huge proportion of its language features and abstractions, but what's the point then? Just use C.

And this is precisely why embedded engineers do so.

The point is that C will never have proper arrays, strings, bounds checking, real constants instead of magic defines, compile time programming, modules.

Yet somehow it has stayed relevant for more than 45 years.

Go figure.

One cannot teach new tricks to old dogs.

What has made relevant was the rise of FOSS UNIX clones and the GNU guidelines for using C.

All major desktop OSes were already on the path of adopting C++ frameworks. BeOS, Windows, OS/2, Mac OS.

Thankfully C has been shown the door on Apple, Google, Microsoft, Sony, Nintendo and ARM platforms.

Others will follow.

FOSS UNIX clones and old time embedded developers are the only ones keeping C relevant.

Probably the most important C++ ideal is that you don’t pay for what you don’t use, and you _still_ get a more expressive language than C if you don’t use those features.

Your cons:

- Virtual functions. Don’t use them if you don’t want to incur an extra indirection.

- STL. Don’t use it if you don’t want to.

- Large binary size. Don’t spin up hundreds of template instances, problem solved.

- No RTTI? Literally a compiler switch away.

Some pros:

- constexpr

- true type safe generic programming (at the cost of code size if you get crazy with it and like TMP)

- type safe no-overhead containers (std::array, std::tuple)

- RAII ownership semantics (I.e. std::unique_ptr)

- Can still emit a C ABI

- Multi-paradigm at a language level, rather than procedural with other paradigms bolted on ad-hoc

The reason C++ (IMHO) is always a better choice than C is that if you don’t want to incur the cost of feature X, drop down to how you would do it in C and it works exactly the same and you can still use the zero-cost C++ features that will never land in C.

I would agree with this assessment in most cases. The problem is it's hard to determine which language features perform worse than C. But usually the cost of being wrong is low or inconsequential for modern hardware.

But it is not true for embedded systems. And the industry disagrees too, and continues to use C because of this.

> But it is not true for embedded systems. And the industry disagrees too, and continues to use C because of this.

What resource constrained embedded systems even exist? Fucking lightbulbs run JavaScript these days. SmartCards have been running Java for 20 years. What is the actual market you're so up in arms about that you think is choosing C on technical merits instead of purely inertia? What is this "performance constrained embedded" market to you?

What is this "performance constrained embedded" market to you?

The very definition of embedded systems?

I would say it is more much likely the many folks who work on embedded systems in C are right and do so with technical merit. And it is much more likely that it is you who is wrong. Which is quite clear as someone who seems to advocate using JS on embedded hardware for serious tasks. A detail that is seriously amusing in its own right.

I worked on embedded systems for trains and we used C++. Many people use C just because the chips they target only have a C compiler.

Where is that definition written on?

Are we talking about PICs with 4KB, or ESP32 that would run circles around the Amstrad PC 1512 that I used to play Defender of the Crown on?

A system that already had quite a few high level programming languages available for us to learn programming on.

Ah maybe we are speaking about my Visa card running a set of tiny Java Card applications on 256KB, to handle my authentication processes at POS.

> The very definition of embedded systems?

Microcontrollers got a lot faster and more powerful over the years. What problem in that space kept up with that speed advancement? Who is paying for hyper-optimized C/asm code instead of just spending an extra $.50 for a dual-core 240mhz ESP32? And who is doing that decision and has decided C is somehow faster than C++?

> Which is quite clear as someone who seems to advocate using JS on embedded hardware for serious tasks.

I never advocated for it. I said it happened.

Can you do high speed sampling, motor control, signal analysis,... in javascript in realtime? Usually microcontrollers have several memory region with different speeds and access characteristics. Can you fine tune in which parts of the memory part of your javascript program will go into? Can you write a driver for a high speed bus or low speed for that matter in javascript on a mcu?

No, but you can use JavaScript on ARM mbed OS, running on $5 ARM-based microcontrollers.

https://os.mbed.com/javascript-on-mbed/

On Samsung's SmartThings embedded devices

https://smartthings.developer.samsung.com/docs/index.html

SigFox customized IoT solutions

https://build.sigfox.com/

Or on the other side from ARM mbed's offering, a beefy ARM 580 MHz with 96 MB (RAM + Flash), using JavaScript alongside Rust.

https://tessel.io/

Now if your only definition of embedded are constrained micro-controllers that aren't even able to cope with standard ISO C89 without having additional proprietary compiler extensions, then naturally everything else are just miniature desktop computers.

Meanwhile some forward looking companies keep on delivering innovative products.

Parent poster wondered if there were such embedded applications that can't be done with javascript. And I listed some. That doesn't mean i'm implying that javascript can't be use on a MCU! Where did you get that idea from?

I don't think I ever gave a definition of an embedded system, but you can quote me to prove me wrong. A PC embedded in a slot machine for instance can also be called an embedded system. A more important distinction is whether the system must run in real time or not. And you can have a beast of a microprocessor and you will not be able to do low level real time processing with it in javascript.

> RAII ownership semantics (I.e. std::unique_ptr)

RAII is baad for performance. Really bad. It's an OOP programming style that can be used to glew a few high-level constructs together. But if you use it on the lower levels, you will miss out on a lot of (systemic) optimization possibilities, because RAII gives you piecemeal construction/destruction pairs.

constexpr/generic programming might have a few nice applications, but in most places they're used without a real need (from what I've seen), while leading to extremely slow compiles. (Those in turn lead to longer turnaround times, worse program quality, worse efficiency...).

std::array, like std::vector, is some nice sugar compared to the C primitives, but if you use them at API boundaries you will get bad coupling effects.

The C++ programmers that I look up to all write basically C with maybe a few select C++ features. Many never write "class" and never include C++ headers. Some just write plain C.

Curious, do you look up to Mike Acton?

I've never seen him programming, but he seems like a very knowledgeable guy. I find a lot to agree with in his talks.

Yeah, because now he is one of the devs leading the High Performance C# effort at Unity.

People do change their mind.

I know, and I can't follow your conclusions. (Btw. come to Handmade Seattle, he will give a talk there ;->)

For someone that used to advocate C, did not like C++ features, I find peculiar that he now stands behind C#.

Yes, I do know that HPC# is somehow constrained, yet even as C# subset, it still supports quite a few features that Mike wasn't so found of having to deal with in C++, as per his CppCon talk.

We don't know how he's using C# there. What are the clues that you base your guesses on? I don't know a lot of C#, but it can probably be used responsibly in a high performance setting.

It seems he's still working on the same ideas and themes that his (in)famous talk from CppCon was about. Check this out, for example: https://twitter.com/Icetigris/status/1116771416072806401 . As far as I'm concerned, that's basically C.

Btw, have you updated your idea of ECS and "Data-oriented design" in the meantime? (Man, I've really started to be annoyed by this buzzword!)

Well, for starters it is impossible to use C# without some form of OOP and its type system goes against the spirit of "Performance begets safety" from C devs like Mike.

Are you just trying to win an argument? Please look at what they actually do, and then maybe let's have a discussion about that. In any case I'm sure Mike is way more pragmatic than you imply, and couldn't care less about language labels as long as he can do his thing.

Btw for extreme data-oriented programming you might actually be able to afford GC, provided you have value-type compound types like I think C# has. That's because there are only a few allocations around. Not that GC matters in this case - you don't free the global tables anyway except maybe at the end. And bounds checks might be easier to optimize out (?).

If C++ is just as fast as C except when it's not, then it's not as fast as C.

More importantly, it's not easy to tell when you're going to hit performance issues with C++. You need an obscene amount of knowledge of both the language as well as hardware to get a sense of what to use to match C performance.

> Huh? C is certainly faster than C++.

It literally isn't. Quite often the reverse is true, as C++ gives you better tools for compile-time optimizations.

> That is a terrible example. You shouldn't be using the STL (i.e. std::string copying) for performance critical sections to begin with.

I think you're confusing the example with a recommendation to use the STL?

std::string is just an example of SSO. The concept is generally applicable and used in more places than just the STL.

Although "You shouldn't be using the STL for performance critical sections to begin with." is also complete fucking bullshit anyway. There's nothing wrong with the performance of eg std::array. Or unique_ptr. Or vector. Or a variety of other things in the STL. You're not outperforming any of those with raw C.

It's not 2003 anymore so stop acting like it.

It is 2019 and even game programmers still avoid STL for performance reasons.

And optimization is taken to another level from that on embedded systems.

I hate to remind you again, but it is 2019 and C is still very relevant on embedded systems. And Javascript, not so much. Snicker

Surely not the same game programmers that use Unreal with its C++ GC.

What does that have to do with anything?

Yes, those same game programmers avoid STL.

No one is advocating replacing C++ with C on capable hardware, troll.

> It is 2019 and even game programmers still avoid STL for performance reasons.

No they don't. They avoid parts of the STL. GDC round-table as far back as 2001 had 40% of shipping games using the STL: http://www.tantalon.com/pete/gdc01_round_table_report.htm

Many of the containers in the STL have performance problems. Yes avoid those (and avoid the god-awful iostream). But large chunks of the STL have no performance issues at all, and re-inventing it is just a waste of your time.

Games don't avoid things like std::vector for performance issues, they use their own thing because they want even more features.

Reply on news.ycombinator.com