va_arg *, and why not.

[Mark Mitchell]

>>>>> "Geoff" == Geoff Keating <geoffk@cygnus.com> writes:

    Geoff> 1. implement va_copy, by using an autoconf test to
    Geoff> determine whether or not va_list is an array type; or
    Geoff> 2. only pass around va_list *.

    Geoff> This breaks the bootstrap on powerpc-linux.

Fascinating.

I see that libiberty's vasprintf has:

  static int
  int_vasprintf (result, format, args)
       char **result;
       const char *format;
       va_list *args;
  {
    const char *p = format;
    /* Add one to make sure that it is never zero, which might cause malloc
       to return NULL.  */
    int total_width = strlen (format) + 1;
    va_list ap;

    memcpy ((PTR) &ap, (PTR) args, sizeof (va_list));

and a bit later:

  int
  vasprintf (result, format, args)
       char **result;
       const char *format;
  #if defined (_BSD_VA_LIST_) && defined (__FreeBSD__)
       _BSD_VA_LIST_ args;
  #else
       va_list args;
  #endif
  {
    return int_vasprintf (result, format, &args);
  }

Does that not work on your platform?  If not, this code should be
changed, too.

I think that the right solution here is that the diagnostic buffer to
contain a va_list, and that we initialize it either with va_copy (if
available) or an as-yet-nonexistant libiberty equivalent, presumably
using the `memcpy' trick shown above.

Gaby, please take care of this as soon as possible so that we can get
Geoff back into bootstrap-land.

--
Mark Mitchell                   mark@codesourcery.com
CodeSourcery, LLC               http://www.codesourcery.com