byte_order.c breaks if byte_order.h uses the bswap_32() fallback macro

[Bloody2]

Hello everyone,

If byte_order.h is run through clang (in my case, clang++), the file concludes that if all else fails, the fallback macro shall be used:

#define bswap_32(x) ((((x) & 0xff000000) >> 24) | (((x) & 0x00ff0000) >> 8) |
(((x) & 0x0000ff00) << 8) | (((x) & 0x000000ff) << 24))

Problem: byte_order.c contains code like: bswap_32( *(src++) )

The above will accidently execute "*(src++)" multiple times as "x" in the fallback macro (instead of just once).

Also, to check if 32/64-bit bswap intrinsics can be used, tests like the following:

#elif (defined(GNUC) && (GNUC >= 4) && (GNUC > 4 || GNUC_MINOR >= 3))

should be expanded to simply check for clang as well, like:

#elif defined(clang) || (defined(GNUC) && (GNUC >= 4) && (GNUC > 4 || GNUC_MINOR >= 3))

Luckily, you don't need bswap_16 which appeared in later clang versions, while 32/64-bit swappers were added early and can simply be assumed to be present as long as clang is defined.

Other than that, your SHA/SHA3 sources compile & run just fine if compiled as C++ btw.

Great implementation - small, easy, nice & fast. Thank you very much!

[rhash]

Hello. The bug was fixed in release v1.3.5, please retest.

[Bloody2]

Confirmed. Sorry for the alarm, didn't notice that part has meanwhile changed, too.
Just one of those "rare bugs" that sometimes go unnoticed. Sorry again.