Re: Are 'goto' + trampolines broken?

[Øyvind Harboe]

On Sat, 2005-05-14 at 12:29 -0700, Richard Henderson wrote:
> On Sat, May 14, 2005 at 11:59:28AM +0200, Øyvind Harboe wrote:
> > void test(void (*t)(void))
> > {
> >     int i;
> >     for (i=0; i<10; i++)
> >     {
> >             t();
> ...
> >   int skip=0xdeadbeef;
> > 
> >   void do_goto (void)
> >   {
> >     goto l1;
> >   }
> >   
> >   test(do_goto);
> ...
> > l1:
> >     if (skip!=0xdeadbeef)
> > }
> > 
> > int main ()
> > {
> >   recursive (10, abort);
> >   abort ();
> > }
> 
> Well of course your program always aborts.  Control never enters
> the recursive part of recursive().  It flows straight through and
> returns, which hits the abort in main.

Ah. So the goto acts like a longjmp in this context...

Running test below produces

$ gcc test.c -g -O1
$ ./a.out 
j=11
$ 

-- 
Øyvind Harboe
http://www.zylin.com

#include <stdio.h>

int j;

void test(void (*t)(void), int i)
{
  j++;
  if (i>0)
    {
      test(t, i-1);
    } else
      {
        t();
        /* never reached */
        abort();
      }
  j--;
}

void bar (void)
{
  __label__ l1;

  void do_goto (void)
  {
    goto l1;
  }

  test(do_goto, 10);

l1:;
}

int main ()
{
  bar();
  printf("j=%d\n", j);
  return 0;
}