On Wed, Oct 17, 2018 at 06:12:41PM +0200, SZEDER Gábor wrote:
> On macOS there is a MIN macro already defined in the system headers,
> resulting in the following error:
> 
>       CC sha256/block/sha256.o
>   sha256/block/sha256.c:133:9: error: 'MIN' macro redefined [-Werror,-Wmacro-redefined]
>   #define MIN(x, y) ((x) < (y) ? (x) : (y))
>           ^
>   /usr/include/sys/param.h:215:9: note: previous definition is here
>   #define MIN(a,b) (((a)<(b))?(a):(b))
>           ^
>   1 error generated.
>   make: *** [sha256/block/sha256.o] Error 1
> 
> A simple "#undef MIN" solves this issue.  However, I wonder whether we
> should #undef the other #define directives as well, just to be sure
> (and perhaps overly cautious).

I'll undefine the macros at the top.

For MIN, I'm going to go with the simple approach and pull pieces from
the block-sha1 implementation, which doesn't require this code path...

> Some GCC versions (e.g. gcc-4.8 with -O2 -Wall -Werror) complain about
> the above line:
> 
>       CC sha256/block/sha256.o
>   sha256/block/sha256.c: In function ‘blk_SHA256_Final’:
>   sha256/block/sha256.c:174:2: error: dereferencing type-punned pointer will break strict-aliasing rules [-Werror=strict-aliasing]
>     put_be64(ctx->buf + trip, ctx->length);
>     ^
>   cc1: all warnings being treated as errors
>   make: *** [sha256/block/sha256.o] Error 1
> 
> Something like this makes it compile:
> 
>   void *ptr = ctx->buf + trip;
>   put_be64(ptr, ctx->length);
> 
> However, it's not immediately obvious to me why the compiler
> complains, or why that intermediate void* variable makes any
> difference, but now it's not the time to put on my language lawyer
> hat.
> 
> Perhaps an old compiler bug?  Clang in general, newer GCC versions, or
> gcc-4.8 with -Wall -Werror but without -O2 don't seem to be affected.

...or this code path.  Presumably GCC 4.8 and macOS are happy with the
code that we already have in block-sha1.
-- 
brian m. carlson: Houston, Texas, US
OpenPGP: https://keybase.io/bk2204