《Lessons learned from c/c++ defects》总结

15 June 2014
版权声明：本文基于署名 2.5 中国大陆许可协议发布，欢迎转载，演绎或用于商业目的，但是必须保留本文的署名elloop(包含链接)

basic question

item1: n « 1 + 1 == 4n , not 2n + 1

item2: what will happen?

#define perimeter(x, y) 2*x + 2*y
volume = perimeter(x, y) * high

correct

#define perimeter(x, y) (2*(x) + 2*(y))

item3: macro pollution:

#define map __gnu_cxx::hash_map // in a.cpp
// files which include 'a.cpp' will be polluted by map define.

item4: 多语句宏使用错误，应该用{}把多语句宏括起来，或者最好是用inline函数代替宏.

item5: what’s wrong?

int main() {
    char c ;
    while ((c = getchar()) != EOF) {
        putchar(c);
    }
    return 0;
}

correct : int c;

  • item6: signed int 和 unsigned int比较时候，前者会转成后者，如果前者是负数，转换结果会影响比较结果。

  • item7: see:

struct data
{
    int flag:1;
    int other: 31;
}
int status()
{
    return 1;
}
int main()
{
    struct data test;
    test.flag = status();
    if (test.flag == 1)
    {
        printf("true");
    }
    else
    {
        printf("false");
    }
    return 0;
}

correct: flag is signed bit, value only be 0 or -1. so change to unsigned int flag: 1;

  • item8: compare float:
float f = 1.0 / 3;
float f2 = 0.333333;
// wrong
if (f == f2) {
    // same
}

// right
if (fabs(f - f2) < 0.00001)
    // same
  • item9: 求相反数
int minInt = 0xffffffff;
minInt = -minInt;
printf("%d", minInt);

// right: -1 最小负整数的相反数需要单独判断
if (minInt == 0xffffffff) 
{

}
  • item10: right?
int main()
{
    size_t size = sizeof(int);
    while (--size >= 0)
    {
        cout << size << endl;
    }
    retrun 0;
}

correct: –size > 0

  • item11: struct init style.
struct rect
{
    int l, w;
}

// bad style.
struct rect r = {10, 20};

// good style.
struct rect r = { r.l = 10, r.w = 20};
  • item12: 模板的两段编译
template<typename T>
class A {
    protected:
        int num;
};

template <typename T>
class B : public A<T> {
    void init() {
        num = 0;
    }
};

won’t compiling pass in latest complier but will pass in old complier 新的编译器模板两段编译：

  1. 扫描所有非依赖名称
  2. (实例化模板)扫描所有依赖名称(只有实例化的时候才能消除歧义的名称) reason: num是一个非依赖参数, 但是但是定义在基类中的，目前的编译器还没有实现第一阶段到模板基类中查找名称，因此会报错：num undeclared.

旧版的编译器没有实现两段编译，对模板类的语法解析时，所有的名字查找都留在了实例化的时候进行，所以可以通过编译

correct: this->num 将会使num变成依赖型名称

lib functions.

  • item1: sprintf: segmentation fault.
char src[] = "aaaaaaaaaaaaaaaaaaaaa";
char buf[10] = "";
int len = sprintf(buf, "%s", src);

suggestion

  1. snprintf: 会检查返回值n，如果n>len(buffer), 会重新分配空间并再一次调用snprintf.
  2. asprintf: 不预先分配buf内存，复制过程中根据实际复制源的大小动态分配内存，参考libc手册
  • item2: snprintf parameters:
char buf[10] = "";
char src[10] = "hello %s";
int len = sprintf(buf, sizeof(buf), src);

// *correct*
int len = sprintf(buf, sizeof(buf), "%s", src);
  • item3: return value of snprintf:
char buf[10] = "";
char src[] = "aaaaaaaaaaaaaaaaaaaaaaaaaaaa";
int len = snprintf(buf, sizeof(buf), "%s", src);
buf[len] = '\0';
printf("buf = %s, buf_len = %d\n", buf, len);

snprintf返回实际写入到buf的字符个数（假设buf大小没有限制） e.g.: “%s” src = “123” -> len = 3 “src=%s” src = “123” -> len = strlen(“str=”) + strlen(src) = 6 “%s” src = “1234567890123” -> len = 13

correct

int len = snprintf(buf, sizeof(buf), "%s", src);
printf("return len: %d\n", len);
if (len > sizeof(buf) - 1) {
    printf("error: src len is %d, buf size is %d not enough", len, sizeof(buf));
}
else {
    printf("buf=%s, len is %d\n", buf, strlen(buf));
}
分享到