Re: [Chicken-users] Some allocation and object-passing tests

[Felix Winkelmann]

Am Tue, 11 Nov 2003 11:00:21 +0000 hat Category 5 <address@hidden> 
geschrieben:

> I've written a small test program to try out the various means of
> passing Scheme objects safely through a chain of C calls.

Thanks for the code, I was able to reproduce the problem.

> #>
> #define TRIES  5
> #define SCHEME_CALLBACK  scheme_callback1
>
> int dispatcher(void (*fn)(unsigned char *), unsigned char *udata) {
>   int i, n = TRIES;
>   for (i = 0; i < n; i++)  {
>     (*fn)(udata);
>     sleep(1);
>   }
>   return(0);
> }

This is the problem: you gc-protect the argument properly, but
inside the dispatcher, `udata' becomes invalid after the first loop
iteration - the gc in the callback changes the address of the value, the 
gc-protected
pointer is updated, but your local `udata' variable *still points
to the "old" address* ! A solution: protect the
value inside the dispatcher:

int dispatcher(void (*fn)(unsigned char *), unsigned char *udata) {
  int i, n = TRIES;
  C_word *data[ 1 ];
  data[ 0 ] = &((C_word)udata);  /* casting could be avoided using C_word 
instead of unsigned char * */
  C_gc_protect(data, 1);
  for (i = 0; i < n; i++)  {
    (*fn)(udata); /* location of udata will be updated automatically on 
each GC */
    sleep(1);
  }
  C_gc_unprotect(data, 1);
  return(0);
}

With these changes, your test-code works fine on my machine.


cheers,
felix