via:reddit

Here is the whole story:

Going back to C
by buddhabrot

When I was 12 someone gave me a Metrowerks IDE to learn programming on my mac. I learned a lot with it, and got programming in C. I never really wrote a large program in it, only things like ROM hacks that extracted levels and models from snes/psx games I liked. Later, I got a windows PC and messed around with MFC in C++. That was the end for me, programming UI’s was too much work in C++ and I couldn’t handle the encyclopedic knowledge needed to program the Microsoft framework.

Later, I learned Java, which was also the language most commonly in my college. Making UI’s was a blast with it, I could factorize anything and really loved making as generic code as possible with it, even for small projects. Everything I made was made in Java.

Later in uni, I started learning functional languages, we used Haskell in one class. I got intrigued with Haskell and started solving Euler problems, up to this day I still have not written anything useful in Haskell but I love using it.

In the mean time I did some jobs for writing web sites. I used perl and php. After a while I only used php and used it more frequently.

After working for a while as a programmer I started to love higher-level languages, most of all Ruby. I became convinced that higher-level interpreted languages were the future because the slower execution times were to become meaningless. For any web projects I started using Ruby on Rails and to this day it still is my favourite framework.

Then, I learned Erlang because everyone was talking about it, and it showed me how to unleash parallellization, in whole new ways. I still think Erlang is the most amazing language I have ever seen. But then I found out it was also rather slow for the things I had in mind with it. I just didn’t find any use for it and never completely learned the language.

Last week I wanted to make a big poster of a fractal. I wanted to compute it on a friend’s octocore, and write a fast parallellized program. At first, I used Ruby, thinking it would perform better than I thought. And I quickly abandoned the idea because it was too slow.

Then I rewrote the same single-threaded algorithm in Haskell and saw that it was much, much quicker. But when I tried to parallallize it, I couldn’t fathom how to do it in Haskell and abandoned the idea. I tried to write a parallellized version in Erlang, and while it spreak across the machine nicely, each individual process was way too slow, the parallellized algorithm was hundreds of times slower than the single-threaded haskell.

Because I really wanted that poster, I decided to switch to C because I had this idea that nothing could fail in C. And I started to code in a language I hadn’t used for years, even a decade. And suddenly it hit me, everything made sense. Sure, I was writing so much boilerplate but it didn’t bother me. I suddenly realized that everything I wrote would be inside the program’s code, and that nothing more would. It was a great feeling.
I used simple pthreads to do the parallellization, and I used libgd to output my data to a png file. I started learning about mutex lock hierarchies to make the parallellizations overlap without endangering thread safety, it was all ugly code and the API’s felt horrible compared to what I was used to.. but it all worked out. The resulting program was thousands of times faster than anything I had written.

And now, I am seriously wondering where I have been all this time. Sure, I still think C is incredibly clunky. But now I got the dust off of my old course on writing compilers (Modern compiler implementation in C), and I want to write a compiler in C for a small programming language. I want to get down to earth again, and write things that are efficient and just run very quickly. I want to write lots and lots of code and know exactly what every line does.

Has anyone else had this feeling, this evolution? Am I going mad? Do you think C will last, and there are intersting jobs left for pure C developers?

I have the same question: Are there intersting jobs left for pure C developers?

If you have problem to compile Broadcom 802.11 Linux STA driver (version 5.10.27.6) on Kernel 2.6.27 and get error message like the following:

/home/ye/downloads/bcm4328/hybrid_wl/src/wl/sys/wl_iw.c: In function ‘wl_iw_get_scan’:
/home/ye/downloads/bcm4328/hybrid_wl/src/wl/sys/wl_iw.c:934: warning: passing argument 1 of ‘iwe_stream_add_event’ from incompatible pointer type
/home/ye/downloads/bcm4328/hybrid_wl/src/wl/sys/wl_iw.c:934: warning: passing argument 3 of ‘iwe_stream_add_event’ from incompatible pointer type
/home/ye/downloads/bcm4328/hybrid_wl/src/wl/sys/wl_iw.c:934: warning: passing argument 4 of ‘iwe_stream_add_event’ makes pointer from integer without a cast
/home/ye/downloads/bcm4328/hybrid_wl/src/wl/sys/wl_iw.c:934: error: too few arguments to function ‘iwe_stream_add_event’
/home/ye/downloads/bcm4328/hybrid_wl/src/wl/sys/wl_iw.c:939: warning: passing argument 1 of ‘iwe_stream_add_point’ from incompatible pointer type
/home/ye/downloads/bcm4328/hybrid_wl/src/wl/sys/wl_iw.c:939: warning: passing argument 3 of ‘iwe_stream_add_point’ from incompatible pointer type
/home/ye/downloads/bcm4328/hybrid_wl/src/wl/sys/wl_iw.c:939: warning: passing argument 4 of ‘iwe_stream_add_point’ from incompatible pointer type
/home/ye/downloads/bcm4328/hybrid_wl/src/wl/sys/wl_iw.c:939: error: too few arguments to function ‘iwe_stream_add_point’
...

Then you need to apply the patch: hybrid_wl-5.10.27.6_patch-2.6.27

How to apply the patch:

1. Download the STA driver and the patch.
2. Extract the STA driver and switch to the root of source tree (where Makefile is located).
3. patch -p1 -E < path_to_patch/hybrid_wl_5.10.27.6_patch-2.6.27

Here is my solution:

class Foo
{
public:
    Foo(const Foo& f) {}
    static Foo getAnInstance()
    {
        return Foo();
    }
private:
    Foo() {}
};

Explanation:

To prevent users from creating objects on heap, I basically can’t give them any public constructors (ctor); therefore, I need to write at least one ctor myself and make it private since if I fail to do so, the compiler will generate a public default ctor for me and that is exactly what I don’t want. Then the user can only use the static member function getAnInstance to get an object to use.

Analysis:

In fact, my solution contradicts the explanation because I actually have a public ctor — the copy ctor. The copy ctor makes the following usage work:

Foo f0 = Foo::getAnInstance();

But because there is a copy ctor, after the user get his/her first object, he/she can do the following:

Foo* f1 = new Foo(f0);

This is saying that my solution can only guarantee the first object is on stack and it can’t go further.

So, my solution fails to solve the question.

p.s. I haven’t really seen the point of the interview question, but it’s interesting to think about ;)

Update:

The correct solution is to overload the new operator and make it private. Credit to DANIWEB

However, in my current assignment, I have to use a makefile provided by the prof. and there are dependency cycles among classes. The following is a simplified class digram shows the relationship:

IMAGE IS MISSING, SORRY!

I screwed up the header files; I spent almost one hour to solve it. I tried a lot different combinations of #includes and forward declarations, but none of them satisfied the preprocessor and/or the compiler.

After I became tired, I decided to leave the keyboard alone. I picked up my pen and a piece of paper and drew a small class diagram like the one above but much uglier. Suddenly, I found the problem guy was the class B because it had two outgoing dependencies and one incoming. Say that B’s dependencies count (dc[B]) = outgoing – incoming = 1, and dc[A] = -1, dc[C] = 0. The greater the number is, the class has more problems.

The problem is clear. Satisfying C’s dependency is a piece of cake, I just need to forward declare class A in class C’s header, c.h, then I include c.h header in b.h. Since the declaration of A is in c.h, I don’t need to declare it again in B’s header. Then, I include b.h into class A’s header file a.h.

Done! I’m not sure the program would work correctly, but now, it can be compiled.

Now I can imagine how hard the problem could be if hundreds or thousands files/classes are there sitting with cycles; therefore, the original design is so important and I believe that’s why we have to take tons of design/software engineering courses in the university.

cout << hex << an_integer;

And if I want to show “0x” before the value, I need to do the following:

cout.setf(ios::showbase);
cout << hex << an_integer;

The above code works well for either positive or negative integers, but not the value ZERO. If I run the following code:

#include <iostream>

using namespace std;

int main() {
    cout.setf(ios::showbase); //1
    cout << hex << 100 << endl;
    cout << hex << 0 << endl;
    return 0;
}

I am expecting this:

0x64
0x0

However, the actually output is:

0x64
0

The hex number notation “0x” is missing for the value 0.

I also tried this:

cout.setf(ios::hex | ios::showbase);
cout << 0;

It doesn’t work, neither.

This is weird to me. I am wondering why it is and is there a solution without any “ugly” hacks such look like:

cout << "0x" << hex << an_integer;

Update: After asking some of my friends, the solution above seems to be a decent solution to the goal I want to achieve; otherwise, code might become uglier at end of the day. Why not take it easy?