va_ref

[Ryan Lortie <desrt at desrt dot ca>]

Hello GCC List.

I am not on this list, so please keep me cc:'d on any replies.

I recently found something about C that I do not like at all.  I was
enraged and furious!  I blogged!

http://blogs.gnome.org/desrt/2007/08/22/isoiec-98991999-e-%c2%a7-67537/

I am proposing the following feature for GCC to mitigate the problem:  a
new macro (defined in <stdarg.h> under -std=gnu99 or an appropriate
#define) called va_ref().

va_ref() would be defined in such a way as to allow the following
program to be correct and to output

--snip--
1 = 1
2 = 2
--snip--


#include <stdarg.h>
#include <stdio.h>

static void
print (int x, va_list *app)
{
  printf ("%d = %d\n", x, va_arg (*app, int));
}

static void
do_both (va_list ap)
{
  print (1, va_ref (ap));
  print (2, va_ref (ap));
}

static void
varfunc (int ignore, ...)
{
  va_list ap;

  va_start (ap, ignore);
  do_both (ap);
  va_end (ap);
}

int
main (void)
{
  varfunc (0, 1, 2);

  return 0;
}

Essentially, on machines where va_list has a structure or scalar type,
you would have something to the effect of

#define va_ref(x) (&(x))

and on machines with va_list as an array you would have something like

#define va_ref(x) ((va_list *)(x))



That is all.

Cheers