Re: va_list vs. __builtin_va_list

[Robert Lipe <robertl at sco dot com>]

Richard Henderson wrote:
> On Sun, Nov 28, 1999 at 02:04:56PM -0800, Mark Mitchell wrote:
> >     Robert> Should I push this toward the all GCC approach (i.e. do
> >     Robert> whatever it takes to get va_list #defined to be
> >     Robert> __gnu_va_list or __builtint_va_list) or should I push it
> >     Robert> toward an all native definition?
> > 
> > RTH is probably the master expert on this, but I think you'd be better
> > off pushsing this towards the all-GCC approach.
> 
> *shrug*  You already put in bits for irix6 making the exact
> same change.  I suppose it wouldn't hurt to do the same for
> unixware.

It's actually OpenServer suffering from this and not UnixWare, but I
really don't expect the rest of you to keep things like that straight.  :-)

> Robert, see BUILD_VA_LIST_TYPE in mips/iris6.h.

Excellent hint and exactly spot-on.  Thank you.  Grafting those bits in
does indeed cure the testcase I was looking at.  Barring any suprises on
the nightly testsuite run, I'll commit something like this.

Index: sco5.h
===================================================================
RCS file: /cvs/gcc/egcs/gcc/config/i386/sco5.h,v
retrieving revision 1.31
diff -u -p -r1.31 sco5.h
--- sco5.h      1999/11/28 21:44:43     1.31
+++ sco5.h      1999/12/01 04:13:25
@@ -927,6 +927,13 @@ dtors_section ()                                          \
 
 #define MAX_OFILE_ALIGNMENT (32768*8)
 
+/* Define the `__builtin_va_list' type for the ABI.  On OpenServer, this
+   type is `char *'.  */
+#undef BUILD_VA_LIST_TYPE
+#define BUILD_VA_LIST_TYPE(VALIST) \
+  (VALIST) = build_pointer_type (char_type_node)
+
+
 /*
 Here comes some major hackery to get the crt stuff to compile properly.
 Since we can (and do) compile for both COFF and ELF environments, we