perlun.eu.org · About the blog · Programming · Select language / Välj språk

Why C is a dangerous programming language

In this blog post, I describe one particular aspect of the C language that can lead to unpleasant surprises if you’re not careful.

I was writing some code for the Perlang standard library which implements printing of floating-point numbers, using the freely available double-conversion library from Google. To be able to write tests for this, I wrote a wrapper function which overrides puts() with some custom logic. The function looks like this;

const char *captured_output = NULL;

extern "C" void __real_puts(const char *s);

extern "C" void __wrap_puts(const char *s) {
    if (puts_mocked) {
        captured_output = s;
    }
    else {
        __real_puts(s);
    }
}

Then, in my calling code, I would do something like this (very simplified example with all error handling removed):

const int PRINT_BUFFER_SIZE = 100;

void print(double d)
{
    char buffer_container[PRINT_BUFFER_SIZE];
    double_conversion::Vector<char> buffer(buffer_container, PRINT_BUFFER_SIZE);
    int length;
    int point;
    bool status = FastDtoa(d, double_conversion::FAST_DTOA_SHORTEST, 0, buffer, &length, &point);

    char formatted_number[length + 1];
    int i = 0;

    while (i < length) {
        // TODO: Add decimal point at the right place
        formatted_number[i] = buffer[i];
        i++;
    }

    formatted_number[length] = '\0';

    puts(formatted_number);
}

This worked quite fine, in the sense that the expected value (12345) was being printed to the screen.

There was just one problem: the assertion in my test was failing miserably. Here's the assertion (CHECK_EQ is an assertion method defined in the double-conversion library mentioned above):

TEST(PrintDouble) {
    puts_mocked = true;
    perlang::print(123.45);
    puts_mocked = false;

    // This is the failing assertion
    CHECK_EQ("123.45", captured_output);
}

The output from running the tests looked like this:

/home/per/git/perlang/src/stdlib/cmake-build-debug/cctest print
/home/per/git/perlang/src/stdlib/test/print.cc:27:
 CHECK_EQ("123.45", captured_output) failed
#  Expected: 123.45
#  Found:    ␛
Signal: SIGABRT (Aborted)

What?? Why?! I don’t understand anything. I left the project at the end of the day with a feeling that not much was working as intended...

Then, the morning after when I was still in bed, I realized what was probably happening. It was a typical “C mistake”.

Remember my __wrap_puts method above? Take a close look at it again. It takes a parameter and saves the value of this parameter for later use. Then look at my print() method above: it declares a stack variable, builds up the correct string inside this variable, then calls puts()... and then returns.

I've highlighted "stack variable" here, because this is precisely the problem here. Pointers to the stack cannot be saved like this, it's doomed to fail. Unfortunately, the stack will be overwritten between the puts() call and the method which checks the captured_output content.

The solution

extern "C" void __wrap_puts(const char* s)
{
    if (puts_mocked) {
        // Note: calling this multiple times will both leak memory and overwrite the captured
        // output from previous calls. Don't use this in the real world without adding suitable
        // free() calls.
        captured_output = strdup(s); // <-- calling strdup(s) here is the fix
    }
    else {
        __real_puts(s);
    }
}

By duplicating the string, which can be either located on the stack or on the heap, we are safe, and the assertion will now fail as expected instead:

/home/per/git/perlang/src/stdlib/cmake-build-debug/cctest print
/home/per/git/perlang/src/stdlib/test/print.cc:27:
 CHECK_EQ("123.45", captured_output) failed
#  Expected: 123.45
#  Found:    12345
Signal: SIGABRT (Aborted)

(The reason it fails is because the implementation is not yet complete; it doesn't add the decimal point at the right place. This is fine and a perfectly "normal" software bug. Figuring out why a garbage value was being printed is a much more "annoying" kind of bug to me.)

Conclusion

"C makes it easy to shoot yourself in the foot; C++ makes it harder, but when you do it blows your whole leg off" Bjarne Stroustrup

With great power comes great responsibility. The fact that you can easily allocate objects on the stack is actually a really nice and useful thing, and one of the things that has historically given (and perhaps still gives) C and C++ an advantage for writing performance-critical code. To be able to completely eliminate memory allocation and deallocation for short-lived objects is simply very nice. But: like many other advanced features in C (like pointer arithmetics...), it comes at a price. And the real nasty part is that there are very few "safety belts" in C. So arguably, the greatest strengths of the C language are also its greatest weaknesses.

The careful reader might be wanting to point out that my example code above is technically not C but C++ - I know. Traditional pointers (e.g. const char *) are much more "C-style" than "C++-style" though. In C++, references are typically favoured over plain-old pointers.