No testcase attached...

Created attachment 58266 [details]
test case for compiler bug

Sorry about that :) Should be there now

Any news on this? Would be nice to know if it actually is a bug in the compiler
or not. Sorry to be nagging about it, but I suddenly stopped trusting my
compiler after seeing this :)

I've made a self contained testcase from it (bellow), verified the bug
in 2.96-RH and 3.0.x and found it is works just fine in 3.1.
As it works in 3.1, it is lower priority to me ATM.

typedef struct
{
  char *str;
  int len;
} A;

A *foo (void)
{
  static A x;
  x.str = "1234";
  x.len = 5;
  return &x;
}

int bar (const char *x)
{
}

char *baz (const char *x)
{
  return (char *) x;
}

int
main (int argc, char *argv[])
{
  char *s, *t;
  A *u;

  u = foo ();
  t = (*u->str == '\0') ? u->str : (u->str + 1);
  s = baz ((*u->str == '\0') ? u->str : (u->str + 1));
  bar (s);
  bar (t);
  if (*s != '2')
    abort ();
  return 0;
}

I had someone test it on 3.1 sparc and he saw the problem there too. I'll try to
get the version from him.

He used 3.1.0 from sunfreeware. Self compiled.

Close this as fixed in the latest release?

Confirmed ok in 8.0