Fix for LLVM wchar_t literals not being four byte

aligned, affecting OS X port.

* configure: Detect a SIZEOF_WCHAR_T when detecting integer
type that will hold a pointer. In the lit_align test,
if we are on Apple Mac OSX, use a lit_align of 2,
so the logic kicks in for padding literals and handling
misalignment.

* lib.h (litptr): Add a case for LIT_ALIGN < 4 and
SIZEOF_WCHAR_T == 4. In this case we do the arithmetic
on the pointer using short *, and then convert to
wchar_t.

* HACKING: New section 2.4.3 about new wchar_t hack.

1 files changed, 4 insertions, 1 deletions

diff --git a/lib.h b/lib.h
index a95c2e3f..e55a3516 100644
--- a/lib.h
+++ b/lib.h
@@ -327,9 +327,12 @@ INLINE val static_str(const wchli_t *str)
 
 INLINE wchar_t *litptr(val obj)
 {
-#if LIT_ALIGN < 4
+#if LIT_ALIGN < 4 && SIZEOF_WCHAR_T < 4
  wchar_t *ret = coerce(wchar_t *, (coerce(cnum, obj) & ~TAG_MASK));
  return (*ret == 0) ? ret + 1 : ret;
+#elif LIT_ALIGN < 4 && SIZEOF_WCHAR_T == 4
+ short *ret = coerce(short *, (coerce(cnum, obj) & ~TAG_MASK));
+ return coerce(wchar_t *, (*ret == 0) ? ret + 1 : ret);
 #else
  return coerce(wchar_t *, coerce(cnum, obj) & ~TAG_MASK);
 #endif