summary |
shortlog |
log |
commit | commitdiff |
tree
raw |
patch |
inline | side by side (from parent 1:
89bf67f)
Previously PLATFORM_BYTE_ORDER and IS_LITTLE_ENDIAN were not defined,
so the little endian code was always compiled in.
This will have the effect that the "SHA1s" on big endian architectures
will change (i.e. become actual sha1s). So someone re-indexing their
database could conceivable lose tags on messages without a message-id
header.
#define bswap_32(x) ((rotr32((x), 24) & 0x00ff00ff) | (rotr32((x), 8) & 0xff00ff00))
#define bswap_32(x) ((rotr32((x), 24) & 0x00ff00ff) | (rotr32((x), 8) & 0xff00ff00))
-#if (PLATFORM_BYTE_ORDER == IS_LITTLE_ENDIAN)
-#define bsw_32(p,n) \
- { int _i = (n); while(_i--) ((uint32_t*)p)[_i] = bswap_32(((uint32_t*)p)[_i]); }
+/* The macros __BYTE_ORDER__ and __ORDER_*_ENDIAN__ are GNU C
+ * extensions. They are also supported by clang as of v3.2 */
+
+#ifdef __BYTE_ORDER__
+# if (__BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__)
+# define bsw_32(p,n) \
+ { int _i = (n); while(_i--) ((uint32_t*)p)[_i] = bswap_32(((uint32_t*)p)[_i]); }
+# elif (__BYTE_ORDER__ == __ORDER_BIG_ENDIAN__)
+# define bsw_32(p,n)
+# else
+# error "unknown byte order"
+# endif
+# error "macro __BYTE_ORDER__ is not defined"
#endif
#define SHA1_MASK (SHA1_BLOCK_SIZE - 1)
#endif
#define SHA1_MASK (SHA1_BLOCK_SIZE - 1)