Why I Still Like C
mitchgolden.com
mitchgolden.com
- Abstractions will always be leaky
and
- be prepared for when an abstraction leaks.
As for the ideology part:
think for yourself and try to see beyond the hype
And I have to say I absolutely agree.
For example, I've written maybe a handful of lines of assembly, and maybe as many lines of inline assembly. However, understanding assembly language has saved me many times. Compiler bugs exist. Sometimes you don't have access to debug info and a debugger, just the instruction pointer and a register dump. Sometimes the compiler just can't optimise well enough.
C is just one level of abstraction higher, so that's probably the level that's going to leak through when your high-level-language interpreter fails for some bizarre reason. (recent Ruby segfaults anyone?)
It's not turtles all the way down, but rather a menagerie of different abstractions, and it's useful to know your zoology, as it were.
I feel bad for people that think imperative-style programming is the only way things can be done.
C is not the anti-Java. It's Java minus modern features and with more rope to hang yourself with. (Then again, someone actually told me that spending an entire day with Valgrind hunting down a memory leak was fun.)
If you want power, non-Java-ness, and modern features, learn a language like Lisp, Haskell, OCaml, etc. And learn Perl for actually getting things done.
There's nothing wrong with learning and using C. It may not be your favorite flavor, but it's a perfectly valid and reasonable choice.
I didn't say not to use C; there is plenty of good software written in C. Like... everything :)
But, I don't think it's very valuable to know. It doesn't put you in a good position to write correct, maintainable software. There are no tools for that, only for twiddling bits. Therefore, I think C should be relegated to interfacing with UNIX (which you Just Have To Do) and small number-crunching routines. In those cases, there's really nothing to screw up (UNIX interface -- allocate resources, mutate them, deallocate; number crunching, call a function with some arguments). For both of those uses, C has been very successful. Just don't write a big application with it, there are much better tools for that.
I'd actually claim that C is very valuable to know, and while I'm at it, so is assembly. Pretty much all high level languages are written in C (or in a language that is itself written in C). Anyone that doesn't understand enough about how a modern CPU operates and cannot implement a compiler/interpreter for a higher level language in C should not claim to be a hacker.
I've been hearing how obsolete C is ever since high school (1994) yet pieces of successful programs not written in C-family is an order of magnitude less common: everything form web-servers to browsers, OSes and office suites and imaging software - all C. Meanwhile I've never had to install JVM on any of my computers, somehow I never run across a single piece of useful software written in Java (with exception of evaluating Jython and toying with Eclipse&Netbeans at work).
Perhaps it's running on my phone. If so, that's probably why the damn thing feels so anemic compared to Objective-C powered iPhone.
I don't know how C programmers do it, but somehow they just get things done.
Moreover, C remains the only choice of writing the most portable code ever. There isn't a reasonably popular CPU on this planet that doesn't have a C compiler for it, while Java runs only on "java computers".
You seem to think that Java is C's competition. It's not. I think we can all agree that Java is one of the worst programming languages ever created. (The VM is nice though.) Nobody is telling you to use Java or C++. They are telling you to use a dynamic language (Perl/Python/Ruby) or a functional language (Haskell/OCaml/Lisp).
I don't know how C programmers do it, but somehow they just get things done.
Sure, but their programs usually leak memory (ever use Firefox?), and their libraries handle errors by terminating the whole program (see PulseAudio; any error condition immediately calls exit(1)).
There isn't a reasonably popular CPU on this planet that doesn't have a C compiler for it, while Java runs only on "java computers".
I don't optimize my life for solving problems I don't have. Every machine I want my code to run on has Perl, Lisp, Haskell... whatever. If C was the highest-level language I could use, then obviously I'd use it. But it's not, so I don't.
Finally, gcj will compile Java down to machine code; ghc compiles Haskell to native code. So I don't ever see a situation where I will absolutely have to use C for some reason (other than interfacing with C libraries, of course).
That's a poorly written program. You can just as easily System.exit(1) in Java, or to do it a more Javaesque way, throw new Error("im too lazy to do this right").
One (exit) makes the code simpler, but makes the consuming application extremely flaky; the other bloats the code with a weird API where functions modify their arguments:
errorcode_t do_some_work(int arg1, int arg2, int *result)
It's possible to write code that way, but why would you want to? This is a solved problem; why keep using a square wheel when round ones are available?
(Yes, there is longjmp. Still a lot of work to use that square wheel.)
In the end, don't think I care what programming language you use. It doesn't matter. But I do think that C should be dying off now; it is much less useful than it was 30 years ago.
Of course it does, in fact you just said it yourself - long jumps. Back in late 90s I worked with the firmware for the Point of Sale terminals, PIN pads, etc. It was written in C and its error handling was exception-based. Behind the scenes it was just a handful of long/setjmp wrappers, but it did nevertheless implement semantics of try/catch/etc.
Additionally, "exiting or returning false" are not the only two options, nor they even the most commonly used ones. At least in the projects that I was exposed to. Kernel code (e.g. Linux, BSD) routinely uses int as a return value and it still somehow manages to be both readable and functional without being "bloated" or using "weird API".
Yet another thing to consider is that optional exceptions (such as those in C++) come with a non-negligible performance hit, so it is considered an absolute no-no to use them in a "fast path" parts of the code.
> But I do think that C should be dying off now
But it just doesn't. Bummer :)
David Hanson's book "C Interfaces and Implementations" explains how this is done and provides (very high quality) source code you can use/learn from.
In case you're familiar with gcj, what's the status of that project? Their page hasn't been updating lately, is it being actively maintained? It looks like the most painless way to integrate some Java code I have with Ruby... and I don't wan to bet on a dying horse :)
And as a side note, regarding "Beating the Averages": there are many ways to beat the averages, and beating them ONLY by using a higher level language is like beating them with Aeron chairs: yes, some languages give you productivity boost over others, but I disagree with Paul on magnitude of that effect - in fact I'd say that availability of certain libraries beats any boost provided by language alone.
And, to conclude, here's a little example from real life: a group of programmers I know were working on some nifty AI stuff for automated document processing, and they were doing it in C++, which I disliked. They just discovered a totally different market in the embedded space, which allows them to grow and make money. Thanks for their language of choice (C++), they were able to beat an average Java/Python/Lisp-powered startup that doesn't have access to that segment.
Just saying... :)
C will not be obsolete for a very long time. It is however completely silly to compare it with Java. Completely different and created for different purposes.
You've missed the point. Learning C is not about imperative-style programming. It is about seeing one abstraction level down relative to your favorite programming language.
http://news.ycombinator.com/item?id=235727
Using:
And C is not inferior to Java. They serve different purposes.
Java is #1 in the industry. I'm not saying I like it but your school is just trying to make sure you can easily get a job after school.
But anyway, no arguing about this because it's been beaten to death (and I just woke up from a nap, don't anger me! :P)
It's like a very sharp knife: you can cut yourself as well as whatever it is you're trying to slice. It's really easy to write broken or inflexible code in it.
This is true of all powerful tools. I've seen horribly unmaintainable Lisp and Haskell. I've also seen very clean and maintainable Perl. C is no different than any other programming language, although it does actively discourage good code. You can ignore the discouragement, but why not use a language that encourages maintainability, or at least automatically deallocates memory when you're done using it?
Anyway, I don't think this article shows why the author likes C, or why someone else should learn C. Sure, learn C if you want to learn C. But you can also learn about your computer by just learning about your computer.
If you want a visceral relationship with your hardware, get a soldering iron and go to town. That's a respectable hobby, and confers mega geek-points. If you want to write software and be productive at it, learn a real language.
C is a powerful tool, and every programmer should know it. But it's woefully underpowered as a fulltime software development tool.
Anyway, I think programmers think too much in black and white. There are people that say you never need to know how the computer works. Then the rest say that you always need to think about how the computer works. The reality is somewhere in the middle. There have been a few cases in my programming career where knowing how the hardware actually worked was helpful. But it was never critical, just good to know.
(Another similar area is data structures. Yes, it's good to know how merge sort and quick sort and B+ trees work. That direct knowledge has helped me a number of times, either because I needed my own tree or because thinking about how merge sort worked helped me solve a related problem. But the other 99% of the time, I'm not thinking about sorting or the tree balancing algorithm. I just use libraries. So theoretically, it's possible to write good programs without knowing that stuff.)
Sure, an IT wonk can churn out working software. But they'll never aspire to build the world's next great startup. Have some standards, man. You can't be a true hacker if you're cutting corners like this. :)
If knowing obscure details of data structures made you rich, I would be rich :)
You can do an awful lot of useful work while relying on voodoo, but eventually the voodoo will catch up with you and you'll end up with an ugly mess where a better trained developer would produce an elegant hack. That's no less true for a web developer using rails than it is for an embedded systems driver developer.
Basically, I've never known a great hacker who doesn't understand CPU architectures and at least a little bit of digital logic design.
As for C specifically, let's just agree to disagree. I'll put of the sum total of "great hacks" written in C up against any amount of C++ or Java you can find. It's true that for some problems (web development being a good example) there are better tools (scripting languages, databases) for the domain. But that alone doesn't make C "woefully underpowered".
most likely implemented in C.
Don't get me wrong, it's wise to know your craft. That's never a bad thing. But it's foolish to use a tool that is, frankly, not as powerful.
Saying a tool is "powerful" doesn't mean "you can do lots of stuff with it." You CAN saw down a tree with a screwdriver. But a chainsaw is a much more powerful tool for that job. Yes, a screwdriver can do lots of things that a chainsaw can't, which only proves that it's a more versatile tool. The power of a tool depends on the context of the problem you're trying to solve.
I'm guessing that screwdrivers are used in the assembly of chainsaws. They still suck for cutting down trees. Same with C vs just about every other language. C is a fairly small step up from assembly. An important step, no doubt, but for most tasks that software developers face, especially on the web, it's not appropriate.
But that's not the same as saying that someone who has that background is as unproductive in C as you are. Seriously: spend some time writing something serious in C, you might like it better. It will certainly seem more powerful than it does to you right now.
I spent 5 years writing almost 100% C and C++ in college, with detours into Lisp, Java, VB, and Ada. (Actually, I'd grown up on Basic, so VB wasn't a detour, really.) After school, I worked at a VB shop, but built a few side-projects in C++. I've had to fix things that were broken, and I don't have a problem with the linker or debugger semantics. I've used templates, and both has-a and is-a object extension. I am certainly not an world-class expert in it, but it's not like I tried it once, got burned, and decided I hated it.
I'm a web developer because I like web development better. For a variety of reasons, not the least of them the challenge of building code that is so portable it will run on 12 different browser/os combinations, and the opportunity to work in many different languages, I find front-end development much more rewarding.
My aversion to C is based on a simple rule of thumb:
(usefulness of a feature) / (tokens required to implement it)
Another way to express this is that the elegance of a solution is the number of tokens that are not essential to the solution and do not aid in understanding the intent of the code. For example, compare Javascript's closures with Lisp's or Erlang's syntax. On the other side, compare any of those with the mess of class and object boilerplate in C++ or Java or PHP.
Conceptual cruft is even more pervasive. If encapsulation requires class/object constructs, then there is no way to get past that.
Some programmers will be able to create more elegant solutions than others, of course. But without changing languages, you can't get past the cruft that is built into language. (Or, in the case of C, the cruft that was not removed from the language.)
Again, "power" does not mean "you can do anything with it." C can be used to tell a computer to do anything that computer can do. But I'd argue that that's a naive view of the power of a programming language. There are things that are trivial to express in other languages, and require several lines to do in C. Just something as simple as comparing that two strings hold the same value requires about half as much conceptual overhead in almost any other language.
Or is there a version C with lexical closures, first-order functions and strings, and a garbage collector that I'm not familiar with?
I'm not sure what you mean by You don't have the scaffolding around you to provide a fallback for changes.
The claim that C programming takes more knowledge up front than web work does is frankly presumptuous and a bit misguided. I've seen the Javascript that C experts write. It's terrible. In fact, I'd argue that since quality web work requires an understanding of semantic HTML, CSS, Javascript, maybe some Flash and/or Canvas chops, and probably at least one kind of server-side language like PHP or Java, it requires much more knowledge up front than almost anything else out there.
I like C because it's pretty "close to the metal" but much easier and more portable than assembly. I can usually visualize what's going on under the hood if I need to. That's much harder with higher level languages. And on top of that it's a useful language to know.
Knowing assembly would also help you understand how computers work, even more so than C. But it's not as useful these days.
Are you sure you just didn't imagine it? ;-)
When you are maintaining a very very large project, the best thing about C is that when you read the code it is fairly obvious what is going on, and you don't have worry about some innocent looking code causing a bunch of other crap the be executed behind your back.
On the downside of C, name space collisions are a problem and the ability to hide data and functions.
I vote they should add C++ style classes to fix the name space problem, and private and public keywords to hide data and functions. I don't want all of the object-oriented B.S. like constructors and destructors that makes it difficult to read and debug other peoples code.
Don't get me wrong.....I think some features of C++ are very useful, especially operator overloading which has allow us to make it easier to code big programs that handle lots of longer intergers and weird integer sizes like 56 bits, or 112 bits, or 128 bits. Yeah, these days we have uint64_t in C99 compilers, but back in the mid-90's we had to write code in C++ just because it had operator overloaded classes, which we also had to write.
Even though I haven't stated earlier...I do real-time embedded software...so I'm not going to be writing an interrupt in Perl or Python anytime soon. C is not going away in the real-time / embedded software / driver world anytime soon. It is still very very common!
Of all of those, only C/C++ offers extremely high performance as its key feature. Being primarily a shorthand for assembly, C occupies a permanent place as the systems and performance language of choice and will remain there. If your application demands performance, use it. If not, then don't.
That's a good insight - what language you use is also a form of optimization.
Only a inexperienced fool says such a thing! The bottom line is that it all depends on your what you are trying to do. For embedded / real-time / drivers / O/S, both C and C++ are perfect, but for middleware and scripting and many other high level things they aren't the best language.
q mustn't be changed to point to something else (but what it points to can be changed -- except that in this particular example, q will point to a string constant, which shouldn't be modified).
Both of the above.
int i[10]; <- allocates memory for 10 integers
&i <- pointer to the first element in the array (i.e. i[0]).
correct or not?
int i[10]; does allocate memory for 10 integers (on the stack)
So i has the type int•. A pointer to a value: an array. A pointer to a value is the same thing as a pointer to the first value in the array. That's why you can use ++i, increment the pointer, and it will point to the second value in the array (++i => •i == i[1])
So &i is, wait for it, a int••, that is, a pointer to an array. So if you have an array of arrays, then you need to dereference twice. E.g. argv. It's a char••, so an array of strings which is an array of arrays of characters.
int i[10];
you can't do
++i
because modification of the value of i is not allowed.