Comment 26 for bug 1347147

Kerberos is miscompiled by gcc-4.8. The impact is detailed at, but here is a reduced test case. The expected return is 0, but when compiled with gcc-4.8 -O2, it returns 1.

$ cat bug.c

struct node { struct node *next, *prev; } node;
struct head { struct node *first; } heads[5];
int k = 2;
struct head *head = &heads[2];

int main()
  node.prev = (void *)head;
  head->first = &node;

  struct node *n = head->first;
  struct head *h = &heads[k];

  if (n->prev == (void *)h)
    h->first = n->next;
    n->prev->next = n->next;

  n->next = h->first;
  return n->next == &node;

$ gcc-4.7 -Wall -O2 bug.c -o bug; ./bug; echo $?
$ gcc-4.8 -Wall -O2 bug.c -o bug; ./bug; echo $?
$ gcc-4.9 -Wall -O2 bug.c -o bug; ./bug; echo $?
$ dpkg -l gcc-4.7 gcc-4.8 gcc-4.9
ii gcc-4.7 4.7.4-2ubuntu1 amd64 GNU C compiler
ii gcc-4.8 4.8.3-6ubuntu1 amd64 GNU C compiler
ii gcc-4.9 4.9.1-3ubuntu2 amd64 GNU C compiler

I bisected the point where the problem disappeared between 4.8 and 4.9 at r202525. However, I don’t understand why. I’m scared by the fact that r202525 was intended to fix a “missed-optimization” bug (bug 58404).