bad dead code detection ?

[David Jobet]

Hello,

I try to write an exception handling library, and I'm forced to play with gcc
inline assembly.
You will find below a sample program.

The trick is to register address of local labels in a global static table. Later
when an exception is launched, I manage to jump to this label to do clean up.
The problem is gcc detects the label cannot be reached and dead code elimination
completely remove the exception handling code from the executable when compiled
with -O3.
When compiled without -03, the code is not removed, but the label is moved from
the *good* location to the return label.

I'm wondering if it's a bug, or if there is a way to prevent this.
When one take the address of a local variable, it cannot be put in a register.
I'm wondering if it should not be the same with code taking address of local
labels ?

Kind regards

David

---
void g(int throwError)
{
  // STATIC DATA **************************************************************
  // strings
  // --snip--
  static unsigned int initialised = 0;

  // nosica variables *********************************************************
  // --snip--
  void *jumpToLabel;

  // static initialisation
  if (!initialised)
  {
    ___cleanUpPoints[1] = &&g_cleanUpPoint; // registering in static table
    initialised = 1;
  }

  // nosica code **************************************************************

  // --snip--

return_label : // without -O3, the assembly code shows g_cleanUpPoint to be the
same as return_label
  return;

g_cleanUpPoint :

  // --snip--

  ___currentReturnAddress(pc); // get 4(%%ebp)
  jumpToLabel = ___nextCleanUpPointByReturnAddress(pc);
  ___nonLocalGoto(jumpToLabel);// similar to a setjmp
}

Nothing happens unless first a dream (Carl Sandburg)