Carbon: high level programming language that compiles to plain C | Hacker News Reader

Carbon: high level programming language that compiles to plain C

home.kpn.nl

48 pointsflexterra39 comments

I wish projects like this wouldn't force you to install their libraries in order to play with them.

Sure, it's just /usr/local, but @#$#@$@# it's annoying when you have to make install on the dependent library before the core application will install.

Use scons. or a makefile. or cmake. Or build a static binary. I don't care; but look at the go source archive for an example of how to do this right.

This isn't a good first touch of a project:

./configure: line 12296: syntax error near unexpected token `CO2,' ./configure: line 12296: `PKG_CHECK_MODULES(CO2, libco2-1.0 >= 0.1.2,,as_fn_error $? "required module missing" "$LINENO" 5)'

hint: you'll notice autoconf isn't in the list above. :P

It's is one of life's small pleasures when programs (useful programs) compile fast and with relative simpicity.

If you want to experience this, try www.sigala.it/sandro/ and cutils. There's a links to it on the OP's page. Programs as UNIX filters, written with flex and bison. Beautiful.

I like this OP because he includes a grammar file. If every programmer that tries making his own language did this, the world would be a better place. Did he also include the .l file in the src? I'm too lazy to look. I think they should include those too. It makes the whole thing easier to understand (and modify).

I too get annoyed with having to install the author's own libs. Sometimes these libs are better than the standard ones (e.g. everyone knows C's stdio sucks). If the special libs are fixing something that needs fixing, I'm OK with using them. But nine times out of ten, that's not the case.

For this sort of thing, i.e. C code generation, I still like the LISP's that generate C.

Reminds me of https://live.gnome.org/Vala

When the first thing I read about a language it on it's description page is "compiles to X" or "runs on Y", this is definitely a sign that the language ha NOTHING INTERESTING TO OFFER whatsoever. Adding insult to injury, underneath is a bulletted list of language features that any language worth it's salt has them! And please... "Named constructors"... WTF?!

(If someone wants to be mean, he could simply compare this feature-wise with one of the Scheme dialects that compile to C...)

Not every language needs to be one of those that changes the way we think about programming. It is perfectly fine to experiment on mundane languages by adding mundane features. Incremental improvement is a good thing and it is too early to just write off this because it seems to be un-interesting.

I'm sure this project was extremely interesting to the author - maybe from a different perspective than yours.

> (If someone wants to be mean, he could simply compare this feature-wise with one of the Scheme dialects that compile to C...)

I'm not being mean here, but if someone really wanted to be..

> not every languages needs to change ... way we think about programming. > I'm sure this project was extremely interesting to the author

No, but neither does(should) every new language be posted and upvoted to HN.

If it's upvoted, then the HN crowd finds the language interesting. If not, then it will fade away like every other article that doesn't get votes does.

The Example link contains, I think, only two things that aren't valid C++. "self." (could be replaced by this-> or simply nothing in C++) and the ["stdio.h"] int printf... syntax for including just one thing.

Two and a half things I guess: the methods of MyObject are accessed by main but not declared public (C++ is private by default)

I don't know if that's good or bad, but it's not very compelling to me.

Without a source-level debugger, I see little advantage relative to COS, which is C, but has a more advanced object and type system than Carbon. http://news.ycombinator.com/item?id=1192791 https://github.com/CObjectSystem/COS

There's less typing for the programmer than in COS as far as I can tell. And productivity is an important improvement.

This is going to need better documentation than a Yacc description if it is going to catch on.

There's example code here, http://home.kpn.nl/pt.kleinhaneveld/carbon/example.html, for what it's worth. Pretty pedestrian, from my perspective.

It's odd to see compilation to C as a main selling point.

Some of us rather like C...

C is eternal. Have you tried Go?

I have not... Should I?

Check out the std lib: http://golang.org/src/pkg/ -- that's canonical Go code.

It's my language of choice for personal projects. A neat and compact language that is very trivial to pick up, feels light, and yet sufficiently subtle to intellectually engage your mind.

Very productive, good performance, and very portable.

http://www.youtube.com/watch?v=HxaD_trXwRE

caveat: /imho/ the runtime is not yet fully cooked. ETA is Go1.1 (end of year I believe) for beefed up internals.

Sure, but that doesn't make it a good intermediate representation for a programming language. I really like CoffeeScript, but compiling to it would be insanity.

That said, there are some good reasons for compiling to C. Not every architecture has a backend for LLVM yet, after all.

Some would say "I really like JavaScript, but compiling to it would be insanity." And yet CoffeeScript is there. How is that better than a language which compiles to C?

Unfortunately, there is a massive reason to compile to JavaScript... browsers.

If you read the second part of my comment, you'll see that I agree that there is a legitimate reason to compile to C. It turns out to be identical to the legitimate reason for compiling to JS - some platforms can't be targeted directly, so you make do with the options you have.

I was not disputing the conclusion that it can make sense to compile to C. I was just saying that "people like C" doesn't support that conclusion.

Yeah, which is why traditionally a LOT of languages, particularly early in their careers, have compiled down to C. Really aside from JVM targeted languages, it still tends to be the trend.

So it is pretty weird for that to be highlighted as the big distinctive feature.

LLVM has C backend, though it is not fully supported. It seems like it would.be helpful for bootstrapping LLVM for a new architecture and for A/B testing of other backends' performance and correctness.

> Yeah, which is why traditionally a LOT of languages, particularly early in their careers, have compiled down to C. Really aside from JVM targeted languages, it still tends to be the trend.

What about .net/mono targeted languages, javasvcript targeted languages? And custom interpreters jit(frequently but not always written in c)

CLR targeted languages are few and far between, as are JavaScript targeted languages, so I didn't consider them a big enough factor to impact the "trend".

People focused on entirely new languages tend not to target JITs first for the same reason they tend to not target direct machine code first: it just adds complexity to getting a proof of concept out the door. LLVM, the JVM, and yes, the CLR are starting to change this, but aside from the JVM, there are still only a handful of languages going that route.

Well, it automatically implies a pretty high baseline of performance and (if it's done right) unmatched portability.

"it automatically implies a pretty high baseline of performance"

I doubt that. It may be (mostly) true for languages that match the capabilities of C well, but if a language does not match C well, it certainly isn't 'automatically' because the optimizer in your C compiler will not help much in making the implementations of such features efficient.

For example, compiling JavaScript to C, it would be non-trivial to make the result as fast as the current JavaScript engines because the C code would likely have to do the same zillions 'is this a string?' checks that interpreters do (probably even more of them because it is harder to see the wood for the trees in your translated-to-C code)

For example, it would require lots of work to support, for example, Lisp's number system, Dylan's union types, any variant of 'eval', or, to take an extreme example, to map Intercal code sequences to efficient C operations such as '+'.

Not really. There is a C compiler for every platform worth its salt.

My point was that usually a language does more to sell itself. The selling point of C++ wasn't "here's a language that compiles to C" (which it was in the beginning), but rather "Here's a safer, OO version of C with many new features."

While not discouraging the original author, Vala from Gnome has taken this thinking a lot further. Vala is actually quite fun to use even for non-GUI projects. And it doesn't have to depend on GTK libraries to produce working programs.

https://live.gnome.org/Vala/Documentation

This is going to be rather confusing, I actually thought someone accidentally reinvented Carbon, https://developer.apple.com/carbon/.

I realise Apple's Carbon doesn't compile to C but my brain's transderivational search just ignored that fact.

Writing Carbon that calls Carbon. Let's call this act Magnesium because ... fusion.

Instead he reinvented something closer to Objective-C.

Apple's Carbon also isn't a language, but rather an API.

Compiling to C puts some significant constraints on your language. For example, C doesn't have COMDAT support, thread local storage, exception handling is limited to setjmp/longjmp, static uplevel function links are a problem, symbolic debug info is going to all look like C, etc.

It's just ubiquitous.

I'm not sure I understand how C's lack of high-level exceptions makes a difference for a high-level language that compiles down to C. The high-level language can provide exception support, just like large-scale C packages sometimes do using longjmp.

Similarly, are you really suggesting that C programs can't have thread-local storage?

setjmp/longjmp tends to be pretty primitive, compared to decent eh support.

Standard C does not have thread local storage.

Are you missing the fact that we're talking about using C as a substrate for a higher-level language?

Reply on news.ycombinator.com