strstr: behavior different using different versions of gcc

Question

I'm porting a program that works with Ubuntu 8.04 (gcc version 4.2.4) to 10.04 (gcc version 4.4.3). I have the following code:

#include <stdio.h>
#include <string.h>

int main(void) {
  char p[100] = "////abcd";
  char *t;

  /* Remove duplicate slashes, saving only one of them */
  while (t = strstr(p, "//"))
    strcpy(t, t + 1);

  printf("%s\n", p);

  return 0;
}

The result should be /abcd, which it is with gcc 4.2.4. With 4.4.3, the output is /accd.

Can you suggest me a code change that will give the correct output using both versions of gcc, and preferrably explain what is going on here.

Thanks in advance!

Answer 1

You've just been lucky.

From the strcpy documentation:

The strcpy() function shall copy the string pointed to by s2 (including the terminating null byte) into the array pointed to by s1. If copying takes place between objects that overlap, the behavior is undefined.

The strings overlap in your case, your program invokes undefined behavior.

A possible reason why it used to work but no longer does is that strcpy could have been implemented as a builtin by GCC like memmove (i.e. safe in this situation), but this changed to a non-safe version for performance reasons. (This is pure speculation.)

To fix it, use memmove rather than strcpy, with something like:

while (t = strstr(p, "//")) {
  memmove(t, t+1, strlen(t)); // strlen(t)'s bytes worth of data
                              // will cover the string bytes left
                              // and the \0 terminator
}

That's not terribly efficient, but it will work portably - memmove has to handle overlapping memory areas.