Fix stdarg on x86-64

this partly reverts 1803762e3 to fix stdarg on x86-64 again.  I've tried
to retain the apple specific changes from that commit.
Also include stdarg.h first in tcc.h, maybe that helps as well.

include/stdarg.h

@@ -4,7 +4,24 @@
 #ifdef __x86_64__
 #ifndef _WIN64
 
+#ifndef __APPLE__
+//This should be in sync with the declaration on our lib/libtcc1.c
+/* GCC compatible definition of va_list. */
+typedef struct {
+    unsigned int gp_offset;
+    unsigned int fp_offset;
+    union {
+        unsigned int overflow_offset;
+        char *overflow_arg_area;
+    };
+    char *reg_save_area;
+} __va_list_struct;
+typedef __va_list_struct va_list[1];
+#else
+/* This is sometimes a void* on TCC, which makes it unlikely to
+   work with va_copy, but until we have something better ... */
 typedef __builtin_va_list va_list;
+#endif
 
 void __va_start(va_list ap, void *fp);
 void *__va_arg(va_list ap, int arg_type, int size, int align);

lib/bt-log.c

@@ -1,9 +1,9 @@
 /* ------------------------------------------------------------- */
 /* function to get a stack backtrace on demand with a message */
 
+#include <stdarg.h>
 #include <stdio.h>
 #include <string.h>
-#include <stdarg.h>
 
 int (*__rt_error)(void*, void*, const char *, va_list);
 

lib/va_list.c

@@ -2,16 +2,14 @@
 
 #if defined __x86_64__
 
-#include "stdarg.h"
-
 /* Avoid include files, they may not be available when cross compiling */
 extern void *memset(void *s, int c, __SIZE_TYPE__ n);
 extern void abort(void);
 
 /* This should be in sync with our include/stdarg.h */
-#define __VA_GEN_REG    0
-#define __VA_FLOAT_REG  1
-#define __VA_STACK      2
+enum __va_arg_type {
+    __va_gen_reg, __va_float_reg, __va_stack
+};
 
 /* GCC compatible definition of va_list. */
 typedef struct {
@@ -24,45 +22,43 @@ typedef struct {
     char *reg_save_area;
 } __va_list_struct;
 
-void __va_start(va_list ap, void *fp)
+void __va_start(__va_list_struct *ap, void *fp)
 {
-    __va_list_struct * _ap = (__va_list_struct*) ap;
-    memset(_ap, 0, sizeof(__va_list_struct));
-    *_ap = *(__va_list_struct *)((char *)fp - 16);
-    _ap->overflow_arg_area = (char *)fp + _ap->overflow_offset;
-    _ap->reg_save_area = (char *)fp - 176 - 16;
+    memset(ap, 0, sizeof(__va_list_struct));
+    *ap = *(__va_list_struct *)((char *)fp - 16);
+    ap->overflow_arg_area = (char *)fp + ap->overflow_offset;
+    ap->reg_save_area = (char *)fp - 176 - 16;
 }
 
-void *__va_arg(va_list ap,
+void *__va_arg(__va_list_struct *ap,
                int arg_type,
                int size, int align)
 {
-    __va_list_struct * _ap = (__va_list_struct*) ap;
     size = (size + 7) & ~7;
     align = (align + 7) & ~7;
     switch (arg_type) {
-    case __VA_GEN_REG:
-        if (_ap->gp_offset + size <= 48) {
-            _ap->gp_offset += size;
-            return _ap->reg_save_area + _ap->gp_offset - size;
+    case __va_gen_reg:
+        if (ap->gp_offset + size <= 48) {
+            ap->gp_offset += size;
+            return ap->reg_save_area + ap->gp_offset - size;
         }
         goto use_overflow_area;
 
-    case __VA_FLOAT_REG:
-        if (_ap->fp_offset < 128 + 48) {
-            _ap->fp_offset += 16;
-            return _ap->reg_save_area + _ap->fp_offset - 16;
+    case __va_float_reg:
+        if (ap->fp_offset < 128 + 48) {
+            ap->fp_offset += 16;
+            return ap->reg_save_area + ap->fp_offset - 16;
         }
         size = 8;
         goto use_overflow_area;
 
-    case __VA_STACK:
+    case __va_stack:
     use_overflow_area:
-        _ap->overflow_arg_area += size;
-        _ap->overflow_arg_area = (char*)((long long)(_ap->overflow_arg_area + align - 1) & -align);
-        return _ap->overflow_arg_area - size;
+        ap->overflow_arg_area += size;
+        ap->overflow_arg_area = (char*)((long long)(ap->overflow_arg_area + align - 1) & -align);
+        return ap->overflow_arg_area - size;
 
-    default: /* should never h_appen */
+    default: /* should never happen */
         abort();
         return 0;
     }

libtcc.c

@@ -928,15 +928,15 @@ LIBTCCAPI TCCState *tcc_new(void)
     tcc_define_symbol(s, "__builtin_extract_return_addr(x)", "x");
 #endif /* ndef TCC_TARGET_PE */
 #ifdef TCC_TARGET_MACHO
-	/* emulate APPLE-GCC to make libc's headerfiles compile: */
-	tcc_define_symbol(s, "__APPLE__", "1");
+    /* emulate APPLE-GCC to make libc's headerfiles compile: */
+    tcc_define_symbol(s, "__APPLE__", "1");
     tcc_define_symbol(s, "__GNUC__", "4");   /* darwin emits warning on GCC<4 */
-    
+
     /* avoids usage of GCC/clang specific builtins in libc-headerfiles: */
     tcc_define_symbol(s, "__FINITE_MATH_ONLY__", "1");
     tcc_define_symbol(s, "_FORTIFY_SOURCE", "0");
-#endif /* ndef TCC_TARGET_MACHO */
     tcc_define_symbol(s, "__builtin_va_list", "void *");
+#endif /* ndef TCC_TARGET_MACHO */
     return s;
 }
 

tcc.h

@@ -24,9 +24,9 @@
 #define _GNU_SOURCE
 #include "config.h"
 
+#include <stdarg.h>
 #include <stdlib.h>
 #include <stdio.h>
-#include <stdarg.h>
 #include <string.h>
 #include <errno.h>
 #include <math.h>