[Bug tree-optimization/61964] [4.8/4.9/4.10 regression] krb5 database propagation enters infinite loop; reduced test case

["vries at gcc dot gnu.org" <gcc-bugzilla@gcc.gnu.org>]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=61964

vries at gcc dot gnu.org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |vries at gcc dot gnu.org

--- Comment #8 from vries at gcc dot gnu.org ---
Using this patch on the example from the description field, I can modify the
example on the command line:
...
$ diff -u bug-orig.c bug-mod.c 
--- bug-orig.c    2014-07-31 14:00:50.663275717 +0200
+++ bug-mod.c    2014-07-31 14:01:49.403276412 +0200
@@ -11,7 +11,7 @@
   struct node *n = head->first;
   struct head *h = &heads[k];

-  if (n->prev == (void *)h)
+  if (FORCE n->prev == (void *)h)
     h->first = n->next;
   else
     n->prev->next = n->next;
...

1. -DFORCE="" gives the original
2. -DFORCE="1 ||" forces the condition to true
3. -DFORCE="0 &&" forces the confition to false

In this experiment, we don't use tree-tail-merge:
...
$ gcc -DFORCE="1 ||" bug-mod.c -O2 -fno-strict-aliasing -fno-tree-tail-merge &&
./a.out ; echo $?
0
$ gcc -DFORCE="1 ||" bug-mod.c -O2 -fstrict-aliasing -fno-tree-tail-merge &&
./a.out ; echo $?
0
$ gcc -DFORCE="0 &&" bug-mod.c -O2 -fno-strict-aliasing -fno-tree-tail-merge &&
./a.out ; echo $?
0
$ gcc -DFORCE="0 &&" bug-mod.c -O2 -fstrict-aliasing -fno-tree-tail-merge &&
./a.out ; echo $?
1
...

The last result seems to suggest that the example is not type-safe.

My understanding is that the problem is in the line:
  n->prev->next = n->next;
where we effectively do:
  /* ((struct node*)&heads[2])->next = node.next */
which is type-unsafe.